Advances in Psychology Volume 62 Volitional Action Conation and Control

VOLITIONAL ACTION Conation and Control

ADVANCES

IN PSYCHOLOGY 62 Editors:

G. E. STELMACH

P. A. VROON

NORTH-HOLLAND AMSTERDAM. NEW YORK OXFORD. TOKYO

VOLITIONAL ACTION Conation and Control

Edited by

Wayne A. HERSHBERGER

1989

NORTH-HOLLAND AMSTERDAM. NEW YORK . OXFORD TOKYO

ELSEVLER SCIENCE PUBLISHERS B.V. Sara Burgerhartstraat 25 P.O. Box 21 1, 1000 AE Amsterdam, The Netherlands

Distributors for the United States and Canada: ELSEVIER SCIENCE PUBLISHING COMPANY, INC. 655 Avenue of the Americas New York, N.Y. 10010, U S A .

ISBN: 0 444 88318 5

0ELSEVIER SCIENCE PUBLISHERS B.V., 1989 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior written permission of the publisher, Elsevier Science Publishers B.V. / Physical Sciences and Engineering Division, P.O. Box 1991, 1000 BZ Amsterdam, The Netherlands. Special regulations for readers in the U.S.A. - This publication has been registered with the Copyright Clearance Center Inc. (CCC), Salem, Massachusetts. Information can be obtained from the CCC about conditions under which photocopies of parts of this publication may be made in the U.S.A. All other copyright questions, including photocopying outside of the U.S.A., should be referred to the copyright owner, Elsevier Science Publishers B.V., unless otherwise specified. No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Printed in The Netherlands

V

PREFACE During the last several decades the behavioral sciences have been undergoing what is arguably a Kuhnian scientific revolution, with radical behaviorism giving way to considerations of cognition and conation. Although cognition is perhaps the more familiar of these two terms, conation (concerning the inclination to act purposively) is equally a hallmark of the times. Indeed, the past few years has seen a resurgence of interest in the psychology and physiology of volition that is unparalleled in this century. Not since William James published his Princ@les of psychology in 1890 has so much careful attention been devoted to a consideration of the will. The present book comprises a significant sample, or distillation, of the observations, both rational and empirical, of individuals from diverse disciplines who are contributing to the present renaissance in conation. The book was designed to serve a threefold purpose: (a) to consolidate the gains of the various scholars, relatively isolated in their respective disciplines, (b) to foster and help focus future research on conation and self-control, and (c) to provide practitioners in applied psychology with a broad-based tutorial. William James noted that there are two fundamental things to be understood about voluntary action: First, volitional actions, being desired and intended beforehand, are done with full prevision; that is, they are preceded by anticipatory images defining what those actions are to be. Secondly, these anticipatory images are representations of the intended sensory consequences of the necessary muscular innervation and not representations of the muscular innervation itself. (James’ putative image is not to be confused with von Holst and Mittelstaedt’s efference copy.) The chapters in this book have been authored by individuals with something further to contribute to our understanding of one or both of James’ observations. For example, some authors have been investigating the neurological signals which precede voluntary movements (e.g., Georgopoulos; and, Kornhuber, Deecke, Lang, Lang, & Kornhuber) whereas others (e.g., MacKay & Crammond) have been concerned primarily with the sensory feedback from the effectors involved in such

vi

Preface

movements. And still others, such as those with systems approaches (e.g., Bullock & Grossberg) are concerned with both aspects. The theoretical flavor of the book is largely cybernetic or control theoretic. That is, most of the authors are committed to the proposition that voluntary actions are intentional, self-controlled inputs or sensations (including, in some cases, the sensed corollary discharges of efference), just as James implied. The principal champion of this notion today is William Powers (see Chapters 2 & 13), who used the idea as the title of his influential 1973 book, Behavior: the control of perception. William James also noted that the sensory consequences which define a particular voluntary action may be resident or remote. Sensations arising from muscle spindles are resident sensations; those arising from exteroceptors are remote. A person driving an automobile, for example, is controlling the remote visual consequences of his or her effector activity. The driver is also controlling his or her destination, another remote sensory consequence. Some of the authors, particularly those with a psychological or sociological perspective (e.g., Hyland) are concerned primarily with the control of remote sensory consequences, whereas others, particularly those with a physiological perspective (e.g., Pavloski), focus more upon resident sensory effects. This, of course, is as it should be. The two perspectives are complementary. Volition is a phenomenon of immense practical as well as theoretical significance, and several chapters (e.g., Lord & Kernan) address the applied aspect. Professional psychology is in need of a broader scientific foundation than that provided by 20th century behaviorism. Conative science is a veritable cornerstone for such a new scientific foundation. I believe practitioners will find the observations in this book (even the esoteric ones) uncommonly stimulating, informative, and professionally relevant. The chapters are grouped according to the methodological approach of the author(s) into 5 sections: theoretical, neurophysiological, mathematical, psychological, and practical, in that order. Within each section the chapters are ordered alphabetically, by author. Wayne A. Hershberger DeKalb, Illinois June 1989

vii

CONTENTS Preface List of Contributors

V

xi

GENERAL THEORETICAL PERSPECTIVE 3

1.

The Synergy of Voluntary and Involuntary Action Wayne A. Hershberger

2.

Volition: a Semi-scientific Essay William T. Powers

21

On the Will: An Historical Perspective

39

3.

Eckart Scheerer

PHYSIOLOGICAL PERSPECTWE 4.

Volitional Eye Movements and their Relationship to Visual Attention Burkhart Fischer and Rolf Boch

5.

The Cerebral Correlates of Reaching Apostolos P, Georgopoulos

6.

Will, Volitional Action, Attention and Cerebral Potentials In Man: Bereitschaftspotential, Performance-Related Potentials, Directed Attention Potential, EEG Spectrum Changes H. H. Kornhuber, L. Deecke, W. Lang, M. Lang, and A. Kornhuber

63

73

107

viii

7.

8.

Contents

Cortical Modification of Sensorimotor Linkages in Relation to Intended Action Wlliam A. Macand Donald J. Crarnmond

169

Cerebral Correlates of Auditory Attention

195

R. Naatanen 9.

The Physiological Stress of Thwarted Intentions Raymond P. Pavloski

215

SYSTEMS-MODELING PERSPECTWE 10.

11.

A Control-Theory Analysis of Interference During Social Tracking W Thomas Bourbon

235

VITE and FLETE. Neural Modules for Trajectory Formation and Postural Control Daniel Bullock and Stephen Grossberg

253

12.

Behavior In the First Degree Richard S. Marken

299

13.

Quantitative Measurement of Volition: a Pilot Study W a r n T. Powers

315

PSYCHOLOGICAL PERSPECTIVE 14.

Some Experimental Investigations of Volition George S. Howard and Paul R Myers

15.

Control Theory and Psychology: a Tool for Integration and a Heuristic for New Theory Michael E. Hyland

335

353

Contents 16.

17.

ix

The Behavioral Illusion: Misperception of Volitional Action J. Scott Jordan and Wayne A. Hershberger

371

Volition and Self-Regulation: Memory Mechanisms Mediating the Maintenance of Intentions Julius f i h l and Miguel &Zen-Saad

387

18.

Levels of Intention in Behavior Richard S. Marken and William T. Powers

409

19.

Involuntary Learning of Voluntary Action Richard J. Robertson

43 1

APPLIED PERSPECTM 20.

A Paradigm Shift in Behavior Therapy: From External Control to Self-Control Dennk J. Delprato

449

21.

Fostering Self-Control: Comments of a Counselor Edward E. Ford

469

22.

Control Theory Applied to Stress Management David M. Goldstein

481

23.

Application of Control Theory to Work Settings Robert G. Lord and Mary C. Keman

493

24.

Effective Personnel Management: An Application of Control Theory James Soldani

515

The Giffen Effect: A Control Theory Resolution of an Economic Paradox William D. Williams

531

25.

Contents

X

Author Index

549

Subject Index

563

xi

CONTRIBUTORS Rolf Boch, Department of Clinical Neurology & Neurophysiology, University of Freiburg, D-7800 Freiburg, Federal Republic of Germany Thomas Bourbon, Department of Psychology, Stephen F. Austin University, Nacogdoches, TX 75962-3046, U.S.A. Daniel Bullock, Center for Adaptive Systems, Department of Mathematics, Boston University, Boston, MA 02215, U.S.A. Donald J. Crammond, Center for Research in Neurological Sciences, University of Montreal, Montreal, Quebec, Canada H3C 3J7 Liider Deecke, Neurological Clinic, University of Vienna, A-1090 Vienna, Austria Dennis J. Delprato, Department of Psychology, Eastern Michigan University, Ypsilanti, MI 48197, U.S.A. Burkhart Fischer, Department of Clinical Neurology & Neurophysiology, University of Freiburg, D-7800 Freiburg, Federal Republic of Germany Edward E. Ford, 10209 N. 56th St., Scottsdale, AZ 85253, U.S.A. Apostolos P. Georgopoulos, Philip Bard Laboratories of Neurophysiology, Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, MD 21205, U.S.A. David M. Goldstein, 801 Edgemoor Road, Cherry Hill, NJ 08034, U.S.A. Stephen Grossberg, Center for Adaptive Systems, Department of Mathematics, Boston University, Boston, MA 02215, U.S.A. Wayne A. Hershberger, Department of Psychology, Northern Illinois University, DeKalb, IL 60115, U.S.A.

xii

Contributors

George S. Howard, Department of Psychology, University of Notre Dame, Notre Dame, IN 46556, U.S.A. Michael E. Hyland, Department of Psychology, Plymouth Polytechnic, Plymouth, Devon, PLA 8AA, England J. Scott Jordan, Department of Psychology, Northern Illinois University, DeKalb, IL 60115, U.S.A. Miguel Kazen-Saad, Department of Psychology, University of Osnabruck, D-4500 Osnabruck, Federal Republic of Germany Mary C. Kernan, Department of Business Administration, University of Delaware, Newark, DE 19716, U.S.A. Anselm W. Kornhuber, Neurological Clinic, University of Ulm, D-7900 Ulm, Federal Republic of Germany Hans H. Kornhuber, Neurological Clinic, University of Ulm, D-7900 Ulm, Federal Republic of Germany Julius Kuhl, Department of Psychology, University of Osnabriick, D-4500 Osnabruck, Federal Republic of Germany Michael Lang, Neurological Clinic, University of Ulm, D-7900 Ulm, Federal Republic of Germany Wilfried Lang, Neurological Clinic, University of Vienna, A-1090 Vienna, Austria Robert G. Lord, Department of Psychology, University of Akron, Akron, OH 44325, U.S.A. William A. MacKay, Department of Physiology, University of Toronto, Toronto, Ontario, Canada M5S 1A8 Richard Marken, Aerospace Corporation, Los Angeles, CA 90009-2957, U.S.A.

Contributors

xiii

Paul R. Myers, Department of Psychology, University of Notre Dame, Notre Dame, IN 46556, U.S.A.

R. Naatanen, Department of Psychology, University of Helsinki, 00170 Helsinki, Finland Raymond Pavloski, Department of Psychology, Indiana University of Pennsylvania, Indiana, PA 15705-1068, U S A . William T. Powers, 1138 Whitfield Road, Northbrook, IL 60062, U.S.A. Richard J. Robertson, Department of Psychology, Northeastern Illinois University, Chicago, IL 60625, U.S.A. Eckart Scheerer, Institute of Cognitive Science, University of Oldenburg, D-2900 Oldenburg, Federal Republic of Germany James Soldani, 13849 N. 64th Street, Scottsdale, AZ 85254, U.S.A. William D. Williams, 1850 Norwood, Boulder, CO 80304, U.S.A.

This Page Intentionally Left Blank

GENERAL THEORETICAL PERSPECTIW

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hershberger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

3

CHAPTER 1 THE SYNERGY OF VOLUNTARY AND INVOLUNTARY ACTION Wayne A. Hershberger The first point to understand about the psychology of volition, according to William James (1890), the preeminent American psychologist at the turn of the century when conative psychology was last in vogue, is that voluntary acts "being desired and intended beforehand, are of course done with full prevision of what they are to be" (Vol. 2, p.487). In turn, the most important point to understand about this prevision or anticipatory image is that it represents an intended sensory consequence of muscular activity and not the muscular activity itself. As James put it,

I trust that I have now made clear what that "idea of a movement" is which must precede it in order that it be voluntary. It is not the thought of the innervation which the movement requires. It is the anticipation of the movement's sensible effects, resident or remote, and sometimes very remote indeed (Vol. 2, p. 522). To illustrate James' meaning, consider the process of emptying a cup of coffee with a series of successive sips each requiring a different muscular effort, the sensible effects of which are all essentially the same: a sip of coffee. Since the voluntary act of taking a sip is literally defined by this common sensory consequence, the act itself must be precipitated by an anticipatory image of that particular sensory consequence which alone defines the voluntary act. In brief, James was suggesting that volitional actions are intended selfcontrolled inputs rather than emitted or elicited outputs.

Volitional Actions Are Not Emitted Contemporary, behavioristic psychologists such as B. F. Skinner (Catania & Harnad, 1988), who define overt behavior as comprising only

4

Wayne A. Hershberger

emitted and elicited outputs cannot, perhaps, legitimately be faulted for failing to find any intentional responses or volitional actions within the scope of human conduct given their definition of overt behavior. But they certainly can and should be faulted for their narrow definition of behavior, which excludes precisely that type of overt behavior which James had previously recognized as comprising volitional action, namely, intended, self-controlled input (Hershberger, 1987a, 1988a, 1988b). Man-made mechanisms that control their input, keeping the value of a monitored variable equal to an intended reference value by means of negative feedback, are called control systems, or closed-loop control systems. The household thermostat and furnace system is a commonplace example. Setting the thermostat of such a system specifies the temperature its thermocouple is intended to sense, not the amount of heat the furnace is going to emit. Having set the thermostat, one can predict the indoor temperature but not the fuel bill. The latter varies with the weather. The indoor temperature, however, is the mechanism’s doing. What these mechanisms do at our command is control the value of their own sensed input, which, of course, is also one of ours: sensed room temperature. Their response to our command is, hence, a particular, self-controlled value of input rather than any particular value of output, either emitted or elicited. This is not to say that the furnace does not emir heat nor that cold weather does not elicit compensatory emissions of heat from the furnace, but only that the control system does not itself control those heat emissions. That is not what it does! What it does, at our command, is control sensed temperature (Hershberger, 1988b, p. 107). In order to avoid confusion later, I draw the reader’s attention to a second example which illustrates several points that tend to be obscured by the furnace and thermostat example. Consider, if you will, the control system, known as power steering, commonplace in modern automobiles.1 Steering provides an historically apt example of control. When Norbert Wiener (1948) coined the term Cybernetics to name a new discipline concerned with control and communication in the animal and the machine, he derived the name from the greek word kubemetes, meaning steersman.

Synergy of Action

5

This control system effectively monitors the value of a variable corresponding to the orientation of the front wheels and keeps that value equal to one expressed in terms of the orientation of the steering wheel. Although much more energy is required to turn the front wheels when the vehicle is at rest than when it is moving, this is of little concern to the driver who uses the steering wheel essentially as a means of communicating intentions about the desired orientation of the front wheels. The forces necessary to actualize an intended orientation under various driving conditions is left to the control loop which realizes or actualizes expressed intentions even as the driver is communicating them through the steering wheel. This example illustrates three important points that the furnace and thermostat example obscures: First, control systems are not necessarily homeostatic; a reference value in a control system may vary, or be varied, continuously. Second, the value of a controlled variable may "track" the value of a rapidly varying reference signal so precisely that they covary almost as if they were one and the same. And finally, control systems may employ proprioceptors, or the like, to control variables such as posture, just as they employ exteroceptors to control the value of variables in an external environment. The operating principle remains the same: the control of input through negative feedback. The principal relevance of these two examples of man-made control systems is that they illustrate graphically the very point which James considered to be of the first importance in understanding the nature of volition or volitional action. In organisms, as in man-made control systems, volitional action is essentially an image + input process, not an organism output process. Nor is volitional action, as described herein, to be regarded as an efference + reafference process. That is, a reference signal, such as James' anticipatory image, is not to be confused with von Holst and Mittelstaedt's (1950) efference copy. I have previously referred to reference signals, such as James anticipatory image, as aflerence copies in order to sharply contrast the concept with von Holst and Mittelstaedt's efference copy hypothesis (Hershberger, 1976, 1983, 1987b). The two notions are readily confused and were confused by von Holst and Mittelstaedt themselves, at least initially. Mittelstaedt acknowledged this in a subsequent paper (1958; also see MacKay & Mittelsteadt 1974) in which he presented a control-systems analysis of their original "functional schemata" and found, in addition to the efference copy they had originally posited, a higher order "command signal" which Mittelstaedt labeled simply "C." Upon analysis (see Hershberger 1976), C proved to --+

6

Wayne A. Hershberger

be a reference signal, or afference copy, which is not surprising given that von Holst and Mittelstaedt were attempting to model volitional action in the first place, among other things.

Old Language Habits Die Hard James notion that volitional action is essentially an image -,input process rather than an organism output process, is, at once, both very simple and very difficult to understand. Although the idea appears to have been shared by other turn-of-the-century psychologists, including many American functionalists, it is not clear that any one of them, James included (see below), fully understood the idea or grasped its remarkable implications. The idea itself is simple enough. The difficulty lies in the fact that the notion is contrary to our traditional Cartesian habits of speech, and that, therefore, understanding the idea involves breaking the grip of the dead hand of habit, something that normally requires drill and practice as well as insight. As it turned out, these early American functionalists did not manage to escape their Cartesian language habits, although it was not entirely for lack of trying. John Dewey's classic critique of the reflex arc concept (Dewey, 1896) was a self-conscious attempt to identify and exorcise whatever vestiges of Cartesian interactionism remained in the new scientific psychology's behavioristic lexicon. Dewey's analysis focused primarily upon the primitive terms stimulus and response. Dewey warned that the reflex arc lacked unity and argued that behavior involves an entire loop or circle. As he put it, "The circle is a coordination, some of whose members have come into conflict with each other" (p. 370). Although he did not achieve a full understanding of closed-loop control, it is apparent that he was on the right track. Dewey's thought-provoking analysis is relevant and worth reading even today, but it is difficult to understand. The difficulty reflects, in part, the mind-bending paradoxes or apparent contradictions that seem inevitably to erupt when one attempts to describe voluntary behavior in the traditional terminology of response and stimulus. --+

For example, consider the following bizarre statement: The response of the furnace and thermostat system is a stimulus, not a response, but this stimulus is a response, not a stimulus. Although this grammatical statement is reasonably correct semantically (see below), it is gibberish, nonetheless. The

Synergv of Action

7

problem is that the terms stimulus and response have two distinctly different pairs of yoked meanings, among others: On the one hand, the two terms refer to receptor input and effector output, respectively $e., to sensory and motor variables). On the other hand, the two terms refer to cause and effect, respectively (i.e., to prod and product, or technically speaking, to independent and dependent variables). This confuses sensory input with cause and motor output with effect, a confusion that lends a specious legitimacy to the tenets of radical behaviorism, but that adds nothing to our understanding of control systems [or volitional action] but confusion itself, as the statement above amply illustrates. Deciphered, the statement reads as follows: The response of the furnace and thermostat system (i.e., what it produces or does) is a stimulus (i.e., a particular value of sensed temperature), not a response (or particular amount of heat output), but this stimulus (or sensed temperature) is a response (Lea, the dependent variable controlled by the system), not a stimulus (i.e., it does not control or cause the temperature being produced) (Hershberger, 1988a, p. 824). In retrospect, it is hardly surprising that the psychology of volition and volitional action languished as stimulus-response psychology flourished, or that functionalism gave way to behaviorism. That is, inasmuch as American functionalists routinely utilized the expressions motor response and sensory stimulus in their analyses of psychological phenomena-including Dewey (1896) himself in spite of his own implicit caveat to the contrary4 was, perhaps, inevitable that one of their number (it happened to be John B. Watson) would conclude that there is nothing scientific that can be said about volition or volitional action? In other words there are good pragmatic reasons why volitional action became an anathema to scientific psychology. However, good reason was not among them. Watson’s conclusion, which Skinner still champions (Catania & Harnad, Bergman (1956) called John B. Watson the “greatest...of the Functionalists” (p. 268). Watson, the founder of behaviorism, went to the University of Chicago to study with Dewey, a founder of functionalism, but as Watson (1936, p.274) acknowledged later he never quite understood what Dewey had to say. We cannot argue with that; he clearly missed the point of Dewey’s critique of the reflex arc concept. Of course, he was not alone in that respect.

8

Wayne A. Hershberger

1988, pp.108-109) is specious; it follows from the mischievous habit of speech noted above, not from a logical train of thought. When behaviorists assert that volitional actions are incompatible with scientific psychology, they also imply that volitional actions are incompatible with those types of behavior that behaviorists have recognized as being scientific, that is, with elicited and emitted outputs. This notion, not surprisingly, is as specious as the fallacy that implies it. The terms voluntary and involuntary denote mutually exclusive categories of behavior, but they do not denote mutually exclusive behaviors! On the contrary, voluntary and involuntary behaviors are always found to go hand in hand in any system that controls its own input. Voluntary behavior actually presupposes involuntary behavior. For example, the flight path of an airplane is the pilot’s (or autopilot’s) doing only to the degree that the pilot’s (or autopilot’s) actions automatically offset any would-be aerodynamic disturbances to the intended flight path. Otherwise, he or she (or it) is merely along for the ride. The two types of behavior comprise a synergistic couple. They are complementary in the strong sense of the term: Like husband and wife, the existence of each type of behavior is distinct from but dependent on the nature of the other. Each has its mate. For every self-controlled input there is a corresponding, disturbance-driven output. And, the relationship is always synergistic (Hershberger, 1987a, p. 1032).

Canonical Self-Control Although control systems can be extremely complex, any system controlling a given parameter appears relatively simple when reduced to its basic canonical form, as is possible, in principle, with any such system, however complex. In its canonical form, the flow chart of any such system is a single negative feedback loop. The flow chart in Figure 1 is a canonical loop mapped onto the interface between an organism (or mechanism) and its environment. Everything above the dotted line is part of the organism (or mechanism). Everything below the dotted line is part of the environment. Note that although the organism (or mechanism) has only one input, the

Synergy of Action

9

(organism or mechanism)

- - - - - - - -(input) - 1 ~ - - - - - - - - - - - - - - \ f -(output)- - - - - (environment)

--

- A

Figure 1. A canonical control loop mapped onto the interface (dashed line) between an organism (or mechanism) and its environment.

control loop has two inputs. One input to the control loop is the reference value specifying the organism’s (or mechanism’s) intended input. The other input to the loop comprises all the environmental factors which potentially disturb the organism’s (or mechanism’s) input. The polarity of the feedback loop is negative. That is, discrepancies between the organism’s (or mechanism’s) intended and actual inputs constitute error signals which negate themselves by driving output so as to offset environmental disturbances and thereby keep the controlled input in close correspondence with the reference value. A coupling of two types of behavior is apparent, one involving controlled input and one involving elicited output. Consider first the

10

Wayne A. Hershberger

controlled input. Inasmuch as the feedback loop keeps the controlled input equal to the reference value, the controlled input is to the reference input as an effect is to its cause or as a response is to its stimulus. And since this stimulus is an intention or intended value, the corresponding response is an intentional or voluntary behavior. This volitional action is represented by the large blocked arrow labeled Intentional actions. The error-driven output is another matter. Inasmuch as the organism (or mechanism) controls its input largely by offsetting environmental disturbances to that input with compensatory output, that compensatory output is to its eliciting disturbance as an effect is to its cause or as a response is to its stimulus. And since this stimulus is an unanticipated, unintended environmental disturbance, the elicited output is in the nature of a Cartesian reflex; that is, it is an involuntary reaction to an environmental stimulus. Note, incidentally, that this stimulus is not an input to the organism (or mechanism). It is an input only to the control loop (I will return to this point later). This involuntary action is represented by the large blocked arrow labeled Compensatory reactions. Note that the two blocked arrows point in a clockwise direction exactly opposite that of the solid arrows comprising the negative feedback loop. The two blocked arrows represent emergent properties of the system. Each represents a lineal cause and effect relationship that emerges from the underlying circular feedback process. To say that they are emergent is not to imply that they are merely putative properties. The blocked, counterclockwise arrows represent immanent and essential aspects of the control process. Nor is the counterclockwise orientation of the blocked arrows to be regarded as whimsical or accidental. That is, it is a mistake to suppose that the two counterclockwise arrows represent two lineal cause-effect arcs comprising the feedback loop itself (ie., two overlapping clockwise arcs). Closed-loop control systems do not control their input by controlling their output. Nor do disturbances elicit compensatory reactions by being sensed (e.g., the outdoor temperature that disturbs the thermostat and furnace system is not monitored by the system’s sensor; the thermocouple senses the temperature indoors not outdoors). In other words, closed-loop control does not involve reciprocal determinism. The feedback loop of a closed-loop control system involves reciprocal influence, not reciprocal control. The control process itself is an emergent property of the feedback loop, and this emergent control, or determinism, is lineal, not reciprocal: The value of the reference signal determines the value of the controlled input, but the opposite is certainly

Synergy of Action

11

not true; ordinarily, the value of the controlled input does not even influence the reference value, let alone determine it. Similarly, the magnitude of the disturbance determines the magnitude of the system's output, but the opposite is certainly not true; ordinarily, the output does not even influence the disturbance, let alone determine it.

William James' Will: Right and Wrong In his chapter entitled "Will," James (1890) argued that the necessary and sufficient antecedent of any voluntary act is the anticipatory image defining the intentional action in question. He wrote, "We may consequently set it down as certain that, whether or no there be anything else in the mind at the moment when we consciously will a certain act, a mental conception made up of memory-images of these sensations, defining which special act it is, must be there!' (vol. 2, p. 492). He also asserted that this anticipatory image "is the only psychic state which introspection lets us discern as the forerunner of our voluntary acts" (vol. 2, p. 501). These two remarks are entirely consistent with the control process described in the paragraphs immediately above. However, James had not conceived the control process illustrated in Figure 1. He did not understand the feedback process by which a control system's reference signal (i.e., anticipatory image) determines the sensory consequences of the system's output (i.e., the system's input). Indeed, it is clear from a reading of James' description of the putative neural processes involved in volitional action that he was thinking of a calibrated input + output system rather than a control system. James neural model was functionally equivalent to any one of a number of more recent stimulus-response theories, such as Greenwald's (1970) ideo-motor theory of performance, based on James' own ideas, or Held's (1961) correlation store hypothesis, based on the reafference principle of von Holst and Mittelstaedt (1950), both of which suppose that neuromuscular outputs are centrally (i.e., neurally) coded in terms of their respective sensory consequences. A complication in all such theories is the fact that neuromuscular outputs do not each have just one particular sensory consequence; rather, for any given output the consequence varies with the circumstance, and circumstances vary endlessly. To paraphrase Heraclitus, one never encounters the same circumstance twice. This implies that an effective neural inventory of all such output/circumstance combinations and their attendant sensory consequences would be exceedingly complex; indeed, an exhaustive inventory would amount

12

Wayne A. Hershberger

virtually to a neural model of the world. The informational load that the organism is expected to bear is staggering. By thus balancing the theoretical burden on the backs of the systems whose functioning they purport to explain, these theories project a specious appearance of parsimony. All things considered, they are anything but simple. However, what is more to the point, neither are they essentially volitional (see Kimble & Perlmuter, 1970). If volition involves merely the selection of an output according to its labeled consequence, wherein lies the distinction between voluntary and involuntary output? Only in the label. The output is the same, whatever one calls it, or however it is labelled or coded. Therefore the distinction is merely semantic. In summary, James’ neurological theorizing does not support the implications of his introspective observations. In this respect, James, like so many others, seems to have been unable to break the Cartesian habit of speech equating actions (including voluntary ones) with neuromuscular outputs. He seems, at once, both to have recognized and ignored the fact that a voluntary movement not only has sensible consequences, but is itself, a sensible consequence of neuromuscular output. James called the sensory consequences registered by receptors in the muscles, tendons and joints, resident effects, and those registered by exteroceptors, remote effects. The control of resident effects is considered in the next paragraph. The control of remote effects, is the process illustrated in Figure 1 above. In Figure 1, the control loop straddles the interface between the organism, or mechanism, and its environment. The controlled input is, therefore, an environmental variable monitored by an exteroceptor such as an eye or a thermocouple. That is, Figure 1 represents the canonical control process involved in such intentional behavior as that of a motorist keeping an automobile, buffeted by variable crosswinds, rolling straight down the highway, or a thermostat and furnace system keeping the temperature indoors at 72 degrees Fahrenheit as the temperature outdoors ranges from, say, 20 to 60 degrees. Resident effects comprise such voluntary actions as movements and postures. Consider, for example, a person lifting a cup to her lips or a power steering system turning the front wheels of an automobile to a desired orientation. In both cases, the intentional behavior involves compensation for variable environmental loads, because the cup may be full or nearly empty, and the automobile may be moving or stationary. However, in neither case is the controlled variable essentially environmental. Rather, in both cases the controlled variable is essentially a parameter of the organism, or mechanism, involving the articulation of

Synergy of Action

13

several of its various parts. Consequently, Figure 1 would need to be modified slightly in order to accurately represent the canonical control of resident effects: Specifically, the dashed line representing the interface between the organism, or mechanism, and the environment should be located coincident with the bottom edge of the control loop. Posture or bodily motion is an environmental variable in the sense that it is publicly observable by others. I can, and often do, observe the orientation of the front wheels of a car facing me at an intersection, just as I can, and daily do, watch my wife across the dinner table lift a cup unerringly to her lips. I see these things, but the control systems need not, because they--the automobile's power steering and my wife's nervous system--are not controlling visual inputs from remote environmental sources. They are controlling the articulation of certain body parts, which are registered either by receptors resident in those body parts, or by corollary discharges of neural efference (or the like), or by both. Corollary discharges of efference (Sperry, 1950), or efference copies (von Holst & Mittelstaedt, 1950) need not be controlled inputs (controlled feedback), but they may be. When they are, any disturbance introduced upstream of the corollary discharge would be offset by compensatory adjustments of output, thereby controlling the value of the feedback signal. Robinson's (1975) model of the saccadic oculomotor control system involves this type of controlled input, or controlled feedback. His model "consists of a single negative feedback system whose forward [efferent] path contains a high gain saturating amplifier with a dead zone (so it is either on or off) and an integrator" (p. 369). The output of the integrator, which corresponds to eye orientation, is the feedback signal that is controlled. The expression "high gain saturating amplifier with a dead zone" means that the system operates in a "bang-bang" fashion exactly analogous to that of a furnace and thermostat system. Bullock and Grossberg (1988) claim that the control of such neural feedback signals is also an integral part of voluntary arm movements. See also their chapter in this volume. Interoceptors, which monitor such controlled variables as body temperature, may be regarded as measuring either resident or remote effects, depending upon the frame of reference assumed. If the organism as a whole is the assumed frame of reference, then interoceptors may be said to monitor resident effects. However, if one takes the control mechanism inside the organism to be the frame of reference, then interoceptors may be said to monitor environmental (i.e., remote) effects. For instance, Cannon (1932) described the control of vital signs such as

14

Wayne A. Hershberger

body temperature as the control of environmental parameters, with the term environment referring to Bernard’s milieu intkrier (1878/1973).

The Involuntary Aspect of Self-Control Whether resident or remote, the consequences controlled by organisms, or mechanisms, are always the joint effects of two antecedents, only one of which is the organism’s, or mechanism’s, output. The other factor comprises all the various influences of the organism’s, or mechanism’s, environment. A closed-loop control system works by pitting these two factors against each other. The environmental factor is customarily called an environmental dkturbance, not only because it has the potential of disturbing the system’s controlled input, but also because it actually elicits compensatory output from the system. Closed-loop control systems control their own input by forfeiting control over their output, which is determined in large part by their environmental disturbances. For example, the amount of heat generated by the furnace of a system controlling indoor temperature depends upon a host of environmental variables all of which tend to influence the air temperature being controlled. These include, but are not limited to, the temperature outdoors, the intensity and direction of the wind, the number and types of windows in the house, the thermal insulation in the external walls and attic, the elevation and azimuth of the sun, the amount of cloud cover, the color of the shingles on the roof, the number of people going in and out, the wattage of incandescent lamps in use, the number of pots cooking on the stove, and the amount of laundry being tumbled in the drier. The thermostat and furnace system responds automatically to all these environmental disturbances with just the right amount of heat to keep the indoor temperature at the reference level, and does so without even measuring these disturbances, either individually or collectively. (It is almost enough to drive a Cartesian mechanist to thoughts of divine intervention.) When we observe an organism or mechanism’s output covarying systematically with an environmental variable, our Cartesian habits of speech (read thought) incline us to suppose that the form of the observed relationship somehow reflects the form of the mediating mechanism. But it should be obvious from the present example of the furnace and thermostat system that this does not hold for control systems. In control systems, output is error-driven, and relatively small errors (i.e., small departures of the controlled variable from the reference value) drive

Synergy of Action

15

relatively large compensatory outputs which serve to nip disturbances in the bud, so to speak. Therefore, the effect of an environmental disturbance upon a control system's output depends entirely upon the influence of that would-be disturbance upon the variable being controlled. That is, it depends entirely upon the relationship obtaining between these two environmental variables, and has nothing to do with the .particulars of the control mechanism involved. For example, if one plotted a furnace's output as a function of outdoor wind speed (holding all other factors constant), the relationship observed would simply reflect the chilling effects of the wind on the house and the air inside, that is all. The function would reflect a law of physics, not a law of behavior. The behavioral law in question is so simple that it drops out of the equation. That law is this: in an ideal control system, a disturbance elicits output whose effects on the controlled variable are equal and opposite to its own; simply put, the output nulls the error signal. For an elegant mathematical exposition of these points, among others, see Powers' (1978) quantitative analysis of purposive systems. To say that the output nulls the error signal is to imply that two essential things need obtain, and essentially only two things. The control loop's output must be able to outmuscle its disturbances, and the polarity of the feedback must be negative. (The system, to be stable, must also be able to detect changes in the controlled variable as it is producing them; hence, a change-slowing factor is sometimes incorporated in man made systems: Chapter 13.) In the example of the furnace and thermostat system, the rate at which the furnace generates heat when it is on must be greater than the maximum rate at which heat is dissipated to the outdoors--however, the control process does not require any particular rate of heat production; that is, the output does not have to be calibrated to the conditions. Further, the polarity of the feedback loop is crucial: the furnace must be switched on or off when the sensed temperature is, respectively below or above the reference temperature. (If the polarity of the switch were reversed, the feedback would be positive and the sensed temperature would "run away" to one of its limits: extremely hot or cold.) Providing that these two conditions are met, a closed loop system will blindly dance to a disturbance's tune and thereby control its own input, keeping it virtually equal to the reference value. Although the system does not detect the disturbances to which it responds, those disturbances are nevertheless mirrored in the system's output (e.g., the weather is mirrored in the fuel bill). A system may thus appreciate the

16

Wayne A. Hershberger

magnitude of a disturbance by recognizing the magnitude of its own output. This perception of the disturbance is entirely a posteriori, however; it comes after, not before, the elicited output. Obviously, this phenomenon provides a firm basis for at least some motor theories of perception (see Coren, 1986; Hershberger, 1983, 1987b). These observations also imply that sensations of innervation, in the form of efference copies or the like, are sometimes reflections of exafference (read environmental disturbance), contrary to what von Holst and Mittelstaedt (1950) initially supposed. Because the polarity of a control loop's feedback must be negative for the system to function properly, environmental factors which disturb that polarity, either neutralizing it or reversing it, are not automatically corrected by that control loop. In order for the polarity of a feedback loop to be controlled against such disturbances of polarity, a higher order control loop is required whose output would reorganize the subordinate loop. Ashby's (1952) classic Design for u Brain treats such self-reorganizing processes at length (also see Campbell, 1956, and Powers, 1953, Chapter 14). Although I do not wish to dwell on the topic of "selfreorganization" here, it is important to note that animal experiments in which the polarity of sensory feedback has been reversed, either optically or surgically, have invariably found evidence of positive feedback in the form of forced circus movements, run-away output or the like. The experiments have used a variety of species, including insects (von Holst & Mittelstaedt, 1950), fish, amphibians, rodents (Sperry, 1951), 4-day-old chickens (Hershberger, 19861, and the oculomotor control system of man (Smith & Molitor, 1969; Yarbus, 1962). In these experiments very little, if any, recovery of function has been observed to result from practice, suggesting that the polarity of some control loops are not readily altered by experience. That is, in animals, some control loops, particularly those already in evidence at birth, may be well-fixed genetically (i.e., "hardwired") and relatively inflexible to change. However, much of the intentional behavior of animals, particularly humans, appears to be mediated by control mechanisms that evolve with experience. For example, when driving a car, one controls the direction of locomotion with the hands not the feet, as evolution would have us do. Also, it is possible for one to steer an automobile while holding either the top or bottom rim of the steering wheel, despite the fact that a rightward motion of the hand has opposite remote effects in the two cases. Not only are these control mechanisms, at least in part, acquired, the latter example involves polarity reversals as well.

Synergy of Action

17

Volition and Hierarchical Intentions Volitional actions are self-controlled, but not all self-control is volitional action. The control processes which regulate the parameters of the milieu int6rier are intentional inasmuch as they involve the maintenance of implicit reference values, but they are not considered voluntary because those reference values are fixed. In order for self-control to fully qualify as being voluntary, the reference value in question must not be fixed, either genetically or environmentally. A consideration of what this implies for hierarchical control mechanisms, such as those comprising complex organisms, merits attention here (also see Marken & Powers’ chapter in this volume). When two loops are joined hierarchically, so that the controlled input of the subordinate loop is the output of the superordinate loop, the controlled input of the subordinate loop, although intentional, is not fully volitional, because, ideally, the intentional actions of the subordinate loop are entirely at the service of two masters, namely, the superordinate intention and its disturbance. The only volitional actions available at the level of the subordinate loop are those which are orthogonal to the intention in question. For instance, if I wish to see either of my arms visibly outstretched at eye level (a superordinate intention controlling a remote effect), the intended posture of my arm (a subordinate intention controlling a resident effect) is fixed with respect to its elevation. Only the arms’ azimuth and its roll (palm up or down) remain optional parameters. Of course, there is also the choice of right or left arm. Our actions order themselves hierarchically in terms of the controlled variables involved. For instance, steering an automobile involves controlling the orientation of the front wheels. But the orientation of the front wheels is determined by the orientation of the steering wheel, and the orientation of the steering wheel is determined by the position of the driver’s arms. Therefore, steering an automobile involves a hierarchy of control. If the car is equipped with power steering, there are at least three hierarchical intentions, or reference signals involved: (a) the driver’s intended direction of visual locomotion, (b) the intended orientation of the driver’s arms, and (c) the intended orientation of the front wheels. Note that I have made proprietary reference only to the first of these three intentions. That is, the reference signal controlling the direction of visual locomotion is the only intention which is uniquely the driver’s. The intention listed last is the reference signal for the power

18

Wayne A. Hershberger

steering. Although the driver is communicating this intention to the power-steering system, it is the power steering’s reference signal not the driver’s; it is not a reference signal for any control loop in the driver. The intended orientation of the driver’s arms is a reference signal for a control loop in the driver, but the value of that intended orientation is dependent in large part upon environmental disturbances. If the driver chooses to drive straight down the highway through a steady crosswind, he or she has no option but to crab the wheel just enough to offset the effects of the wind. So although the driver may have provided the muscle to turn the wheel, the intention to do so was not of his or her choosing; it derives almost entirely from two considerations, (a) the driver’s intention to stay on the road, and (b) the crosswind. If the wind is very gusty the driver will involuntarily be very busy at the wheel without voluntarily intending to do anything but stay on the road. Learning to drive or pilot a vehicle involves the development of control systems in which the control of resident sensory effects are subordinated to the control of remote visual effects (for a discussion of visual locomotion, see Gibson 1966; also see Owen & Warren, 1982). Initially, the novice driver controls the orientation of the steering wheel without effectively controlling the direction of the car. In contrast, the proficient driver has learned to control the car by allowing the environmental circumstances to dictate the orientation of the steering wheel. Inevitably, the more volitional the car trajectory becomes, the less volitional the hand motions become. There is a conservation principle at work. The development of hierarchical control serves to centralize choice or volition, but does not increase it. The amount of will which may be marshalled, or mustered, or brought to a focus is thus limited by the control system’s ordinal size, that is, by the number of orthogonal inputs (or variables) the system is able, in principle, to control. There is no will to be found lying around free.

References Ashby, W. R. (1952). Design for a brain. New York: Wiley. Bergman, G. (1956). The contribution of John B. Watson. Psychological Review, 63, 265-276. Bernard, C. (1973). Lectures on thephenomena of life common to animals and plants. (H. E. Hoff, R. Guillemin, & L. Guillemin, Trans.) Springfield, Illinois: Thomas. (Original work published 1878).

Synergy of Action

19

Bullock, D., & Grossberg, S. (1988). Neural dynamics of planned arm movements: Emergent invariants and speed-accuracy properties during trajectory formation. Psychological Review, 95, 49-90. Campbell, D. T. (1956). Adaptive behavior from random response. Behavioral Sciences, I , 105-110. Cannon, W. B. (1932) Wisdom of the body. New York: Norton. Catania, A. C., & Harnad, S. (Eds.), (1988). The selection of behavior, New York Cambridge University Press. Coren, S. (1986). An efferent component in the visual perception of direction and extent. Psychological Review, 93, 391-410. Dewey, J. (1896). The reflex arc concept in psychology. Psychological Review, 3, 359-370. Gibson, J. J. (1966). The senses considered as perceptual system. Boston: Houghton Mifflin. Greenwald, A. G. (1970) Sensory feedback mechanisms in performance control: With special reference to the ideo-motor mechanism. Psychological Review, 77, 73-99. Held, R. (1961). Exposure-history as a factor in maintaining stability of perception and coordination.Journal of nervous and Mental Diseases, 132, 2632. Hershberger, W. A. (1976) Afference copy, the closed-loop analogue of von Holst's efference copy. Cybernetics Forum, 8, 97-102. Hershberger, W. A. (1983). A conditioned weight illusion: Reafference learning without a correlation store. Perception & Psychophysics, 33, 391-398. Hershberger, W. A. (1986). An approach through the looking-glass. Animal Learning & Behavior, 14, 443-451. Hershberger, W. A. (1987a). Of course there can be an empirical science of volitional action. American Psychologist, 42, 1032-1033. Hershberger, W. A. (1987b). Saccadic eye movements and the perception of visual direction. Perception & Psychophysics, 41, 35-44. Hershberger, W. A. (1988a). Psychology as a conative science. Arnen'can Psychologist, 43, 823-824. Hershberger, W. A. (1988b). Some overt behavior is neither elicited nor emitted. In A. C. Catania and s. Harnad (Eds.), The selection of behavior (pp. 107-109). New York: Cambridge University Press. James, W. (1890). The principles ofpsychology (Vol. 2). New York: Henry Holt. Kimble, G. A., & Perlmuter, L. C. (1970). The problem of volition. Psychological Review, 77, 361-384.

20

Wayne A. Hershberger

MacKay, D. M., & Mittelstaedt, H. (1974). Visual stability and motor control (reafference revisited). In W. D. Keidel (Ed.), cybernetics and bionics. Munich: Oldenbourg. Mittelstaedt, H. (1958). The analysis of behavior in terms of control systems. In B. Schaffner (Ed.), Groupprocesses, Transactionuf the Fifth Conference. New York Josiah Macy, Jr., Foundation. Owen, D. H., & Warren, R. (1982). Optical variables as measures of performance during simulated flight. Proceedings of the Human Factors Society 26th Annual Meeting, 312-315. Powers, W. T. (1973). Behavior: The control of perception. Chicago: Aldine. Powers, W. T. (1978). Quantitative analysis of purposive systems: Some spadework at the foundations of scientific psychology. Psychological Review, 85, 417-435. Robinson, D. (1975). Oculomotor control signals. In G. Lennerstrand, P. Bachy-Rita, C. C. Collins, A. Jampolsky, & A. B. Scott (Eds.), Basic mechanisms of ocular motility and their clinical implications. New York: Pergamon Press. Smith, K. U., & Molitor, K. (1969). Adaptation to reversal of retinal feedback of eye movements. Journal of Motor Behavior, I , 69-87. Sperry, R. W. (1950). Neural basis of the spontaneous optokinetic response produced by visual neural inversion. Journal of Comparative & Physiological Psychology, 43, 482-489. Sperry, R. W. (1951). Mechanisms of neural maturation. In S. S. Stevens (Ed.), Handbook of experimental psychology. New York: Wiley. Watson, J. B. (1936). J. B. Watson. In C. Murchison (Ed.), A history of psychology in autobiography, Vol. 3. Worcester, Massachusetts: Clark University Press. Wiener, N. (1948). Cybernetics: Control and communication in the animal and the machine. New York: Wiley. von Holst, E. & Mittelstaedt, H. (1950). Das Reafferenzprinzip. Natuwksenshafen, 37, 464-476. Yarbus, A. L. (1962). Eye movements and vision. (Trans. by B. Haigh) New York Plenum Press.

VOLITIONAL ACTION, W.A. Hershberger (Editor)

21

0 Elsevier Science Publishers B. V. (North-Holland). 1989

CHAPTER 2 VOLITION: A SEMI-SCIENTIFIC ESSAY William T.Powers Every child wonders, sooner or later, how it is that simply wanting one’s hands, arms, legs, body, head, or eyes to move suffices to create the wanted result. The sense of willing that one’s own body do something is at the same time unmistakable and unexplainable, being unlike any other mental or physical experience. While willing an act seems to suggest that we are masters of our own behavior, experiences of other kinds suggest just the opposite. As children we do as we will when it is playtime, but from the very beginning we find that we also do as we must, when people and events decree that playtime is at an end. Even the passive physical world forces us into action in ways that can seem to push the will aside. With growing force, necessity makes itself known in many forms. The demands of our bodies, saying that we must breathe, eat, drink, stay warm, seek love, and avoid pain, override the will more and more often; one demand leads to another, until by the time we are adults it can seem that we no longer have any freedom to will except as a momentary act of useless defiance. When the rat-race is at its worst, there seems to be an external reason for every slightest act from rising in the morning at the alarm clock’s buzz to swallowing the final nightcap so we can sleep, only to rise, too soon, again. To indulge in any extended period of purely volitional action would be to put unacceptable stresses on the network of behaviors we are forced to adopt, stresses that seize control again and bring us back into the daily groove, will we or nil we. The transition from childhood to adulthood is unpleasant largely because of the sense of steadily diminishing freedom to will. On the one hand, adulthood promises immense freedoms -- driving a car, getting out of school, having one’s own money, going to bed when one pleases, being listened to, understanding how things work, owning and managing things and events. On the other hand, adults obviously do not seem to enjoy these freedoms as much as they ought to. In fact, they seem to act as if they have no great amount of freedom. Every child must at some time

22

William T. Powers

vow not to become like that -- not to give up control of one’s own life. And every child inevitably ends up breaking the vow, perhaps raging but in almost every instance succumbing to all the controlling influences that prove unavoidable. The traditional scientific view of behavior is the adult’s view, not the child’s. But this is not the view of a wise adult; only of an adult who has decided that the sense of will that was given up must somehow have been an illusion. Opponents of the objective, dispassionate analysis of causation that is traditional in science, on the other hand, maintain the child’s view, insisting on the essential freedom of the mind with a child’s faith -- with the same amount of influence on science that children usually have on adults. The puzzle of the will is central in our attempts to understand human behavior. Insisting that creative will is all, the naive child’s view, is neither more nor less correct than insisting that it is impotent or nonexistent, the cynical adult’s view. To understand both will and necessity, we must avoid siding with either view, and try to define the terms of this puzzle in a way that gives us a chance at solving it.

Internal vs. External Causation To speak of volition as a sense of willing is to use one word in place of another, illuminating nothing. While only the individual can sense volition when it is occurring, the ability to sense it confers no particular understanding of it. If sensing it were enough, we would not have these problems. What we must do is find a place for volition in our general understanding of both private and public, but most importantly public, phenomena. Volition can be defined as a cause of behavior that is internal to the behaving system. Speaking generally instead of personally, we can see that human behavior seems to have two kinds of causes. One kind we can easily see, as when a gust of wind makes a man struggle to stand up, or an unexpected sound makes someone jump, or a worker tries harder when the boss threatens to fire her. The other kind is harder to see, because the cause is located where it can’t be observed; the identification of a volitional act always seems weak because all we can say is that there was no apparent external cause. Few of us would dare to claim that we have noticed every possible cause and ruled it out. The weakness of the identification would seem to leave external cause as the most rational choice.

Volition: A Semi-scientific Essay

23

On the other hand, a careful consideration of human behavior, our own or that of others, makes it quite clear that we cannot identify many external causes. While we can pick out salient events such as an explosion or the offer of food and make a case that the ensuing behavior was the result, it is much harder to extend these connections to all behaviors and all events. Given any event chosen from the ongoing stream at random, we normally have no way of predicting what behavior will follow it in any given person. And if we really pay attention to behavior, we must admit that behavior is going on every moment of a person’s life, in an unending continuous flow. It isn’t just that our knowledge of external causes of behaviors is incomplete: it is nearly nonexistent. In sheer quantity, the amount of behavior that has been connected to prior causes is only an infinitesimal fraction of all the behavior that goes on every day. Scientists who have given up completely on internal causation have done so not because of the evidence, but because of an urge to simplify. It is much easier to assert that all behavior is externally caused than it is to envision trying to sort out one class of causes from another. In support of external causation, it has been claimed that in a physical universe, all material objects are caused to behave by the confluence of all current influences on them. But this physical principle is not a premise from which we can conclude that all behavior is externally caused: it is simply a restatement of the assertion in different words. And it is a restatement that ignores all the ways in which organisms differ from the simple point-masses to which the original Newtonian principle was applied. The principle difference is complexity: there is a great deal more going on inside an organism than inside any piece of matter that a physicist or a chemist studies. Most pieces of matter that a physicist studies do not stand up and try to get away. This complexity means not only that there are important processes going on inside the organism at all times, but that these processes may arise from sources that existed at unknown and unknowable times in the past. When a person speaks, the grammar and syntax that shape the speech may have originated in the outside world, but they certainly did not originate just before the utterance. As far as any present-time observer is concerned, the causes of grammar and syntax now are carried in the brain of the speaker, and cannot be traced to anything happening in the current environment. This gives us the first wedge with which to pry open the puzzle. We must admit at least that large parts of the behavior we observe have

24

William T. Powers

origins that are unconnected with the current environment. On this basis we may still claim that some behaviors amount to responses to current stimuli, but we must allow that even more of the behaviors, perhaps most of them, must be under control of processes that are not altered by any present stimuli.

The Logic of Causation Another route we can take involves a closer look at what is supposedly caused, behavior itself. What is behavior? The naive view, which is shared by scientist and laymen alike, is that behavior is whatever organisms do. But what do we mean by "doing?" The two-letter word "do" takes up four column-inches in an old Collegiate dictionary, and far more in an Unabridged. This primitive grunt refers to causing essentially any occurrence that can be named. It asserts agency, but reflecting our ignorance it skips over process. The doer does, but how the doer does it is not mentioned. How does one open a door? Not, we can be sure, by opening it -that is not an answer to a "how" question, but a reassertion of agency in more obscure form. Normally, one opens doors by pushing on them or pulling on them. Opening a door would surely be classed as a behavior, but in fact this behavior is carried out by an organism that is doing something distinctly different from ''opening." It is the door, not the organism, that opens. What the organism does is to apply a force with its muscles: the consequence of this effort is, usually, that the door opens. Most of us will open a sizable number of doors in one day, some familiar and some unfamiliar. We open bedroom doors, bathroom doors, front doors, car doors, supermarket doors, refrigerator doors, cupboard doors, and the doors where we work. There is no linguistic problem with calling all these activities "opening doors," but in terms of the motor actions we carry out, not only are the actions very different over all these instances of the "same behavior," but they are quantitatively different each time we open the same door. What we call behavior is really some repeatable recognizable consequence of our motor actions. Almost 100 years ago, William James pointed out the uncomfortable fact that while these consequences repeat, the actions that bring them about do not repeat. Had James gone on to analyze this observation in more detail, he would have realized that the actions do not repeat for the simple reason that if they did repeat, their consequences would vary. If you turn left to enter a cafeteria, you will

Volition:A Semi-Scientzjk Essay

25

be able to buy your lunch. But if you enter the building by a different door, or if someone is standing in the way, or if the cafeteria is locked, you will not get lunch by turning left. Something else will happen. This is the story of essentially every behavior of any amount of complexity. Circumstances change. The surrounding world influences the outcomes of actions, and those independent influences can change greatly from moment to moment. Sometimes they don’t change, so the same action will have nearly the same result as before. But organisms must produce behavior in the worst-case world, too, and they do. When external influences change, organisms alter their actions to compensate, even to the extent of reversing them or substituting a totally different action. This is a commonplace fact of life: regular behavior is nut brought about by regular motor actions, and regular motor actions would not normally produce regular results. That fact, as simple and obvious as it is, spells great difficulty for the concept of external causation. For external causation to work, the causal chain must remain predictable from beginning to end. There must not be any other causes that contribute to the outcome downstream from the initial cause -- otherwise, anything could happen. If the principle of external causation worked as it is supposed to work, we would predict that disturbing the outcome directly would cause the outcome to change, in exact proportion to the disturbance. What does happen is that an immediate change in the action just cancels the effect of the disturbance. This is the only way in which organisms can possibly continue to produce recognizable behavior. The patterns that result from their motor actions are ordinarily under continuous disturbance, the disturbances arising partly from independent sources in the environment and partly from the varying relationships of the organism to its environment. We see stable patterns; it follows that the actions of the organism cannot be correspondingly stable. This analysis would seem to rule out external causation altogether, but that is not quite the result. What happens instead is that we are made to focus on something outside the purview of the causal hypothesis -- not what changes when stimuli and disturbances occur, but what does not change.

The Logic of Control We do not normally pay attention to the motor acts by which familiar patterns of behavior are created; for one reason, they are hard

26

William T. Powers

to observe. It is not action, but the consequence of action, that is made to repeat by a behaving organism. To understand just how variable those acts must be, we have to understand something about the workings of the physical world. When we see a person reaching out toward the floorselector button in an elevator carriage, we see what seems to be a motion of the arm directed by its muscles toward the button. With a little reflection, we realize that the muscle forces are not aimed in the direction the hand is moving. They are aimed primarily straight up, countering the force of gravity and whatever accelerations of the elevator carriage are occurring. Even an act like reaching out toward something, which seems a direct expression of muscle action, is several steps removed from the actual motor behavior that is going on. The ends and the means are almost never related in any simple straightforward way. Clearly, we do not simply "doll behaviors. That description is just too sketchy. A more accurate description would be that we -- and other organisms -- act in such a way that certain consequences are brought about and maintained. The phrase "in such a way'' has a specific meaning: the way in question can be deduced from observing the consequence and knowing what independent forces are acting to alter the consequence. For instance, if we observe a car being steered straight down a level flat road, and we know that a crosswind is exerting 75 pounds of force on the car to the left, we can be quite sure that the driver is exerting a force on the steering wheel that, relayed through the power steering, the linkage, and the front tires, pushes the car to the right with a force of just 75 pounds. If that were not so, the car could not go straight. When more than one influence adds a sideward force to the car -- the camber of the roadbed, for instance, adding its effects to those of the crosswind -- we can be quite sure that the driver's effort, translated into an effect on the car, is equal and opposite to the sum of all those disturbing forces. That is simply a matter of applying Newton's laws of motion, and observing that the car continues in a straight line. When we see consistent behavior in the presence of independent disturbances, we can deduce that the actions of the organism must be varying so that the resultant is right for producing what we see. This is the basic logic of the phenomenon we know as control. A disturbance that tends to alter the final pattern results immediately in a change of motor action that tends to alter it by the same amount in the opposite direction. The net result is no change, or almost none. It is this lack of change, under circumstances where change is to be expected, that tells us control is occurring.

Volition:A Semi-Scientific Essay

27

This concept of behavior clearly does not fit the conventional causal model. As expressed so far, it seems to rely on variations in actions that are fortuitously just right to prevent disturbances from having disturbing effects. To implicate external causation in this kind of situation, we would have to imagine that the external cause varied in just the way needed (taking the organism’s properties into account) to make behavior change to preserve a particular outcome. We would have to imagine stimuli that act on the driver so as to keep the car exactly in its lane for, say, 100 miles despite the myriad disturbances, mostly invisible, that come and go during the trip. But the driver’s environment doesn’t care whether the car stays on the road or goes wandering off among the sagebrush. The causal explanation requires us to believe not just in one incredible coincidence, but in a never-ending stream of incredible coincidences. To define behavior as a process of control does not require us to explain how this process is brought about: first we define the phenomenon; then we try to understand how it is created. The phenomenon is this: by varying their actions, organisms stabilize certain outcomes of those actions, outcomes that would otherwise change with every change in environmental influences on the same outcome.

The Mechanism of Control The development now turns somewhat technical. The question before us is now how an organism must be organized to produce the control phenomena we observe. The answer to this question has, in fact, been known for some 50 years. If independent external causes cannot account for behavioral changes that control consequences, we must look for the causes elsewhere. The solution of this problem was found by engineers who studied certain types of human behavior in order to replicate it in a machine. The resulting machines were called control systems. The missing factor, these engineers discovered, was that the control system must sense the very consequence or outcome that is to be placed under control. The external cause of control behavior is the outcome itself -- the effect. The cause and the effect are identical. The cause is not independent of the effect. The basic arrangement of a control system is simple. A sensor reports the state of the controlled variable as a correspondingly variable signal, inside the control system. This signal is compared against a

28

William i? Powers

reference signal carried inside the system, and the discrepancy is represented by still another signal, the error signal. The error signal is amplified to produce a physical output, which in turn acts on the same controlled variable. This is the famous feedback loop, the feedback being negative in that any change anywhere in the loop propagates all the way around the loop to arrive at the starting point with the opposite effect. Properly speaking, feedback is a property of the entire closed loop, not of any one part of it. When such a system is properly designed (not a particularly difficult task if the system is simple), the result is not quite what may have been expected. The basic effect is that the sensor signal is held very actively in a match with the internal reference signal. If the controlled variable is disturbed, the beginning of the change due to the disturbance causes a slight departure of the sensor signal from the reference signal; an error develops, which, highly amplified, produces action. The action, simply because of the way the negative feedback loop is arranged, tends to force the controlled variable back toward its undisturbed state, and thus tends very strongly to force the sensor signal back toward a match with the reference signal. Almost as an afterthought, this action opposes the effect of the disturbance. The only generally correct way to describe the action of a control system is as a system in which all influences are in continuous equilibrium all around the closed loop. Applying a disturbance to the controlled variable results in an immediate rebalancing of the equilibrium, the action changing as the disturbance changes, so that the sensor signal is never allowed to depart much from the setting of the reference signal. Intuitively, we want to think of this circle as a sequence of events going around and around. Intuition, in this case, is simply wrong: it is attempting to treat the closed loop as if it were a lineal temporal sequence, and that does not work. Only the mathematics of control theory (or hands-on experience with control systems) can show the essentially simultaneous action of all parts of the system. Intuition must be retrained. If anyones intuition objects to the idea that mathematics can help it, the proof that it can is to be found in a basic property of control systems called "loop gain." Loop gain is the amount by which any variation is amplified as its effects make one complete trip around the loop. Real control systems normally have loop gains amounting to a factor anywhere between 10 and one million. In other words, the effect of a small change in a variable upon itself (via the closed loop) is a

Volition:A Semi-Scientific Essay

29

change from ten to a million times as large as the original, and in the opposite direction. Intuition, of course, predicts disaster. Instead, there is control. One must simply learn control theory to understand how this result can occur. Nothing in our intellectual training has prepared any of us to reason out, unaided, how control systems work. The principles involved, although 50 years old, are unknown to almost everyone but engineering specialists. Using the principles of control theory, engineers have built machines that behave exactly in the way organisms behave. They automatically vary their actions to bring about and maintain specific predetermined consequences of those actions, counteracting disturbances without any specific instructions to do so. They produce consistent outcomes by variable means: they behave just as William James said organisms behave. That is no coincidence: they were modeled on the behavior of organisms, and the engineers who invented them succeeded, serendipitously,in finding the first workable model of a behaving organism.

The Appearance of Control Behavior When an engineer builds a control system, providing a reference signal for it is just a matter of introducing a signal generator into the system. The source of reference signals in organisms is not quite that easy to explain, but we do not need to account for the presence of reference signals to understand their effects. For all practical purposes, reference signals function exactly as intentions are supposed to function. The reference signal specifies an intended state of the sensory input. Action is based at all times on the difference between the sensory input and the reference signal. The action, having a polarity opposite the detected difference, serves to reduce or negate that difference. This negative feedback first brings the external variable to the specified state, and then keeps it there, all the while creating actions that oppose any disturbances that might also act on the variable. Thus completely without any predictions and certainly without any influence of the future on the present, the control system’s reference signal determines the outcome of action. The action of a control system makes its sensory representation of an external variable match its internal reference signal. If that internal reference signal changes, the same organization will force the sensed variable to change in the same way, maintaining the match between sensory representation and reference signal. Thus whatever can vary the

30

William T. Powers

reference signal can cause the external variable to vary in the same way. The behavior of the external variable is then no longer what it would have been with the control system, the organism, absent. Normal physical influences are treated as disturbances, and cancelled by variations in the output actions of the control system. The external variable affected by the action behaves as the reference signal specifies, not as the environment otherwise would make it behave. Reference signals clearly have something to do with the phenomenon we intuitively recognize as volition. The simple alteration of a signal inside the system causes an external variable to behave in a corresponding way. But this causal connection is anything but straightforward, because the motor outputs that appear not only must bring the variable to the right state, but must show added variations that are needed to counteract the effects of unpredictable disturbances. In a great many situations, the outputs required to keep a variable under control are small and even trivial -- or would be, if disturbances were not present. Disturbances, however, are almost always present, and even in perfectly normal environments they have large influences on the variables we are controlling. A driver in a precisely-made car in perfect condition on an absolutely level road would scarcely need to steer at all -- the efforts involved would be miniscule. But if the road tilts and the crosswind blows, or if the car pulls spontaneously to one side, the driver must start exerting significant efforts, efforts that are needed simply to oppose disturbances. Because these efforts do occur, the controlled variable is kept from changing; it obeys the intention, not the disturbances. The logic of control shows us that there are really two major kinds of relationships going on at the same time. One is the relationship between the reference signal and whatever it is that is being controlled. The behavior of the reference signal determines, through feedback effects, the behavior of the controlled variable. At the same time, however, there is another relationship between the system’s actions and independent environmental disturbances. Every disturbance calls forth a change of action that is quantitatively equal and opposite to it, in terms of effect on the controlled variable. We know that this apparent relationship is really the result of small errors induced by the disturbances, errors that are highly amplified to become opposing actions. If we did not have that model of a control system in mind, the appearance would be that the disturbances are directly causing the actions, and the stability of the controlled variable would be just a lucky break for the organism.

Volition: A Semi-scientific Essay

31

We therefore have a dual causal relationship that is seen in the behavior of every control system. The actions of the system appear to be determined largely by external forces that disturb the controlled variable. At the same time, the state of the controlled variable appears to depend only on the will of the control system, which we now recognize to mean on the setting of a reference signal inside the system. The controlled variable remains close to the state specified by the reference signal. We see the arms of the driver urging the steering wheel continuously to the left and right in an apparently random pattern, a pattern we could eventually trace to crosswinds and other variable influences on the car. But the car itself continues its course undisturbed, remaining on the line that the driver intends. What the car is doing seems to be almost unrelated to what either the crosswind or the driver’s arms are doing. These two seemingly different kinds of causal relationship are really just aspects of the way one system behaves in relationship to its environment. Control theory removes the duality, showing us what is really going on. But while it does that, it also explains why we see two different kinds of causation in behavior, external causes and, less obviously, internal causes. The reference signal is the internal cause, and what it causes is the outcome of behavior. The sum of all disturbances is the external cause, and what it causes is the action, or most of the action, that stabilizes the outcome. Control theory thus shows us how it is that outcomes can be voluntary while actions are involuntary ( a nice summing-up that is due to Wayne Hershberger). Once we have this picture clear, we can understand how the driver can intend for the car to stay on the road, and carry out that intention, while being unable to predict or choose the forces his own muscles apply to the steering wheel while bringing about the intended result. When the driver elects to control the position of the car, by that very choice he elects to let the wind and a dozen other invisible disturbances determine his motor actions.

A Hierarchy of Control Motor behavior involves the operation of hundreds of control systems, each associated with controlling the force applied at the attachments of a muscle. Many others sense and control muscle length. But these elementary control systems are not the end of the story: they are used in turn by systems of higher level, which control variables much farther removed from the nervous system. In the example of the driver,

32

William T. Powers

the muscle-force control systems are employed in the larger control loop that involves the steering forces applied to the car, the position of the car on the road, and the visual images that tell the driver about that position. In order to control the appearance of the scene in the windshield, the driver’s primary way of sensing the car’s position, the driver’s brain must compare the scene as it actually is with a reference image (or, if not literally an image, some internal information relating to the visual field). The mismatch between what is sensed and what the internal reference specifies is the basis for exerting forces to the left or right, or for not exerting forces on the steering wheel. The higher-level control loop does not operate the muscles directly; instead it varies reference signals sent to the muscle- force controlling systems (according both to this control-system model and to neuroanatomy). Those control systems automatically make the sensed forces match the reference signals, in the process generating physical forces on the steering wheel. There are probably more than just these two layers of control involved in steering a car, but these two will get us started. The reference signal specifying the car’s intended position is itself variable: the driver is not stuck forever in his lane. When the driver overtakes a slower vehicle, we observe that at some point the car veers left and takes up a new path in the adjacent lane until the vehicle is passed; then it swings back and resumes its former position. In a stiff crosswind this can be an exciting encounter as the car passes into the lee of the other vehicle; at that point the steering effort that has been counteracting the crosswind suddenly makes the car lurch toward the other vehicle, and the steering effort has to be relaxed -- and then proves insufficient as the driver’s car pulls ahead, into the crosswind again. But most drivers manage to pass another car or a truck in a way that seems effortless to an onlooker who does not feel the fluctuations in steering efforts. This passing-event required that the reference position for the visual-motor steering control system be changed for a while, and then changed back. But following the logic of control, we do not ask so much about these changes as about what remained constant because of them. What remained constant was the car’s progression toward its destination. There is no one generic answer to the question of what remains constant -- the driver might be trying to maintain a constant estimated time of arrival, or might just be trying to maintain a good average speed for some unexamined reason. Keeping the speedometer at a certain reading would

Volition: A Semi-Scientific Essay

33

be part of maintaining an average speed, but going around a truck instead of ploughing into its rear is also necessary. Voluntary and involuntary aspects of the behavior shift their roles as we consider higher levels of control. If the driver chooses to exert a specific sensed force on the steering wheel, he has no choice but to create a certain amount of contraction in his muscles. If he chooses to keep the car in a specific position on the road, he has no choice but to set the muscle-force reference signal at whatever level is required by disturbances of the car’s path. In effect the crosswind and other disturbances are determining the setting of the effort reference signal, given the intention to stay in the lane. And now the intention regarding the car’s position relative to the road has to be changed if the forward progress is to remain the same: the presence of the other vehicle makes the changed position mandatory, given the intention to maintain forward progress. Again we ask, what is this forward progress for? Presumably, the driver is not astonished to find himself driving a car down a road: he is going somewhere, perhaps intending to arrive in time to meet someone for lunch. The intention of arriving at a particular place at a particular time has put him on this road, in this car, going at this speed. However, if the perception of arriving in space and time as intended is to be maintained, the reference signal specifying forward progress has to be varied: it must have varied in order to get the car onto this road in the first place, and sooner or later it must vary in order to enter the driveway of the restaurant. To maintain the pattern of the whole trip in the intended form, the driver must periodically vary the intention regarding forward progress, and in the precise way dictated by the starting point, the time on the dashboard clock, and the location of a free parking slot at the destination. The reason for having made and now having kept this lunch date is for the driver to sell a house to the person waiting for him. The driver intends to sell this house. If someone else had called him to ask about it, he would have made a different trip, perhaps not even in a car, and he would have gone to a different destination, perhaps not for lunch. That is because once he has selected the reference condition of selling a house, he has to go wherever a buyer can or will meet him. There is no other way to give his pitch to the prospective buyer: he has no choice. As it happens, our driver was trained as a physicist specializing in nuclear power plant design. Why is he so intent on selling this house? And why was he so intent last week, and why will he be the same next

34

WilIiam rl: Powers

week? Because selling houses is now his only means of making money, the demand for new nuclear power plants having slackened dramatically. This means of making money presented itself, and as he intended to make a reasonable living and no comparable opportunity was found, he had no choice but to take the job. This was the only available occupation that promised to provide the amount of money he intended to make. The intention to make $50,000 per year instead of, say, $25,000, can be traced to the fact that when he lost his job as a power plant designer, he consoled himself in a foolish manner and is now required to pay $25,000 per year in alimony. Actually, his simple needs would be met quite well on $25,000, but the negative $25,000 disturbance due to the alimony required him to set his salary goals correspondingly higher, so he can net enough to provide a sufficient living for himself. Obviously, he intends to make a sufficient living, as he thinks of it, but that intention, plus the disturbance, leaves him no choice but to earn twice as much as he needs. We can now see that it is the alimony disturbance of the driver’s income that explains why, at 11:48:37 this morning, he was exerting a 1.2 kilogram-meter torque to the left on the steering wheel, steering the car to the right around a curve in that ubiquitous crosswind. The highestlevel goal -- plus dozens of external disturbances at several intervening levels of abstract intentions -- required that effort at that time. If we were to carry out this sort of analysis with a real person, we would arrive eventually at levels of intention that would be very hard to trace any higher. Perhaps there is a highest level, having to do with control of abstract concepts like a self, relationships to a society or a family, loyalties to knowledge or culture or religion. Where the highestlevel reference signals come from is an interesting question, but not germane here. The central point of this imaginary excursion up the levels of control is that volition and necessity are not simple matters. It is rather arbitrary to select a momentary intention and treat it as if it came from nowhere and served no higher purpose. It is especially risky, in talking of the will, to talk of free will. What seems free will at one level of analysis is a necessary adjustment to external disturbances at another level. There is nothing wrong with identifying the sense of volition with reference signals in a hierarchical control-system model of the brain. That may well be a correct identification; it is certainly functionally and scientifically plausible. But in order to understand how voluntary and

Volition:A Semi-scientific Essay

35

involuntary behavior interact, we must think of the entire hierarchy, not just one slice out of its middle.

The Web of Intention Even at the lowest level in the human behavioral hierarchy there are control systems, systems that maintain muscle forces, as sensed in the tendons, at levels specified by signals descending the spinal cord from the brain. Those descending signals, while acting as first-order reference signals, are also the actions of higher-level control systems concerned with controlling more abstract or general variables. There must be many major levels of control, perhaps ten or more, in the human nervous system and brain. At the lower levels we have systems that sense and control effort vectors in space, that employ these vectors to control bodily configuration, that vary congifuration reference signals to control movements or transitions. At still higher levels the configurations and movements become the familiar events we recognize as acts, and those acts are maintained in relationships involving many acts and many external objects and events. On top of these levels are all the cognitive levels, in which the world of experience is classified, analyzed symbolically and logically, abstracted to become principles and generalizations, and finally made into coherent concepts like the concept of a self, a society, a science, a material world. Control occurs at all of these levels, each level acting to control its own kind of perception by means of varying the reference signals, which we experience as volition, reaching lower systems. While it may be that human beings control what they experience in terms of certain broadly shared types of perception, the variety of human experiences, circumstances, preoccupations, and problems tells us that within these broad classes, the structures of control that individuals build up as they mature are highly idiosyncratic. It is no simple matter to manage a world that begins as millions of identical sensory signals, and is then subject to multiple levels of interpretation that must, for the most part, be worked out in private and without the aid of an instruction manual. It is no simple matter to discover how one part of this world can be controlled without negating the control of another part of it, at the same or a different level. The high-school senior understands that by going to college and submitting to at least four more years of school, he will be able to enhance his personal power and self-respect, to raise children in comfort, to feel a part of his conception of a larger world.

36

Wlliarn T. Powers

But if he chooses that intention, he will have to tolerate continued supervision by his parents and others, he will have to leave behind the girl he loves, and who will take care of his cat? The loss of volition sensed by the adolescent -- and many who are much older -- is not really a loss of volition, but a gradually expanding network of self-contradictions, a consequence of ignorance about how we work. The physical world and the society into which we are born only set the stage on which our lives are played out: they do not limit our freedom, but simply constitute the means available to us for doing whatever we can make sense of doing. It is up to each of us to learn how to act on and in that world, to learn to perceive its possibilities, and to learn how to organize our intentions regarding that world. Through the miracle of communication each person can learn from the others, but if there are no others who understand human organization, the amount of help available is going to be small. People are very free with advice, but as advisors tend to contradict each other, the useful residue is not as useful as it might be. Look before you leap -- or nothing ventured, nothing gained? Beneath the fuzziness of personal experience there lurk some hard natural laws. The process of control itself, at any level, requires that certain mathematical relationships in space and time be properly established. Fortunately we seem to have the capacity to reorganize until we achieve skillful control. But there are even harder laws. Given a body containing about 800 muscles (depending on how they are counted), it is mathematically impossible to establish control of more than 800 independent variables of experience at the same time. The degrees of freedom of control cannot exceed the degrees of freedom to act. And actually to be able to control that many variables at once, one would have to solve 800 nonlinear differential equations in 800 unknowns. It is unlikely that the nervous system -- even the nervous system of an engineering mathematician -- would be able to realize anything near that potential. And that takes into account only the second level of control. Now we must consider that the variables of the second level, already abstracted once from raw sensory inputs, are abstracted again to yield a new type of experience, and thus a whole new set of potentially controllable experiences. And this adding of new modes of control at new and ever more abstract levels must continue for at least some respectable number of levels. In every case, at every level, the same mathematical problem

Volition:A Semi-Scientijk Essay

37

exists: how to partition the universe of experience so that its parts can be independently controlled without self- contradiction; without conflict. This whole hierarchy of control contains a network of intentions that represent the actions of the control systems above the first level. When, inadvertently, the intentions cancel each other before they can produce any action, we feel a loss of volition, a paralysis of the will. At the highest levels our intentions are reasonably clear, but the the lower levels they may demand contradictory intentions, and so produce none at all, or only an unsatisfactory compromise. We easily become lost in the complexities of managing this physically compact but functionally gigantic structure, the human brain. How many of us could sit down and draw a map of our structures of intentions? Most of us could probably explain fragements of the structure here and there: this act serves that purpose, which in turn was selected as part of satisfying a higher-level intention, and so on for perhaps three or four levels at the most. A few of us might be able to show how the goals we seek at work relate to those we seek on weekends, or how our relationship to our parents interacts with our relationships to our wives and children. It is unlikely that any person alive could draw the whole map, even considering just the parts of it that are actually available to inspection. When we consider our own lives, we see them as if through a moving peephole that limits the size of the picture visible at a given moment, or as if we are shining a penlight around in a dark cathedral, trying to build up a picture of the whole huge room out of images that pass through the small circle of light. The sciences of life, being founded primarily on the old causal model, have little to tell us about understanding the vast structure of the mind. Having long ago dismissed the importance of phenomena such as volition, they have produced essentially nothing that would help us to map out our own organizations, either to understand or to improve them. Control theory, on the other hand, seems to show us the way toward doing something useful in this direction.

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hershberger (Editor) Elsevier Science Publishers B. V. (North-Holland), 1989

0

39

CHAPTER 3

ON THE WILL: AN HISTORICAL PERSPECTIVE Eckart Scheerer Why did the will depart from modern psychology, to be resurrected, in this volume, under the title "volitional action"? Howard and Conway (1986) claim that the demise of the will was caused by the principles of extrospection, rational objectivity, and determinism, taken over by psychology from the physical sciences as a result of the "Baconian revolution." If volitional action is identified with the "free will" as postulated in ethics and theology, then it comes, indeed, into conflict with the principle of determinism and perhaps also of objectivism. From a psychological viewpoint, though, volition can be studied independently of the free will concept, and, in fact, it continued to be a subject of psychological study long after psychology had taken over the methodological standards of the physical sciences. Consequently, the story of the will's demise has to be told in different terms than those chosen by Howard and Conway, the more so because it was a tragedy with dramatic peripeties rather than a simple though perhaps protracted "decline and fall." In approaching the topic of volition from an historical perspective, we are struck by the variety of meanings attached to the term and its derivatives in the Western languages. As far as I see, at least three different basic meanings can be discerned. First, there is a class of movements that are conventionally called "voluntary," to distinguish them from "involuntary" movements such as reflexes or instincts. Second, we have a class of actions mediated by processes such as deliberation, decision, and choice, and opposed to more "impulsive" actions. And third, we have a mental faculty or subsystem called "the will" which is presumed to subserve voluntary movements and volitional actions, and may occur in various degrees of intensity or "will power." In order to understand the heterogeneity of the phenomena (or constructs) subsumed under the category of volition, we need to have a brief look at the philosophical antecedents of the concept.

40

Eckart Scheerer

The Concept of Will: Its Philosophical Background Consider a student in one of the old European universities taking an exam at any time between the 13th and the 18th centuries. His professor confronts him with some easy questions. How many basic mental faculties or "powers" are there? Two, of course, cognition and desire (or appetite, or conation). What is the will? Rational desire. M a t is more important, the intellect or the will? The intellect, because willing occurs only in creatures that have an intellect; also, because volition depends on cognition but not the other way round. To what province of philosophy does the study of the will belong? To natural philosophy (physics) as well as to ethics; in physics we study the movements subserved by the will; in ethics we study the means and ends of volitional (i.e., purposeful and rational) actions. Our fictitious dialogue reflects the core of the received opinions about volition in the West, beliefs that were in a large measure due to Aristotle; that is, beliefs that outlived the Aristotelianism of the middle ages, even after it had been superseded by the "modern," experimental, mathematically oriented science of Descartes, Galilei, Kepler, and their followers. Aristotle had written about volition in his ethical works, but also from a perspective we would today call "biological," in his work On the Soul and in his little book On the Movements of Animals (Nussbaum, 1978). He said more about volition than was sketched in our dialogue. For instance, he was the first to distinguish clearly between voluntary and involuntary actions. Actually, he had even a tripartite division; involuntary movements (e.g., of the heart and of the genitals) were under the control of sense perception but not of thought, while non-voluntary movements (such as sleep/waking and respiration) were devoid of any such control. As far as the will was concerned, its assignment to a mental faculty different from cognition was motivated by its affinity to movement, its dynamic character, and its dependence on practical as opposed to theoretical reasoning. Aristotle also gave much attention to the cognitive processes involved in volitional action. His most influential ideas were the distinction between deliberation (in which we can wish impossible things) and choice (which pertains to the possible only), the assignment of freedom to the act of choice, and the conception of the "practical syllogism," which contains two premises (a general rule and a specific case) and the action itself as the conclusion. From late antiquity on, and for more than a millennium, Aristotle's analysis of volition remained a model to be improved and filled in with

On the Wll: An Historical Perspective

41

more detail, but not to be repudiated in its general outline. In Thomas of Aquinas we find an elaborate scheme of the "top-down" control in volitional action, encompassing judgment, will in the narrow sense, intention, deliberation, consent, choice, and even some kind of anticipated reinforcement; at the bottom of the hierarchy stands a simple "command" to the body, which then utilizes this command on its own account. To be sure, there were controversies around the will in the scholastic period. For instance, Duns Scotus assigned priority to the will rather than to the intellect, and he thought that the will intensifies and clarifies ideas which are in the background of consciousness; but he considered the will to be a rational faculty (Wolter, 1986), and so he stayed in the traditional "cognitivist" perspective on volition, despite his "voluntarism." The philosophers of the scientific revolution and the empiricists and sensualists of the 18th centuries may have ridiculed Aristotle, but they did not doubt his basic presuppositions such as the duality of cognition and volition, the rationality of volition, and the power of the will to move the body. Fundamentally new perspectives on the will arose only in the second half of the 18th and in the first half of the 19th century. One of these was the replacement of the dichotomous classification of mental processes by what Hilgard (1980) has called the "trilogy of mind," that is, cognition, affection, and conation, or thinking, feeling, and willing. The trilogy of mind was "dogmatized" by Immanuel Kant (1781) who drew upon developments in English philosophy and German empirical psychology which had resulted in the introduction of feeling as an independent class of mental phenomena. (In the Aristotelian scheme, feeling had been a concomitant of every mental process). In itself, the trilogy did not mean a deposition of the will; for instance, Kant himself still accepted the traditional definition of willing as an "appetitive faculty based on reason" (1781, p. 38). But it carried the potentiality of assimilating the will into the affective, emotional side of mental life. In turn, this could mean that the will was separated from reason and turned into an irrational factor of mental life. Another development was the emergence of a new voluntarism around the turn of the centuries. Maine de Biran, subsequently called the "Kant of France,'' conceived of the will as the basic force in mental life responsible for the emergence of the dualism between subject and object, or Ego and Non-Ego. In "immediate apperception" (i.e., introspection), on which Maine de Biran founded his psychology, the will is represented as a feeling of effort, which in turn is signalled by the muscle sense (sens musculaire). Like the German philosopher J. J. Engel, who

42

Eckart Scheerer

had introduced the term "muscle sense" in 1802 (the priority usually given to Charles Bell or Thomas Brown is wrong; see Scheerer, 1987a) Maine de Biran appealed to the muscle sense in order to explain the formation of the concept of force through the resistance we encounter in the active handling of objects. Neither Engel nor Maine de Biran were concerned with the question whether the muscle sense was efferent or afferent. However, their descriptions permit the inference that they were primarily concerned with a correlate of the "effort of the will" and thus with an efferent impulse, although James (1880/1983, footnote 22) believed that Maine de Biran was referring to afferent impulses. Actually, James was projecting back a distinction not relevant to these earlier thinkers because the afferent/efferent distinction became obligatory only as a result of Charles Bell's (1826) work resulting in the law named after him and Magendie. Maine de Biran was not the only voluntarist in the first half of the 19th century. Apparently independent of him, the German philosopher J. G. Fichte derived the Ego/Non-Ego dichotomy from the experience of activity, but unlike Maine de Biran he did not consider himself a psychologist and was not interested in physiological or anatomical questions. A similar comment pertains to the metaphysical voluntarism of Schopenhauer. While important in the history of philosophy, these thinkers had little influence on empirical psychology. Concepts like "feeling of effort" and "muscle sense," on the other hand, had a profound influence on the emergent scientific psychology of the 19th century, and though they were initially formulated in the context of voluntarism, eventually they led to the demise of the will as an independent category of psychology.

Spontaneity, Innervation Sensations, and Reaction Times: The Will in Early Experimental Psychology It is well known that experimental psychology started with the investigation of sensations. However, this choice had primarily methodological reasons--among all mental processes, sensations were most easily amenable to experimentation--and did not necessarily imply a philosophical orientation toward sensualism and empiricism. In Germany, where experimental psychology originated, British empiricism and French sensualism generally were held in slight regard; neither philosophy belonged to the philosophical presuppositions of, for instance, Fechner and Wundt. Even Helmholtz, the one pioneer who had a high opinion

On the Will: An Historical Perspective

43

of British philosophy, had a place for the will in his theory of perception. This is perhaps not so surprising if we consider that Alexander Bain, the chief systematizer of British association psychology, wrote a book entitled The Emotions and the Will and accepted the Kantian tripartite division of the mind. Until around 1880, the 'hew psychology" still accepted the concept of the will. Let us look at some details. One component of the will concept is that it is an inner determinant of behavior. In order for an impulse of the will to be generated from within, there must be some spontaneous activity of the nervous system. In the middle of the 19th century, the concept of spontaneous activity was not considered to be incompatible with a scientific outlook on psychology. Three examples will suffice. In 1859, Bain drew up a list of evidence for the spontaneity of movement, encompassing awakening (where movement was supposed to precede sensation), the activity of young animals in general, and certain temperaments where activity was stronger developed than sensibility. In 1888, he still summarized his position thus: "Movement precedes sensation, and is at the outset independent of any stimulus from without" (Bain, 1883, p. 303). In 1860, Fechner built his psychophysics on the notion of oscillatory "psychophysical excitation," which he conceptualized as being endogenous and modulated, rather than elicited, by external stimulation (Scheerer, 1987 b). And in 1863, Wundt declared that (spontaneous) drives were the primordial form of mental activity and the base from which both representation and volition were derived. During the same period it was generally assumed that at least some perceptual phenomena required the participation of efferent processes. The idea was a logical continuation of the "feeling of effort" concept developed within the framework of philosophical voluntarism, but it was now applied to the explanation of phenomena observed and occasionally measured in the laboratory and under clinical conditions. Efferent impulses subserving perception were called by Wundt, in 1863, "innervation sensations" or "innervation feelings"; other authors (e.g., Helmholtz) linked them more explicitly to the will by using terms like "effortloor "impulse of the will". Unfortunately, for the sake of terminological clarity, many French authors had a tendency to apply the term "muscle sense'' to efferent processes. Bain (1855) spoke of "outgoing1' (as opposed to "ingoing," i.e., afferent) impulses, a usage quite near to the current "outflow" vs. "inflow" terminology. Innervation sensations (see Scheerer, 1987a, for a fuller treatment) were invoked in four different contexts: (a) the stability of the visual

44

Eckart Scheerer

world despite a moving retinal image produced by active eye movements, (b) differences between active and passive appreciation of weights, (c) clinical data from anaesthesia, paralysis, and amputation, and (d) the effects of paresis of eye muscles. Despite wide variations of individual viewpoints, the theoreticians of innervation sensations shared two common assumptions: The motor impulse originating in the brain (a) is available to consciousness, and (b) it somehow combines with afferent impulses to produce a given perceptual phenomenon. The need to postulate innervation sensations arose whenever the afferent excitation pattern was ambiguous and its perceptual effect depended on the absence vs. presence of voluntary movements (or the intention to produce them) on the part of the observer. Let us clarify this statement by adducing a classical example--egocentric visual localization. How do we determine, on the basis of the retinal image, whether seen objects are in movement or at rest? Johannes Miiller (1838) formulated three rules: (a) Movement of the retinal image results in perceived movement when our eyes and our body are at rest. (b) When we follow a moving object with our eyes and as a consequence its retinal image does not move, then we use "either the sensations from the moving eye muscles or the impulses sent to the eye muscles from the sensorium" (p. 363), to judge that the object moves. (c) When both the retinal image and the eye muscles move in correspondence with each other, then we judge that the object is at rest. Thus, Miiller accorded the centrifugal impulse a role in egocentric visual localization, but not to the extent of excluding the afferent impulses originating from eye movements, The scale tipped in favor of an efferent explanation (i.e., by means of innervation sensations) when it became clear that passive movements of the eye (produced by applying pressure to the eye ball) and the mere intention to move the eyes in the absence of actual eye movements (in cases of eye muscle paresis) resulted in perceived visual movement, albeit in a direction opposite to each other. The relevant observations had been made briefly after 1850, and for about a quarter of a century everybody (e.g., Wundt, 1863; Helmholtz, 1867; and Mach, 1886) was convinced that active (voluntary) and passive movements differ from each other with respect to their perceptual effect, and that these differences can and must be explained by the match or mismatch between the efferent impulse (i.e., innervation sensations) and its reafferent effect (e.g., the movement of the retinal image). A final research domain in which the will figured prominently was the fractionation of reaction times into stages, a practice that had been

On the Will: An Historical Perspective

45

started by Donders in 1865 (Brozek, 1970). Among the mental operations that were "timed" in this way we regularly find included such traditional components of volition as choice and decision. Wilhelm Wundt considered the reaction time method as a useful technique for measuring the temporal dynamics of the "apperceptive" level of mental functioning. Because apperception to him was an internal act of volition, he assigned the study of reaction times to the section on the will in his Grundziige der physiologischen Psychologie of 1874.

The Demise of the Will: Reflex Physiology, Evolution Theory, and Neurology It may be said that the pioneers of experimental psychology took the will for granted. That is, they used the will in their explanations of other mental phenomena, rather than analyzing it in its own right. The downfall of the will began when it was converted from an explanatory concept into a subject matter of independent investigation. This development was not intrinsic to psychology but depended on the ascendancy of reflex physiology, evolution theory, and clinical neurology. The contributions of these disciplines may be briefly summarized as follows. For methodological reasons, the study of reflexes was for a long time restricted to spinal animals, particularly the decapitated frog. However, around 1860 a breakthrough was made in that the effects of brain stimulation on spinal reflexes were studied. The Russian physiologist Sechenov, who was one of the first to work along this line, in 1866 published a monograph entitled The Reflexes of the Brain, in which he asked for the application of the reflex arc model to all kinds of mental phenomena and, inter uliu, described volition as a reflex arc where the first, afferent, part was inhibited. Sechenov was perhaps the most vociferous spokesman of a general trend toward assimilating psychology into reflex physiology, a trend which tended to relativize the distinction between the brain (as the seat of voluntary action) and the rest of the central nervous system, and asked for an analysis of voluntary action in terms of its sensory antecedents. Traditional boundaries were also washed out by evolution theory. Ever since Aristotle had tied volition to intellect, creatures lacking reason were supposed to be incapable of volitional action. The very notion of instinct originally conveyed the idea that God or Providence had provided animals with special gifts for purposeful action in order to make up for

46

Eckart Scheerer

their lack of reason. In the Cartesian tradition, where animals were even denied a soul, they were a fortiori devoid of volition. Stressing the continuity between brute and man, Darwin and his followers did away with the assignment of instinct to animals and of volition to people. Rather, the task was now to derive volition from various forms of involuntary actions (reflexes, drives, instincts) by means of a genetic analysis. Finally, clinical neurology (and psychiatry; the two disciplines were not separated in the period) tended in various ways to obliterate the clear demarcations contained in the traditional picture of the mind and its bodily substrate. Two somewhat disparate instances were particularly important. The first pertains to the interrelations between sensation and movement as revealed by neurological disorders (see Scheerer, 1987a, for more details). Motor disturbances were shown to have sensory causes; for instance, ataxia (on the face of it a purely motor disorder) was shown to result from damage to the sensory tracts of the spinal cord, and thus the importance of sensory feedback from the muscles for the control of movement was stressed. The second line of attack arose from the preoccupation of neurologists with hypnotism (or "somnambulism," as it still was called by some). Either during trance or as a consequence of post-hypnotic suggestions, hypnotic subjects displayed voluntary behavior in the absence of the subjective experience of volition. Once again, the dividing line between "voluntary" and "involuntary" seemed to become fuzzy. Taken together, these developments resulted in a "paradigm shift" in the psychology of volition. The beginning of the shift is indicated by publications such as G. E. Muller (1878), Ribot (1879), James (1880/1983), and Schneider (1880; 1882), and its consummation by Munsterberg (1888; 1889) and James (1890). It comprised a genuine shift of perspective; the will was now seen as a result of sensations and images, while before sensations and images had been made dependent on the will. As a result, the will lost its status as an independent element of mental life. Let us briefly look at some details in terms of the concepts introduced in the previous section. How could voluntary actions fall under the control of sensations or images? The detailed answers to this question varied somewhat among the proponents of the paradigm shift, but the general outline was always the same (Hildebrandt, 1985). It appealed to a "forward movement'' of the sensory consequences of actions. Suppose that the original motor equipment of the mind consists either in reflexes or in spontaneous

On the will: An Historical Perspective

47

movements. In either case, the execution of the movements will give rise to movement sensations. The movement sensations are associated with the conditions under which the movement was made, and, if these conditions are reinstated, they elicit the movement sensations but now in the form of an image of the movement, which in turn elicits the movement itself. Thus, consequences of actions became antecedents of actions, and movement sensations were converted into images, representations, or expectations of past movements. As a result, volitional action was no longer spontaneous or inner-directed but depended on the previous occurrence of some sensory or imaginal event; and the will was assimilated into the associative network of mental life. The process just described bears surprising similarity to modern accounts in terms of Pavlovian or operant conditioning, but it was a theoretical construction and did not rest on laboratory evidence on the learning process. Consequently, it was confronted with the factual question of whether movement images had, indeed, the power to elicit movements. An affirmative answer to this question was provided by the principle of ideomotor action. Introduced by Carpenter in 1852, the principle initially was meant to provide a naturalistic explanation for seemingly paranormal phenomena such as table lifting. Carpenter made it clear that ideomotor action was involuntary and under normal conditions was inhibited by the will. However, also in 1852, Lotze drew attention to everyday phenomena such as the involuntary imitation of seen movements and the automatic nature of many everyday actions such as walking. Such movements, he thought, were initiated by motor images, and he ventured the suggestion that "more complicated series of movements (even those comprising the content of a crime)" were initiated in this way (Lotze, pp. 294-296). Volitional action was reduced to the voluntary combination of involuntary elements. Subsequently, among psychologists and psychiatrists the conviction that "every representation of a motion awakens the actual motion which is its object, unless inhibited by some antagonistic representation simultaneously present to the mind" (James,1880/1983, p. 103) became so widespread that James could use it as one of the cornerstones of his theory of the will. What became of innervation sensations once motor images had been recognized as antecedents of voluntary action? They were no longer needed and fell victim to Occam's razor; or, to paraphrase William James (1880/1983), on a priori grounds alone they were a "pure encumbrance'' (p. 88). On the other hand, even James felt the need to base his argument against them on a posteriori evidence also. Two basic lines of

48

Eckart Scheerer

reasoning were put forth by him and others, such as G. E. Muller (1878). First, given that innervation sensations were needed to disambiguate afferent input patterns, they would become superfluous if, under conditions in which they were invoked, the afference in fact was not ambiguous. Thus, in semiparesis of the eye muscles the healthy eye was still able to move, and it did so even when it was occluded. Consequently, the afferent situation was not ambiguous and evaluation of eye position could be done on the basis of the healthy eye. In modern terms, there is a mismatch not between efference and retinal re-afference but between two sources of afference (retinal and proprioceptive). Second, in cases where afferent information was lacking altogether or no sensory source for disambiguation could be found, appeal was made to expectations concerning the movement. Thus, paralytics felt that their paralyzed limbs were heavy because they expected them to move when they wanted to move them. Expectations were different from innervation sensations because they were based on memory images or traces of earlier afferent impulses. The fractionation of reaction times (a topic which James did not touch in his discussion of the will) was the last stronghold of the will as an independent mental element. Ironically, their downfall was initiated in Wundt's own laboratory, when his student L. Lange showed, in 1888, that simple reactions were faster when attention was paid to the movement ("muscular reactions") rather than to the stimulus ("sensorial reactions"). Muscular reactions were interpreted as brain reflexes, but sensorial reactions and choice reactions certainly reflected the dynamics of representations under voluntary control. This, at least, was the opinion of Wundt himself. Not so, argued Munsterberg (1889). He demonstrated that "muscular" reactions could be produced even under conditions of choice and semantic categorization and concluded that the consciously experienced impulse of the will followed rather than preceded the reaction. His general conclusion was that there was no difference between the involuntary and the voluntary dynamics of cognitive processes, and that the experience of volition consisted in stimulusspecific and generalized ("expectation") sensations of tension and strain.

Three Perspectives on the Will: Miinsterberg, James, and Wundt Around 1890 the case of innervation sensations "was not open and shut" (Boring, 1942, p. 528). Some authors retained the term but

On the will: An Historical Perspective

49

reinterpreted it such that it now meant the afferent sensations underlying the "feeling of effort." More commonly, it was dropped and replaced by "kinesthesis," introduced in 1880 by Bastian as a blanket term covering motor sensations and images of peripheral origin. Nevertheless, even after the will had lost the status of an independent mental element, there was still the need to explain volitional action in terms of other mental processes. The three main types of "theory of the will" are exemplified by the names of Munsterberg, James, and Wundt. All of them pointed to further developments in the psychology of volition, though later authors were rarely aware of their historical predecessors. The most "radical" (in the sense of reductionist) theory was Munsterberg's (Scheerer, 1984). His initial theory of the will, which earned him his early fame and the call to Harvard, was strictly associationist and "dissolved" the experience of volition into a complex of peripherally excited sensations. In 1900, he switched to a "centralist" theory (the "action theory") in which sensations were supposed to gain access to consciousness only if they were accompanied by a central motor discharge; yet the discharge in question was still instigated by afferent processes, and the whole scheme may be said to antedate the stimulusresponse bond of the behaviorists. More importantly, Munsterberg adhered to a strict epistemological division between scientific psychology and the humanities. All aspects of human experience that relate to active striving and the realization of values were to be dealt with by the humanities; scientific psychology was restricted to the study of conscious contents, as registered by a passively onlooking Ego. Insofar as volition encompassed the experience of an actively striving subject, it could, as a matter of principle, not be studied by a psychology adhering to the methods of the natural sciences. Volition, in the proper sense, was an exclusive subject-matter of the humanities. William James's (1890) chapter on the will is too well known to need a summary here. James succeeded in steering a middle course between abolishing the will altogether and treating it in a speculative and constructive manner, after the fashion of the old arm-chair psychology. The first strand of his thinking is represented by his polemics against innervation sensations and his adoption of the ideomotor principle, the second strand by his mental fiat and his discussion about different types of decision. Moreover, he indulged in the type of theorizing about the "neural machinery" that was so dear to his contemporaries and rendered his theory of the will scientifically respectable. The newly coined term "kinesthesis" allowed him to keep clear of the trappings of the ''muscle

50

Eckart Scheerer

sense" concept and to subsume all kinds of reafferent sensations under one unitary label. On the other hand, he was probably the first to make a distinction between proprioceptive (Ikesident") and exteroceptive ("remote") reafference and to point out the importance of the latter. Another distinction he made proved to be prophetic. Volition, so he wrote in 1880 in referring to his mental fiat, ''is a psychic or moral fact pure and simple," and 'the supervention of motion upon its completion" belongs "to the department of physiology exclusively" (p. 107). In fact, he himself was perhaps the last to bring about a synthesis of the mental and the physiological aspects of volition, and, after him, the psychology of the will was destined to be either physiological and reductionist or outrightly mentalistic. Wilhelm Wundt's theory of volition presents a paradox. Before pointing out the paradox, it must be said that Wundt was not immune to the criticism of the innervation sensation concept. Although he did not abandon the term, after 1883 he reinterpreted it to bring it more in line with the thinking of his contemporaries. In the final version of his theory (Wundt, 1910, pp. 37-41), and in almost literal anticipation of Teuber's (1966) notion of "corollary discharge," he maintained that there were ''movement sensations of central origin" but interpreted them as resulting from the collateral discharge of motor impulses into neighboring sensory areas of the cortex. At the same time, he denounced James for appealing to some "transcendent, abstract will" in his fiat conception (Wundt, 1911, p. 272). And herein lies the paradox, for Wundt applied the label "voluntarism" to his own psychology. The paradox is dissolved when we notice that he did not consider the will as a particular type of mental element but rather as the prototype of all mental activity. More specifically, he assimilated willing into affective processes, which displayed a typical pattern of build-up and release of tension. Depending on whether relaxation was brought about by muscular movements or by changes in the course of ideas and emotions, Wundt distinguished between outer and inner voluntary actions. Somewhat surprisingly, he assigned genetic priority to inner voluntary action, that is, apperception, where relaxation resulted from bringing ideas into the focus on consciousness by means of inhibitory processes generated in the frontal lobes. From the standpoint of motives, Wundt distinguished between drives (one motive only), voluntary actions in the strict sense (several motives but one gains ascendancy) and choice (consciously experienced conflict between motives). Again, all forms of volitional activity, when analyzed introspectively, dissolved into sensations and emotions; the specificity of the will

On the Will: An Historical Perspective

51

consisted not in the elements, but in their combinations and above all in their time course. Before discussing the continuation of the themes adumbrated by Munsterberg, James, and Wundt, we must briefly look at the final act in the downfall of the will, which proves to be an American tragedy.

How the Will Disappeared from American Psychology When John B. Watson (1913) supplemented his behaviorist manifesto by the analysis of some current psychological concepts, he chose "image and affection" and did not deem it necessary to discuss volition in behaviorist terms. This was no coincidence. The will had already departed from American psychology, and its exile was prolonged but not caused by behaviorism. Watson wanted to open "a free passage from structuralism to behaviorism," and so it is appropriate to look at Titchener's psychology first. Although Titchener posed as a spokesman of Wundt in the United States, he had little use for the latter's psychological voluntarism. In fact, he restricted apperception--the key concept of his master's system--to "attributive clearness" and disposed of the idea that it was an internal act of volition. Furthermore, he was a strict champion of kinesthetic sensations and images which invariably were reported by his introspective observers; and finally, he opposed Wundt's three-dimensional theory of feeling and thus the tension/relaxation dimension on which Wundt's affective theory of the will was based. In sum, in his system of psychology, which in its final version reduced the subject-matter of psychology to sensation, there was as little place for volition as in the behaviorism of Watson. Whatever their differences on other scores, they were united in favor of a strict peripheralism. There was, however, outside of Titchener's structuralism, a more specific development which tended to eliminate essential ingredients of James's theory of the will. First, the "idea of movement" had to go, when Woodworth (1906) concluded, from experiments involving the introspection of the "immediate antecedents" of voluntary movements, that an act may be thought of without any representative or symbolic image. Furthermore, when "kinesthetic images" were reported, they were often quite unlike the movement to which they referred. Woodworth started his work independently of the Wurzburg school in Germany, but when the latter became known to him, he pointed out that he had discovered "imageless thought" in the sphere of volition (Woodworth, 1907). The

52

Eckart Scheerer

second phase consisted in Thorndike's (1913) attack on the principle of ideomotor action. Thorndike felt that belief in the principle was a mere superstition. In his view, the ability of an idea to produce a movement depended on the laws of habit formation rather than on the "amount of likeness between the members of the pair" (Thorndike, 1914). Woodworth and Thorndike were no behaviorists, inasmuch as they accepted the testimony of introspection. However, even if introspection was not rejected, the "imageless thought" doctrine, as applied to volition, did away with all antecedents of action except habits and current stimulation; and, the demise of the ideomotor principle meant that the relation between action and its antecedents was entirely arbitrary. Behaviorism rested on the rejection of cognitive mediators and maintained the essential arbitrariness of stimulus-response connections; and, in both respects, Watson's "principal contentions" had been prepared by "tendencies initiated by the psychologists themselves" (Watson, 1913, p. 423).

The Resurgence of the Will in European Psychology In the first half of the 20th century European and American psychology drifted apart. There was no behaviorist revolution in Europe, and new theoretical developments were largely indigenous. One symptom of this is that there was a fresh interest in volition that did not have a parallel in the United States. One characteristic feature of German psychology between the two world wars was the emergence of a new psychology based on the humanities @ezkteswissenschafilichepsychologie). Miinsterberg's conception of a bipartite psychology came true in the Weimar period (1918-33), which witnessed a 'krisis" resulting from the conflict between a scienceoriented and a humanities-oriented outlook on psychology. The ''humanistic'' psychologists were not really interested in the classification of mental phenomena, and so they did not discuss the question of the will's independent status. But Wilhelm Dilthey, from whom they derived their inspiration, had endorsed the "trilogy of the mind," and reference to the will was not considered problematic by the humanistic psychologists. Like Munsterberg, they linked the will to the concept of value; volitional action was seen as the "realization" of values which were defined in terms of objectively existing mental structures. Particular emphasis was laid on the typology of value systems, such as the "economic"versus the "aesthetic form of life" (Spranger, 1927), and on the conflict of values in mental development (Spranger, 1928). By stressing the real-life qualities and the

On the Will: An Historicul Perspective

53

supra-individual determinants of volitional action, the new psychology based on the humanities provided an important corrective to laboratory investigations of volition. Such investigations were conducted in great number during the Weimar period in continuation of work started before World War I. The protagonist of the flourishing experimental study of 'the will was Narziss Ach. Outside of Germany, he is known best for his first work (Ach, 1905) in which he used the method of systematic introspection in the context of reaction time experiments. One result--the ''presence of impalpably given knowledge," named "awareness" by Ach--belonged to the pioneer demonstrations of "imageless thought" by the Wurzburg school. But Ach was more interested in volition than in cognition, and he spent most of his subsequent career investigating its phenomenological and functional properties (Ach, 1935). To Ach, evidence for a volitional element of mental life rested, in the main, on his demonstration of "determining tendencies," which were supposed to derive from the internal representation of a goal and to direct the course of mental processes in accordance with the goal representation. Initially, Ach described the experience of willing, in somewhat Jamesian terms, as the "mental assent" to responding in consonance with some previously acquired determining tendency. In 1910, he devised an ingenious method for the study of volition: subjects were first given extensive practice with pairs of meaningless syllables and then were required to perform some other activity, such as rhyming or vowel substitution, on the stimulus terms of the previously acquired paired associates. When subjects were switched to the rhyming tasks, there was a strong tendency for the previous learned responses to be emitted, and in order to overcome these (associative) reproduction tendencies, the subjects deployed what Ach called a "primary act of the will." Primary, "energetical" volition was, in Ach's (1910, 1935) view, an irreducible mental state with unique structural and functional qualities. Structurally, it was defined by four "moments": relation to the task, experience of "I really want to do this," pronounced strain sensations, and an effortful attitude. Functionally, it was defined as a means to attain the goal by overcoming inhibitory conditions. The strength of the determining tendencies set up by a primary act of volition could be measured by their "associative equivalent," that is, by determining the maximum number of paired-associate learning trials which did not produce intrusions or prolonged reaction times after switching to the second task. In the final version of his theory, Ach stressed the largely unconscious nature of

54

Eckart Scheerer

determining tendencies; once a certain task had been adopted by the subject, the selection and control of subsidiary mental processes is performed at an unconscious level. Even the act of willing could become automatized by frequent repetition; automatization was indicated by the lack of interference from a concurrent secondary task. Ach also formulated quite a few quantitative laws of volition; for instance, the ''law of specific determination" (determinations are realized more quickly and safely the more specific they are) and the "law of motivation from difficulty" (increased difficulty results in a spontaneous increase of effort of the will). Ach used to refer to his theory of the will as the "psychology of determination," and this alone is an indication that he did not consider the will to be free. In fact, he contrived an experimental procedure, known as the "prediction method," which allowed the experimenter to correctly predict the choices of subjects under conditions where the subjects themselves stated that they were totally free in their choices. The "feeling of freedom" was a dependent, not an independent variable in his theory. Ach was the most active, but by no means the only psychologist of the will in Weimar Germany. In 1911, the Belgian psychologist Michotte had suggested another method for the investigation of volition, where subjects were given a choice between performing different arithmetical operations; he reported that their decisions sometimes where reached without any sensory or imaginary content (Michotte & Prum, 1911). Subsequently, many German experimental psychologists (including Lindworsky, 1923, the author of a much-read monograph on the will) defined the will in terms of an activity which is experienced as originating from the Ego. From this perspective, Ach was criticized for having introduced extrinsic elements in the theory of the will; the phenomena on which his theory was based could arise from purely associative factors. Another line of criticism centered on Ach's reliance on meaningless materials and strict laboratory experimentation; for instance, Rohracher (1932) collected introspective reports while his subjects were refraining from food intake for a whole day and concluded that in volition, as opposed to drive, the Ego is active. More consequential was the criticism of Ach's methods and theories with which Kurt Lewin (1917, 1922) started his scientific career, for it gave rise to a theory of the will which stressed--like Wundt's theory, but in very different terms--the affective side of volition. Lewin initially was interested in the measurement of will power as proposed by Ach.

On the will: An Historical Perspective

55

Accordingly, he repeated Ach's experiments with a slight modification: after having thoroughly memorized pairs of syllables, subjects were required to read them and to refrain from attempts at reproduction; there were no intrusions or lengthened reaction times as required by Ach's theory. Other experiments showed that Ach's results could either be replicated or not be replicated on the same previously learned syllable pairs, depending on whether or not they were embedded in a context of neutral syllables. Lewin concluded that neither the "law of association" nor a conflict between reproductive and determining tendencies were operative in Ach's experiments. Instead, the results depended on the dynamics of task-specific sets (such as a recall set or a rhyming set) and the subject's spontaneous adoption of the easier activity when one and the same result could be attained by different activities. Thus, the will seemed to be gone once more, but it was resurrected in Lewin's classical Berlin work on the psychology of affect and action. In the introduction to the well-known series containing the work by Zeigarnik, Ovsiankina, Dembo and others, Lewin (1926) expressed the opinion that the psychology of the will should be approached from a causal-dynamical point of view, instead of focussing on the experiential qualities of volitional acts such as decision and resolve. The central concern was to be the problem of self-control, and this would mean a new approach to the experimental investigation of drive and affect, which were intimately connected with the problem of self-control. The "dominant theory of the will''--Lewin here referred to Ach, of course-was wrong in assigning central importance to the coupling between goal representations and the occasions for the execution of actions corresponding to them. Such a coupling should increase in strength upon repeated occurrence of the occasion; but once I have thrown a letter into a letterbox, I will not repeat this action when a second letter-box comes into sight. Thus, intentions function rather like needs in that they generate a tension state and are satisfied when they are acted on; they are quasineeds. Another divergence with Ach concerned the relationship between the strength of an intention and the likelihood of overcoming inhibitory influences; according to Lewin, very strong intentions often resulted in poor performance. Given that intentions are quasi-needs, do they still qualify as acts of the will? No, said Lewin; the defining characteristic of volitional actions is that they are "controlled," that they occur in opposition to some real need or quasi-need. In other words, they are relatively free from the dynamics of the psychological field, while instinctual and impulsive actions

56

Eckart Scheerer

are completely determined by the field. Intentions, on the other hand, are defined by foresight and preparation; they set up a psychological field which without them would not exist or would exist in another form. Thus, we arrive at a fourfold classification of actions (field vs. controlled, intended vs. non-intended), where the combination "volitional action without intention" is not self-contradictory but corresponds to actually existing cases, such as the spontaneous act of saving somebody from a fire (against field forces but without preparatory intention). In reading Lewin's (1926) paper, one is struck to find quite friendly references to psychologists working in the humanities tradition. It seems that Lewin, like other Gestalt psychologists of the time, looked at his work as overcoming the crisis of psychology in the sense that he accepted the general aims of the psychological humanists but differed from them as far as scientific methodology was concerned. To the extent that he succeeded in integrating the divergent trends prevalent at his time, his work can be considered as the culminating point that the investigation of the will could reach in German interbellurn psychology. When he went to the United States, his interests shifted toward group dynamics and social psychology and lost their specifically German flavor, not only linguistically but in terms of the basic presuppositions guiding them. This meant that the will was again lost, and in fact the most accessible source on Lewin's contribution to the psychology of volition is Koffka's (1935) Principles of Gestalt Psychology, more so than Lewin's own writings in the English language.

Concluding Remarks The "cognitive revolution" in psychology was precipitated by developments occurring outside of psychology, such as communication theory and computer science (Scheerer, 1988). Once psychologists had convinced themselves that cognitive theorizing was not incompatible with scientific methodology, they started to look back at the history of their field and discovered that pre-behaviorist psychology had been cognitive all along. A similar constellation is to be expected in the impending "volitional revolution." The development was initiated by outside influences, in this case control theory, and it has to struggle with presuppositions that the will, by its very definition, is not amenable to scientific study. As far as these apprehensions are based on the supposed freedom of the will, the present essay has, I think, dispersed them; at no point in the history of psychology has the investigation of the will been

On the will: An Historical Perspective

57

prevented by a "freedom of the will." But history may point to another obstacle for the scientific study of the will. Perhaps the will is not entirely of a "natural kind" and will have to be dissected when we start to *'carve nature at the joints.'' This has been argued for the concept of consciousness (e.g., Wilkes, 1988) and it might equally well be argued for the concept of the will.

References Ach, N. (1905). h e r die Willenstatigkit und das Denken. Gottingen: Vandenhoeck & Ruprecht. Ach, N. (1910). h e r den willensakt und das Temperament. Leipzig: Quelle & Meyer. Ach, N. (1935). Analyse des Willens. Berlin, Wien: Urban & Schwarzenberg. Bain, A. (1855). The senses and the intellect. London: Parker. Bain, A. (1883). The emotions and the will (3rd ed.). London: Longmans, Green & Co. Bastian, C. (1880). The brain as an organ of the mind. London: Kegan Paul. Bell, C. (1826). On the nervous circle which connects the voluntary muscles with the brain. Philosophical Transactions of the Royal Sociev, 116(1), 163-173. Boring, E. G. (1942). Sensation and perception in the history of experimental psychology. New York: Appleton-Century-Crofts. Brozek, J. (1970). Wayward history: F. C. Donders (1818-1889) and the timing of mental operations. Psychological Reports, 26, 563-569. Carpenter, W. B. (1852). Electro-biology and mesmerism. Quarterly Review, 93, 501-557. Engel, J. J. (1802). Uber den Ursprung des Begriffs der Kraft. Quoted after: J.J. Engel, Schrifien, Vol10 (2nd ed.) (pp. 101-122). Berlin: Mylius 1844. Helmholtz, H. (1867). Handbuch der physiologischen Optik Leipzig: Voss. Hildebrandt, H. (1985). Ideomotorik: Ein neues Paradigma fur ein altes Problem? Perception & Action, Report # 65. Universitiit Bielefeld: Zentrum fur Interdisziplinare Forschung. Hilgard, E. R. (1980). The trilogy of mind: cognition, affection, and conation. Journal of the History of the Behavioral Sciences, 16, 107-117. Howard, G. S., & Conway, C. G. (1986). Can there be a science of volitional action? American Psychologist, 41, 1241-1251. James, W. (1890). The principles of psychology. New York: Holt.

58

Eckart Scheerer

James, W. (1983). The feeling of effort. In F. H. Burkhardt, F. Bowers, & I. K. Skrupskelis (Eds.), The works of WilliamJames: Essays in psychology (pp. 83-124). Cambridge, MA: Harvard University Press. (Original work published in 1880) Kant, I. (1781). f i t & der reinen Vernunfl. Riga: Hartknoch. Koffka, K. (1935). Principles of gestalt psychology. London: Routledge & Kegan Paul. Lange, L. (1888). Neue Experimente uber den Vorgang der einfachen Reaktion auf Sinneseindriicke. Philosophische Studien, 4, 479-510. Lewin, K. (1917). Die psychische Tatigkeit bei der Hemmung von Willensvorgangen und das Grundgesetz der Assoziation. Zeitschn’ftfiir Psychologie, 77, 212-247. Lewin, K. (1922). Das Problem der Willensmessung und das Grundgesetz der Assoziation: I. Psychologische Forschung, 1, 191-302. Lewin, K. (1926). Vorsatz, Wille und Bedurfnis. Psychologische Forschung, 7, 330-385. Lindworsky, J. (1923). Der WilEe (3rd ed.). Leipzig: Barth. Lotze, R. H. (1852). Medicinische Psychologie oder Physiologie der Seele. Leipzig: Weidmannsche Buchhandlung. Mach, E. (1886). Beitruge zur Analyse der Empfindungen. Jena: G. Fischer. Maine de Biran, F. P. (1807). De l’apperception immediate (MCmoire de Berlin), J. Echevierra (ed.), Paris: Vrin 1963. Michotte, A., & Prum, E. (1911). etude experimentale sur le choix volontaire et ses antkctdents immediates. Archives de Psychologie, 10, 113-320. Miiller, G. E. (1878). Zur Grundlegung der Psychophysik Berlin: Hoffmann. Miiller, J. (1838). Handbuch der Physiologie des Menschen, 2. Band, Coblenz: Holscher. Munsterberg, H. (1888). Die Willenshandlung:Ein Beitrag zurphysiologischen Psychologie. Freiburg: Mohr. Munsterberg, H. (1889). Willkiirliche und unwillkurliche Vorstellungsverbindung. In H. Munsterberg (Ed.), Beitriige zur aperimentellen Psychologie, (pp. 64-188). Heft 1. Freiburg: Mohr. Miinsterberg, H. (1900). Grundzuge der Psychologie. Leipzig: Barth. Nussbaum, M. (1978). Ahtotle’s De Motu Animalium. Princeton: Princeton University Press. Ribot, T. (1879). Les mouvements et leur importance psychique. Revue Philosophique, 8, 371-386. Rohracher, H. (1932). Theorie des Wllens auf qerimenteller Grundlage. Leipzig: Barth.

On the will: An Historical Perspective

59

Scheerer, E. (1984). Motor theories of cognitive structure: A historical review. In W. Prinz & A. F. Sanders (Eds.), Cognition and Motor Processes (pp. 77-98). Berlin, Heidelberg, New York, Tokyo: Springer-Verlag. Scheerer, E. (1987a). Muscle sense and innervation feelings: A chapter in the history of perception and action. In H. Heuer & A. F. Sanders (Eds.), Perspectives on Perception and Action (pp. 171-194). Hillsdale, N.J.: Lawrence Erlbaum. Scheerer, E. (1987b). The unknown Fechner. Psychological Research, 49, 197202. Scheerer, E. (1988). Towards a history of Cognitive Science. International Social Science Journal, 40(1), 7-20. Schneider, G. H. (1880). Der thierische wille. Leipzig: Abel. Schneider, G. H. (1882). Der menschliche Wile vom Standpunkte der neueren Entwicklungstheorien (des "Danuinismus").Berlin: Diimmler. Sechenov, I. M. (1956). Reflexes of the brain. In I. M. Sechenov, Selected physiological and psychological works (pp. 31-139). Moscow: Foreign Languages Publishing House. (Original work published 1866) Spranger, E. (1927). Lebensfomzen (6th ed.). Halle: Niemeyer. Spranger, E. (1928). Psychologie des Jugendalters (10th ed.). Leipzig: Quelle & Meyer. Teuber, H. L. (1966). Alterations of perception in brain injury. In J. Eccles (Ed.), Brain and conscious experience (pp. 182-216). Berlin, Heidelberg, New York: Springer-Verlag. Thorndike, E. L. (1913). Ideo-motor action. Psychological Review, 20, 91-106. Thorndike, E. L. (1914). Ideo-motor action: A reply to Professor Montague. Journal of Philosophy, Psychology and Scientific Methods, 12, 32-37. Watson, J. B. (1913). Image and affection in behavior. Journal of Philosophy, Psychology and Scientific Methods, 10, 421-428. Wilkes, K. (1988). ---,yishi, dum, and consciousness. In A. J. Marcel & E. Bisiach (Eds.), Consciousness in contemporary science (pp. 16-41). Oxford: Clarendon Press. Wolter, A. B. (1986). Duns Scotus on the will and morality. Washington, D.C.: Catholic University Press. Woodworth, R. S. (1906a). The cause of a voluntary movement. In: Studies in philosophy and psychology, Garmun memorial volume (pp. 35 1-392). Boston: Houghton-Mifflin. Woodworth, R. S. (1906b). Imageless thought. Journal of Philosophy, Psychology and ScientiJic Methods, 3, 701-709. Wundt, W. (1863). Vorlesungen iiber die Menschen- und Thierseele. Leipzig: voss.

60

Eckurt Scheerer

Wundt, W. (1874). Grundziige der physiologischen Psychologie (1st ed.). Leipzig: Engelmann. Wundt, W. (1910). Grundzuge der physiologischen Psychologie: Vol. 2 (6th ed.). Leipzig: Engelmann, Wundt, W. (1911). Grundzuge der physiologischen Psychologie: Vol. 3 (6th ed.). Leipzig: Engelmann,

PHYSIOLOGICAL PERSPECTIVE

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A . Hershberger (Editor) Elsevier Science Publishers B. V. (North-Holland), 1585

0

63

CHAPTER 4 VOLITIONAL EYE MOVEMENTS AND THEIR RELATIONSHIP TO VISUAL ATTENTION Burkhart Fischer and Rolf Boch The eyes are moving virtually all the time (even during sleep), although rarely are we aware of these movements. Conversely, we can also consciously move our eyes. Therefore, a movement of the eyeball (defined by its rotational position in the orbit as a function of time) may be either (a) the consequence of our own decision, (b) the effect of a reflex, or (c) a superposition of both. For example, although eye movements are largely involuntary movements, they generally can be generated or suppressed voluntarily. However, this generalization needs qualification, because certain types of eye movements occurring under certain conditions are never under our voluntary control. The question of which types of eye movements are to be regarded as voluntary presupposes a prior classification of the types of eye movement to be considered. Therefore, in our attempt to answer the question of volition, we start from a classification of eye movements based upon the physical parameters of the movements and on the physical conditions under which these movements occur. This classification is illustrated in Figure 1. Then, in a second step we (a) try to identify those types and aspects of eye movements that are under volitional control, and (b) discuss their relationship to visual attention.

Classification of Eye Movements Eye movements are ordinarily distinguished or classified in terms of their velocity, as being either fast, slow (say, below 2Oo/s), or essentially zero (stationary). The three circles in Figure 1 represent the three velocity domains: fast, slow, and close to zero.

64

Burkhart Fischer and Rolf Boch

MOVEMENTS of the EYE fast eye movements unvoluntary movements\ spontaneous eye movements self paced saccades anticipated saccades resting

eye

fixating

visual1y guided saccades corrective saccades

long

Figure 1. Diagram classifymg the different types of eye movements referring to their velocities. Fast eye movements are shown on the outer, slow movements on the middle, and no movement on the inner circle. Nystagmic eye movements break down into slow and fast phases, belonging to those eye movements collected on the middle and outer circle, respectively. An imaginary vertical midline divides the concentric circles into two halves with involuntary eye movements in the left and eye movements under at least some voluntary control in the right half. The thin horizontal line corresponds to the fact that drifts and microsaccades occur in the state of resting and fixating eyes, respectively. With respect to their reaction times, visually guided saccades are subdivided into express, regular, and long latency saccades (lower right). For details see text.

Volitional Eye Movements and Visual Attention

65

Fast Eye Movements Fast eye movements are called saccades. They occur when the direction of gaze is suddenly shifted from one object at one position in space to another object at another position. Saccades can be subdivided into a number of different types depending on the conditions under which they occur: Spontaneous saccades are made in total darkness. Anticipated and corrective saccades occur under laboratory conditions; the former occurs when a subject being asked to respond to a sensory stimulus with a saccade initiates the eye movement prior to the stimulus; the latter occurs as a result of the eye missing the position of a target to which a subject is saccading. Anticipated saccades are slower in velocity than normal saccades of the same size and they often fail to reach the target position. Instead they undershoot and then are followed by corrective saccades. Selfpaced saccades are--by definition--voluntary saccades that a subject makes (without an external signal) on his own decision. fisually guided saccades are the most common eye movements we make. Every 200 - 300 ms the eye jumps from one position to the next if we are free to look at our visual surroundings. Under laboratory conditions visually guided saccades are typically initiated in response to the onset of a light stimulus. In this case one can classify the saccades as express, regular, or long on the basis of their reaction time (latency) being either extremely short (ca. 100 ms), intermediate (ca. 150 ms), or long (ca. 220 ms), respectively.

Slow Eye Movements Slow eye movements are made when we try to follow an object that moves relative to the head regardless of whether this relative movement is a consequence of our own body movements, or the consequence of a movement of the object, or both. These slow movements are therefore called pursuit eye movements. Vergence eye movements (counterrotating the two eyes) are also of slow velocity even when they are made to shift the gaze as quickly as possible from a near to a far object or vice versa.

No Eye Movements Tiny eye movements occur during time periods when we try to keep the image of a (small) resting object on our fovea. That is, during periods of fixation, the eye is not always at rest (no eye movement); rather, during these "fixation" periods the eyes also can move by slow drifts and fast microsaccades.

66

Burkhart Fkcher and Rolf Boch

Combinations of Fast and Slow Eye Movements Complex combinations of eye movements occur reflexively when the head moves. These reflexes, distinguished on the basis of the receptor detecting the motion, are of two types: vestibular and optokinetic. Both consist of slow phases when the eyes pursue a target and quick phases that look like saccades that bring the eyes back into an appropriate operating position in the orbit. Under laboratory conditions these reflexes lead to sequences of alternating quick and slow phases, called nystagmus.

Eye Movements and Volition

As pointed out already, movements of the eyes may be generated voluntarily. This, however, is a generalization which holds true only for certain types of eye movements and only for certain aspects of them. In this section we will go through the classification scheme represented in Figure 1 in an attempt to assign the attribute "volitional" or 'hot volitional" to each type of eye movement, or at least to some aspect thereof. Both slow and fast eye movements, as well as the state of "no movement,'' can be achieved by volition: we can decide to pursue a moving object, we can decide to keep the eyes still, and we can decide to initiate a saccade. However, we have no voluntary control over the velocity of the smooth pursuit movement, we have no voluntary control over the occurrence or suppression of slow drifts, we cannot initiate microsaccades voluntarily (but they may be suppressed by instruction; Steinman, Haddad, Skavenski, & Wyman, 1973), and we have no voluntary control over the velocity of saccades; instead there exists a fixed relationship between size and duration of saccades (Fuchs, 1967), the so-called ''main sequence." Nystagmus is a combination of eye movements almost completely controlled by reflexes (vestibular and/or optokinetic) and therefore--by definition-will be assigned as "not volitional". Among the various types of saccadic eye movements, self-paced saccades are "volitional" and spontaneous saccades are Ifnot volitional", both by definition. Furthermore, anticipatory saccades are also "not volitional," again by definition. Corrective saccades occur only after visually guided saccades that have missed the target. Most of the time the observer is unaware of his mistake in amplitude, and corrects the error involuntarily. All the remaining types of saccades are correct, visually guided, fast changes of the direction of gaze distinguished only on the basis of their reaction time (express, regular, long), Even though these eye movements are volitional, because they can all be initiated or suppressed by the subject's own decision,

Volitional Eye Movements and Viual Attention

67

one cannot decide to initiate a certain type and not another. For example, a subject who usually produces a high percentage of express saccades cannot decide on a given trial to generate a regular saccade. In other words, he has no voluntary control over his reaction time other than to make it extremely long, that is, to suppress the saccade proper and to make another one some time later. In conclusion, not a single type of eye movement is generated and controlled completely by our will. Even self-paced saccades have a velocity which we cannot select independently from its size. Nevertheless we allocated the term "voluntay" to the types of eye movements at the right side of Figure 1 and "involuntary1'to those at the left side. What is meant is only that these types of eye movements are under at least some voluntary control whereas the others are completely involuntary. The volitional control of eye movements is restricted to certain types of movement and it is restricted to the initiation (in some cases) or to the suppression (in other cases) of the movement. This is a basic difference from the control of movements of the limbs, for example, of reaching movements (see chapter by Georgopoulos). Whether or not this difference is related to the fact that there exists no oculomotor cortex but a motor cortex is an interesting but open question.

Eye Movements and Attention Objects that draw our attention are usually the ones we look at. In terms of eye movements this means that we make a saccade to bring the image of the object to the fovea and than we fixate or pursue it. In other words: in everyday life eye movements can be regarded as more or less direct expressions of our attention. (Yet, this is not a fixed relationship because we can pay attention to things we are not looking at.) It is therefore not surprising that the studies of eye movements have been used to study mechanisms of visual attention (for review see Fischer & Breitmeyer, 1987). This is true in particular for the study of visually guided saccades, whereas-interestingly--the study of fixation has been neglected. The close relationship between fixation, attention, and saccades became clear only recently after the discovery of the express saccade defined by its extremely short reaction time of about 70 ms in monkeys (Fischer & Boch, 1983) and around 100 ms in man (Fischer & Ramsperger, 1984). Attempts to understand the conditions under which this type of saccade occurs have shown, that all saccades (except microsaccades perhaps) are arrested by directed or engaged visual attention. This is almost trivial as long

68

Burkhart Fischer and Rolf Boch

as the direction of gaze and that of attention are coincident in space. If, however, attention is directed to an object in the periphery of the visual field one expects a shorter reaction time of a saccade directed to that object. The opposite is true (Mayfrank, Mobashery, Kimmig, & Fischer, 1986): Saccadic reaction times are extremely long when--at the time a saccade target occurs in the field of view--the subject still maintains fixation or pays attention to something in the periphery including the saccade target. On the other hand saccadic reaction times become extremely short (express saccades) when the subject has already disengaged his attention from wherever it has been engaged before. This notion of two states of the attentional system--one "engaged'' with saccades inhibited, the other "disengaged" with saccades permitted--leads to the question of what really happens when we make a voluntary saccade. Is it, for example, possible that, what we experience as a volitional saccade, is preceded by an internal switch from engaged to disengaged attention?

Neural Events Related to Saccadic Eye Movements and Attention Several brain structures--cortical and subcortical--have been investigated during the last 25 years using microelectrode, single-cell recording techniques in monkeys who have been trained to perform specific visuo-oculomotor tasks. The main result is that the visual responses of the cells (in the superior colliculus (Goldberg & Wurtz, 1972), the prestriate cortical area V4 (Fischer & Boch, 1981), the inferior parietal cortex (Robinson, Goldberg, & Stanton, 1978), the frontal eye fields (Wurtz & Mohler, 1976b), the prefrontal area 46 (Boch & Goldberg, in press) are enhanced when the visual stimuli that elicit the responses become targets of saccades. Whereas this enhancement is very closely related to the occurrence of the saccade for the cells in the colliculus (Wurtz & Mohler, 1976a) and in the frontal eye fields (Goldberg & Bushnell, 198l), response enhancement can be observed without saccades for cells in the parietal cortex (Bushnell, Goldberg, & Robinson, 1981) and in the prestriate cortex (Fischer & Boch, 1985) when the monkey just disengages his attention from the fixation point and moves it to the peripheral target. In fact, one can modulate the activity of a prestriate cortical cell by switching the fixation point on and off while the monkey's eyes stay still all the time and a constantly illuminated stimulus is projected into the cell's receptive field (Fischer & Boch, 1985). This occurs after the monkey has learned that the stimulus may change its luminance during the time the fixation point is invisible and that he has to detect this change to receive a reward. It has been argued, therefore, that this type of response modulation reflects the

Volitional Eye Movements and Viiual Attention

69

animal's visual attention moving back and forth from the fixation point to the peripheral stimulus. In the frontal eye fields--by contrast--there are cells which become active before saccades even though the monkey has no visible target. But the saccade must be part of a behavioral task, otherwise the cells remain silent (Goldberg & Bruce, 1986). This experiment shows that the question of whether or not a particular saccade, or aspect thereof, is indeed purposive, and therefore voluntary, is clearly distinguished by cells in the frontal cortex. Patients with frontal lesions lose, to some extent, their voluntary control over saccades in the sense that they are unable to suppress a saccade to a stimulus appearing in the part of the visual field affected by the lesion (Guitton, Buchtel, & Douglas, 1985). Instead they make reflex-like saccades about 100 ms after stimulus onset, which are probably express saccades. The intimate relation of frontal and parietal cortex to the generation and suppression of saccades as a function of fixation and directed attention is further emphasized by the results of electrical stimulation experiments. Saccades usually elicited by electrical microstimulation of these cortical areas are abolished when the stimulation is applied while the animal performs a fixation task (Goldberg, Bushnell, & Bruce, 1986; Shibutani, Sakata, & Hyvaerinen, 1984). Both experiments show that during fixation, pathways that mediate the generation of saccades are inhibited. This of course makes sense, because the decision to fixate contradicts the decision to make a saccade. If one assumes that fixation is the combination of a resting eye and attention being engaged to a foveal stimulus one predicts that engaged attention alone inhibits saccades. It is exactly this conclusion that has been reached by Mayfrank et al. (1986), and by Fischer (1987). A summary of the cortical control of eye movements is given by Fischer and Boch (in press). In conclusion, the attentional processes that usually precede voluntary saccades appear to take place in the prestriate and parietal areas, whereas the volitional aspect comes into play by frontal and prefrontal cortical mechanisms. However, these parts of cortex are intimately and mutually connected and the complex anatomical situation (see Fischer & Boch, in press, for an overview) suggests that the terms "voluntary" and "attentional" may not have simple neural correlates located in different small and well defined cortical areas.

Modification of Saccadic Eye Movements Some eye movements, or aspects thereof, are subject to modification as a function of practice. For example, when learning to read as a child, one

70

Burkhart Fischer and Rolf Boch

learned to make saccades regularly and repetitively. Even as adults our saccadic system, or at least a mechanism which controls the generation of saccades, can be modified by daily practice. This applies mainly to the reaction time of visually guided saccades. It is not just that these reaction times may be reduced by training, but more specifically, and interestingly, that the occurrence of one type of saccade--the express saccade--can be increased in frequency of occurrence from close to zero in untrained subjects (Fischer & Ramsperger, 1986) or monkeys (Fischer, Boch, & Ramsperger, 1984) to almost 100 %. This means that the subject has learned to voluntarily control one or more processes which must be completed before a saccade can be made. Most probably these processes have to do with a change in the attentional system being switched from the engaged to the disengaged state (Fischer & Breitmeyer, 1987). Here we have an intriguing situation in that what would seem to be a prerequisite for successful learning--namely attention--is itself being modified by practice. So far, not a lot is known about the changes that occur in the neural activity of different brain structures during periods of practice. We do know, however, that a certain type of presaccadic activity in prestriate cortex is reduced as monkeys practice visually guided saccades thereby reducing their reaction times drastically (Fischer & Boch, 1982).

Conclusion Although we are intimately and immediately aware (i.e., conscious) of our own "will" and "attention," neither phenomenon, considered as an objective event or process, has directly measurable parameters. That is, although we know about their existence directly by introspection, this subjective "method is not accepted in science. Therefore, we try to get a scientific grip on these phenomena indirectly, by looking at their objective consequences in behavioral or neurophysiological terms. It is only natural, therefore, that attempts to understand such a simple thing as an eye movement should foster philosophical considerations of the essential nature of volition and attention. Or, to put it the other way around, a thorough investigation of the generation and control of voluntary saccades may well be expected to provide objective insights into the neurobiological aspects of volition and attention, terms usually discussed in philosophy and psychology.

Volitional Eye Movements and Viiual Attention

71

References Boch, R., & Goldberg M. E. (in press). The participation of prefrontal neurons in the preparation of visually guided eye movements in the rhesus monkey. Journal of Neurophysiology. Bushnell, M. C., Goldberg, M. E., & Robinson, D. L. (1981). Behavioral enhancement of visual responses in monkey cerebral cortex. I. Modulation in posterior parietal cortex related to selective visual attention. Journal of Neurophysiology, 46, 755-772. Fischer, B. (1987). The preparation of visually guided saccades. Review of Physiology and Biochemical Pharmacology, 106, 1-35. Fischer, B., & Boch, R. (1981). Enhanced activation of neurons in prelunate cortex before visually guided saccades of trained rhesus monkeys. Experimental Brain Research, 44, 129-137. Fischer, B., & Boch, R. (1982). Modifications of presaccadic activation of neurons in the extrastriate cortex during prolonged training of rhesus monkeys in a visuo-oculomotor task. Neuroscience Letters, 30, 127-131. Fischer, B., & Boch, R. (1983). Saccadic eye movements after extremely short reaction times in the monkey. Brain Research, 260, 21-26. Fischer, B., & Boch, R. (1985). Peripheral attention versus central fixation: modulation of the visual activity of prelunate cortical cells of the rhesus monkey. Brain Research, 345, 111-123. Fischer, B., & Boch, R. (in press). Cerebral cortex. In R. H. S. Carpenter (Ed.), Vision and visual dysfunction: Vol. 9. New York: Macmillan Press. Fischer, B., Boch, R., & Ramsperger, E. (1984). Express saccades of the monkey: effects of daily training on probability of occurrence and reaction time. Experimental Brain Research, 55, 232-242. Fischer, B., & Breitmeyer, B. (1987). Mechanisms of visual attention revealed by saccadic eye movements. Neuropsychologia, 25, 73-83. Fischer, B., & Ramsperger, E. (1984). Human express-saccades: extremely short reaction times of goal directed eye movements. Experimental Brain Research, 57, 191-195. Fischer, B., & Ramsperger, E. (1986). Human express saccades: effects of randomization and daily practice. Experimental Brain Research, 64, 569-578. Fuchs, A.F. (1967). Saccadic and smooth pursuit eye movements in the monkey. Journal of Physiology, 191, 609-631. Goldberg, M. E., & Bruce, C. J. (1986). The role of the arcuate frontal eye fields in the generation of saccadic eye movements. In H. -J. Freund, U. Bittner, B. Cohen, & J. Noth (Eds), Progress in brain research: Vol. 64 S. (pp. 143-154).Amsterdam: Elsevier Science Publishers B.V. (Biomedical Division).

72

Burkhart Fischer and Rolf Boch

Goldberg, M. E., & Bushnell, M. C. (1981). Behavioral enhancement of visual responses in monkey cerebral cortex. 11. Modulation in frontal eye fields specifically related to saccades. Journal of Neurophysiology, 46, 773-787. Goldberg, M. E., Bushnell, M. C., & Bruce, C. J. (1986). The effect of attentive fixation on eye movements evoked by electrical stimulation of the frontal eye fields. Experimental Brain Research, 61, 579-584. Goldberg, M. E., & Wurtz, R. H. (1972). Activity of superior colliculus in behaving monkey: II. The effect of attention on neuronal responses. Journal of Neurophysiology, 35, 560-574. Guitton, D., Buchtel, H. A., & Douglas, R. M. (1985). Frontal lobe lesions in man cause difficulties in suppressing reflexive glances and in generating goal-directed saccades. Experimental Brain Research, 58, 455-472. Mayfrank, L., Mobashey, M., Kimmig, H., & Fischer, B. (1986). The role of fiation and visual attention on the occurrence of express saccades in man. European Journal of Psychiatiy and Neurological Science, 235, 269-275. Robinson, D. L., Goldberg, M. E., & Stanton, G. B. (1978). Parietal association cortex in the primate: sensory mechanisms and behavioral modulations. Journal of Neurophysiology, 41, 910-932. Shibutani, H., Sakata, H., & Hyvaerinen, J. (1984). Saccade and blinking evoked by microstimulation of the posterior parietal association cortex of the monkey. Experimental Brain Research, 55, 1-8. Steinman, R. M., Haddad, G. M., Skavenski, A. A., & Wyman, D. (1973). Miniature eye movement. Science, 181, 810-819. Wurtz, R. H., & Mohler, C. W. (1976a). Organization of monkey superior colliculus: enhanced visual response of superficial layer cells. Journal of Neurophysiology, 39, 745-765. Wurtz, R. H., & Mohler, C. W. (1976b). Enhancement of visual responses in monkey striate cortex and frontal eye fields. Journal of Neurophysiology, 39, 766-772.

VOLITIONAL ACTION, W.A. Hershberger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

73

CHAPTER 5

THE CEREBRAL CORRELATES OF REACHING Apostolos P. Georgopoulos Reaching to objects of interest in immediate extrapersonal space is an important motor activity of primates in everyday life. Recent studies of the activity of single cells in various brain regions of behaving primates have provided new insights into the brain mechanisms underlying reaching. These studies are discussed below with emphasis on parametric studies of the relations of the neuronal activity to the direction of reaching. Several brain areas are involved in the initiation and control of reaching. The study of the role of the various areas in this function was made possible by the advent of a technique that allowed the recording of the activity of single cells in the brain of behaving animals during reaching. This technique (Lemon, 1984) is indispensable for the study of neural mechanisms underlying motor aspects of behavior. Typically, monkeys are trained to perform various motor tasks and then microelectrodes are inserted through the dura into the brain area of interest to record extracellularly the electrically isolated action potentials of single cells. This combined behavioral-neurophysiological experiment provides a direct tool and one with fine-grain by which the brain mechanisms underlying performance can be studied.

Posterior Parietal Cortex and Reaching An important finding from such studies has been that several brain areas are involved in reaching, including areas of the cerebral cortex and various subcortical structures. The first cortical area investigated was the posterior parietal cortex (Hyvarinen & Poramen, 1974; Mountcastle, Lynch, Georgopoulos, Sakata, & Acuna, 1975). This cortical region was chosen because posterior parietal lesions in human subjects and monkeys result in motor defects in reaching (see Georgopoulos, 1986, in press, for reviews). Indeed, cells were identified in the superior and inferior parietal lobules (Brodmann’s areas 5 and 7, respectively) that changed activity with reaching in the absence of any peripheral somesthetic driving. These

74

Apostolos P. Georgopoulos

Figure 1. Drawing of a monkey reaching towards a lighted target in an apparatus used by Mountcastle et al. (1975) for studies of the posterior parietal cortex. The animal has just released the key and reached out to touch the lighted switch mounted on the moving carriage. The head fixation apparatus, implanted microdrive, cathode follower, and reward tube are also shown. (From Mountcastle et al., 1975; reproduced with permission). changes in cell activity were studied quantitatively using a behavioral apparatus that allowed 3-dimensional (3-D) reaching to stationary or moving visual targets. This apparatus is illustrated in Figure 1. A trial started when the monkey depressed a key at lap level. As soon as the key was depressed, a red light was lit on a push-button mounted on a semicircular rail in front of the monkey at shoulder level. After a variable period of time the light dimmed which signalled to the monkey to release the key and reach towards and push the lighted button. In some trials the button was stationary whereas in others it moved and then dimmed after a period of time while in motion. Therefore, the task comprised

75

Cerebral Correlates of Reaching

DETECT

MEAN RESPONSE

Figure 2. Impulse activity of an area 5 neuron during repeated reaching movements. The cell never responded to any passively delivered mechanical stimulus to the arm, to visual or auditory stimuli, or during passive-aversive movements of the contralateral arm. "Detect" indicates key release (see Figure l), and "response" indicates contact with the lighted switch. The arrow is mean response time & 1 SD. (From Mountcastle et al., 1975; reproduced with permission.) reaching to both stationary and moving targets. A typical result from reaching to a stationary target is shown in Figure 2. It can be seen that the activity of the cell changed appreciably during reaching. Although the changes in cell activity associated with reaching could be observed in the absence of visual guidance (Hyvarinen & Poramen, 1974; Mountcastle, Motter, & Andersen, 1980), cell activity was usually modulated more strongly when the animal reached with the eyes open. Indeed, a particular class of cells could qualify for an "eye-hand coordination" function because the changes in their activity was most intense when the monkey tracked a moving visual target with both the eyes and the hand (Mountcastle et al., 1975).

76

Apostolos P. Georgopoulos

Premotor Cortex and Reaching It is reasonable to assume that the planning and execution of reaching involves the concurrent or sequential activation of several brain regions. It is well documented that visually guided reaching may be disturbed following damage to the parietal lobe ("optic ataxia"; see Perenin & Vighetto, 1988, for a review) but it is not known how this disturbance is brought about. One idea is that visuospatial information is transmitted from the parietal to the frontal lobe, so that defects in visually guided reaching can be thought of as resulting from a disconnection between these two lobes (see Ferro, Bravo-Marques, Castro-Caldas, & Antunes, 1983, for a succinct discussion, and also Haaxma & Kuypers, 1975). The main recipients of posterior parietal projections are areas located anterior to the motor cortex, i.e., premotor cortical areas (reviewed in Humphrey, 1979), although area 5 also projects directly to the motor cortex (Strick & Kim, 1978; Caminiti, Zeger, Johnson, Urbano, & Georgopoulos, 1985). The premotor areas comprise a number of different sub-areas in both the lateral and medial surface of the hemisphere and project to the motor cortex (Muakkassa & Strick, 1979; Barbas & Pandya, 1987). In fact, these areas seem also to project directly to subcortical structures, including the red nucleus (Humphrey, Gold, & Reed, 1984) and the spinal cord (Hutchins & Strick, 1987), which means that their influence on the motor function need not be exerted exclusively through the motor cortex. Reaching movements pointing to visual or auditory targets were used to study the activity of cells in the premotor cortex of the monkey (Weinrich & Wise, 1982). Three basic classes of cells (n = 205) were distinguished based on the changes of their activity in the task. Movement-related cells (45%) changed activity in relation to the movement. These changes were temporally correlated better with the onset of the movement than with the onset of the stimulus, and were the same whether the stimulus was visual or auditory. Signal-related cell (43%) responded phasically to the presentation of the visual (87/89 cells) or the auditory (2/89 cells) stimulus. Finally, set-related cells (29%) showed changes in activity during an instructed delay period, that is between the onset of a "ready" cue and a "go" signal. These changes in activity were maintained throughout the delay period in most (52/59) of these cells. Many movement-related (57/149) and set-related (36/59) cells showed changes in activity that differed for the two directions of movement (side-to-side) used. Wise, Weinrich, and Mauritz (1986) have argued convincingly that changes in cell activity in the premotor and

Cerebral Correlates of Reaching

IS

lnit

I

I

A

77

TS Flew

I

I

I

TARGET ON

IS

”I

’

I j

r

...........

i

TARGET OFF

.#..............................

TARGET CHANGE

I .......// ....................... 1s

. - IS

TS

1 1.5-3.0s

delay

ME

i................. CHANGEr 0.5s

RT & MT

...........

Figure 3. Sequence and timing of events in the fixed-timing task used by Mauritz and Wise (1986) to study the neuronal mechanisms in premotor cortex associated with anticipation of predictable environmental events. See text for description of the task. The instruction stimulus (IS) either remained on during the instructed delay period (solid line),was turned off after 1 s (dashed line), or changed from one side to the opposite (dotted line). IS, instruction stimulus; Init, trial initiation; TS, trigger stimulus; A, target change or removal;’ Rew, reward; MB, monkey behavior, L, left target; R, right target. (From Mauritz & Wise (1986); reproduced with permission.)

motor cortex are related to the upcoming movement rather than to visuospatial cues themselves. Godschalk, Lemon, Kuypers, and van der Steen (1985) provided direct evidence for this idea by dissociating the direction of a reaching movement from the location or configuration of the visual stimulus that triggered the movement: under these conditions, the changes in the activity of premotor cells in the postarcuate area during a waiting period were related to the upcoming movement than to the visuospatial cue itself. However, Vaadia, Benson, Hienz, and Goldstein (1986) described cells in more anterior frontal regions which

78

Apostolos P. Georgopoulos

changed activity only when the monkey reached for an illuminated or a loud target but not when the same movement was made in the absence of such a target. The results of the studies above underscore the fact that several cortical areas are involved in preparation for and execution of reaching, as pointed out by Humphrey (1979) on grounds of anatomical connectivity. The different roles of various premotor areas in the initiation and control of reaching remain to be elucidated although there is clear evidence for their involvement in preparation for and anticipation of movement (see reviews by Wise, 1985, and by Tanji, 1984). The latter was observed clearly in premotor cell activity in studies by Mauritz and Wise (1986). A visua! pointing task was used. Events occurred in a fixed-timing sequence, as outlined in Figure 3, although the events themselves and some of the times involved were uncertain. A trial was initiated when the monkey pressed the central of three illuminable keys. (One of the two other keys was to the left and the other to the right of the central key.) After 1 s either the left or the right key (randomly selected) became illuminated. This key became the next target and illumination of the target key served as the instruction stimulus (IS). The IS was followed by an instructed delay period during which the monkey withheld its movement pending a subsequent cue. However, before that cue and 1 s after the onset of IS, one of the following three events happened with equal probability: (a) the target light remained on ("target on" condition, Figure 3), (b) the target key illumination was turned off ("target off'' condition), or (c) the target changed from one side to the opposite side ("target change" condition). After an additional 0.5, 1.25, or 2 s delay (randomly chosen) a light emitting diode over the target key was turned on. This served as the trigger stimulus (TS) for the monkey to move its arm and depress the target key to obtain a liquid reward. Figure 4 shows an example of anticipatory premotor cell activity in this task. These anticipatory changes in activity can be seen in both the left and right columns, corresponding to leftward and rightward trials. As described above, following the instruction stimulus there was a fixed 1 s delay after which different events could happen, and the trigger stimulus did not occur until after an additional 0.5, 1.25, or 2 s time period; thus, the three possible trigger-stimulus times were at 1.5, 2.25, or 3 s following the onset of the instruction stimulus. It can be seen in Figure 4 that clear changes in cell activity occurred at and around these three latter times, as shown in the rasters and the three peaks of the histograms following the instruction stimulus. Now, it is remarkable that even when

79

Cerebral Correlates of Reaching

RIGHT

LC?T

TARGET ON

I.

TARGET OFF

TARGETCHANGE

Figure 4. Impulse activity and peri-event time histograms of a premotor neuron showing anticipatory activity before the trigger stimulus (TS). Six raster and peri-event time histograms are shown, 3 for leftward trials and 3 for rightward trials in each for the conditions of the fixed-timing task outlined in Figure 3. Left and right columns have the same format. Each histogram binwidth is 40 ms. The heavy mark on each rater line indicates the time of occurrence, on that trial, of the trigger stimulus (TS). The first two activity peaks at the left of each histogram indicate activity from preceding trials. 0, target removal; A, target change. Scale is in impulses/s. (From Mauritz & Wise (1986); reproduced with permission.)

the trigger stimulus did not appear until a later time (e.g., at 3 s following the instruction stimulus) the changes in cell activity were still present at the other times in which the trigger stimulus was expected (i-e,, at 1.5 and 2.25 s following the instruction stimulus). These results show that the premotor cortex is intimately involved in the monitoring of anticipated

80

Apostolos P. Georgopoulos

external events that have been associated with the elicitation of arm movements.

Figure 5. Schematic drawing of the apparatus used to study free reaching movements in 3-D space. A, monkeys reached towards and pushed lighted buttons mounted at the end of metal rods threaded through a heavy metal plate. The movement trajectory was monitored using an ultrasonic system. B, Schematic diagram of the location of the 9 buttons used. Dotted lines indicate directions of movements. (From Schwartz et al., 1988; reproduced with permission. Copyright by Society for Neuroscience.)

Cerebral Correlates of Reaching

81

Motor Cortex and Reaching Porter and Lewis (1975) studied the changes in activity of motor cortical cells while monkeys reached out and manipulated a handle in front of them. It was found that single cells changed their activity during the task and that the latency of activation of different cells shifted to later times as more distal parts of the limb became involved in the motor act. This question of the sequential activation of motor cortical populations in reaching and grasping was investigated in more detail by Murphy, Wong, and Kwan (1985). The activity of single cells in the forelimb area of the motor cortex was recorded in a task in which monkeys pointed to targets in front of them. The functional relation between the recording locus and the joint of the arm was determined by intracortical microstimulation. Thus a cell could be classified as relating mainly to movements at the shoulder, elbow, hand or fingers. It was found that in the pointing task cells were generally activated sequentially, from proximal to distal, reflecting the sequential engagement of successively more distal parts of the arm. Murphy, Kwan, MacKay, and Wong (1982) also investigated the possible relations between motor cortical cell activity and joint motion and electromyographic (EMG) activity in muscles of the forelimb during reaching. There were three main findings of this study. First, no simple relation was observed between single cell activity and the EMG, even when the muscle from which the EMG was recorded was activated by intracortical microstimulation. Second, single cells related to motion about the shoulder or elbow joints behaved similarly in the task, although the motions produced about these joints could be quite different. And third, the discharge of shoulder-related cells seemed to vary systematically with the movement trajectory. These results indicate that the relations between single cell activity in the motor cortex and components (joint rotation, EMG activity) of reaching are complex.

Parametric Studies of the Direction of Reaching

Motor Cortex Direction of reaching and single cell activity. Reaching movements possess two spatial components, namely direction and amplitude. The relations between the direction of reaching and the activity of single cells in the motor cortex were studied recently

82

Apostolos P. Georgopoulos

(Georgopoulos, Schwartz, & Kettner, 1986; Schwartz, Kettner, & Georgopoulos, 1988; Georgopoulos, Kettner, & Schwartz, 1988; Kettner, Schwartz, & Georgopoulos, 1988). The behavioral apparatus used in these studies is illustrated in Figure 5. It consisted of 9 buttons of which 8 were at the corners and 1 at the center of an imaginary cube in front of the monkey. In a typical trial, the center button was lighted first; the animal was required to push it for a period of time after which it was turned off and one of the peripheral lights was turned on; the monkey then moved towards and pushed the lighted button to receive a liquid reward. Different peripheral lights were turned on in a randomized block design; thus, reaching movements were made which were of the same amplitude but which differed in direction. In fact, the experiment was designed to study the relations between the cell activity and the direction of reaching. Cells were selected for study that changed activity with spontaneous or evoked arm movements outside the behavioral task. Recordings of cell activity were made in the motor cortex contralateral to the performing arm. A salient finding of these studies was that the activity of single cells was broadly tuned to the direction of movement: cell activity was most intense for movements in a particular direction (the cell's "preferred direction") and decreased progressively for movements made farther away from the preferred direction. An example is shown in Figure 6A. The crucial variable on which cell activity depends is the angle formed between the direction of the movement and the cell's preferred direction (Figure 6B); in fact, the intensity of cell activity is a linear function of the cosine of this angle, as shown in Figure 7B. The directional tuning equation, then, is

where di(M) is the discharge rate of the ith cell with movement in direction M, bi and ki are regression coefficients, and O C ~ Mis the angle formed between the direction of movement M and the cell's preferred direction Ci. A directional tuning volume constructed using Equation 1 above is shown in Figure 7C for the cell whose data were illustrated in Figures 7A and 7B. The preferred directions differed for different cells and were distributed in the whole 3-D directional continuum. The data discussed above were obtained using 3-D reaching movements. Very similar results were obtained when a 2-D reaching task was used in which the monkey was trained to move a handle on a planar working surface and point with it towards lighted targets on that surface

83

Cerebral Correlates of Reaching

.

.

.. . .. . . , ........... .... ....... ... ;A:., ,........... . ... .. , L

- I

..A

t.-.

I

I.

I

L,.... I... I

..,..,.4

. _ I

...

...

.

. A _ . .

..,..-,-....... L

.,,!':.:.:

.. , .3:::::,.:::::::. _ _ - ...... . ...... a,..... ,...,. . ......-", ..__...a . . i :... ._(

I...* ........A I

B

-Y

-Z

LJ

Figure 6. Broad directional tuning in 3-D space of a motor cortical cell. A, impulse activity is shown in raster form for eight trials in eight movement directions indicated in the drawing at center. MOV, onset of movement. B, principle of directional tuning. C is the preferred direction of the cell whose rasters are shown in A, M is the direction of a movement; 0 is the angle between C and M. The cell activity varies in a linear fashion with the cos 0: it is maximum when cos 0 = 1, that is when C and M coincide; minimum when cos 0 = -1, that is when C and M are in opposite directions; and in between when 1 > cos 0 > -1, that is when M is between the direction of C and its opposite. (Adapted from Georgopoulos et al., 1986 and reproduced with permission. Copyright by AAAS, 1986)

84

Apostolos P. Georgopoulos

A

7 . 0 .

I . . .

Figure 7. A, impulse activity of a different directionally tuned cell. B, mean discharge rate (+ SD) during total time (from onset of target to end of movement) is plotted against the cosine of the angle 0 formed between the direction of the movement and the cell's preferred direction. C, directional tuning volume. The origin of the coordinate axes is at the origin of the movement and the arrow points in the cell's preferred direction. (From Schwartz et al., 1988; reproduced with permission. Copyright by Society for Neuroscience.) (Figure 8; Georgopoulos, Kalaska, & Massey, 1981; Georgopoulos, Kalaska, Caminiti, & Massey, 1982; see also Kalaska, Cohen, Hyde, & Prud'homme, in press). An example of a directionally tuned cell recorded in the motor cortex is shown in Figure 9. The directional tuning function plotted in Figure 9B is of the same form as that described by Equation 1 above. Finally, it is remarkable that the changes in cell activity relate to the direction of the reaching movement and not to its endpoint (Georgopoulos, Kalaska, & Caminiti, 1985). This is illustrated in Figure 10. Some possible implications of the directional tuning of motor cortical cells for circuits of the spinal cord related to reaching movements have been discussed elsewhere (Georgopoulos, 1988).

Cerebral Correlates of Reaching

85

0

Figure 8. Left: schematic drawing of the 2-D apparatus used to study "drawing" movements of monkeys. The monkey sat at A, in front of the working surface, B. The numbered light emitting diodes (LED) were placed on a circle of 8 cm radius. The monkey held the articulated manipulandum at its distal end (C) and captured a lighted LED within a 10 mm diameter transparent plexiglass circle (D). X-Y motion of the center of that circle was monitored every 10 ms with a resolution of 0.125 mm. Right: a monkey performing the task (side view). Insert shows two trajectories. (From Georgopoulos et al., 1981; reproduced with permission.)

Direction of Reaching and Neuronal Populations The data discussed above focus on the question of the neural representation and coding of the direction of reaching. They indicate that a given cell participates in movements of various directions and that, conversely, a movement in a particular direction will involve the activation of a whole population of cells: how, then, is the direction of reaching represented in a unique fashion in a population of neurons each of which is directionally broadly tuned? To answer this question it was hypothesized that the motor cortical command for the direction of reaching can be regarded as an ensemble of vectors (see Figure 11; Georgopoulos, Caminiti, Kalaska, & Massey, 1983; Georgopoulos e t al., 1986). Each vector represents the contribution of a directionally tuned cell. A

86

Apostolos P. Georgopoulos

I

'A'

P C R l 1 0 .so1

bQ

S1R

1

I

IS

135

225

31s'

DIRECTDN OF MOVEMENT

Figure 9. Broad directional tuning in 2-D space of a cell recorded in the arm area of the motor cortex. Top, impulse activity during five trials with movements in the directions indicated in the drawing at the center. Short vertical bars indicate the occurrence of an action potential. Rasters are aligned to the onset of movement (M). Longer vertical bars preceding the onset of movement indicate the onset of the target (T); those following the movement indicate the entrance to the target window (see Figure 9) and the delivery of reward. Bottom, average frequency of discharge (-t SEM) from the onset of the stimulus until the entry to the target window are plotted against the direction of movement. Continuous curve is a cosine function fitted to the data using multiple regression analysis (see Georgopoulos et al., 1982 for details). (From Georgopoulos et al., 1982; reproduced with permission. Copyright by Society for Neuroscience.)

Cerebral Correlates of Reaching

87

PBt141.501

B

Pet 1 4 4 . 5 0 2

Figure 10. Impulse activity of a motor cortical cell studied in the 2-D apparatus during center -,out (A) and out + center (B) movements, as indicated in the center diagrams. Conventions as in Figure 9. (From Georgopoulos et al., 1984; reproduced with permission.)

particular vector points in the cell's preferred direction and has length proportional to the change in cell activity associated with a particular movement direction: then the vector of these weighted cell vectors (the "neuronal population vector") points at or near the direction of the movement (Georgopoulos et al., 1983; Georgopoulos et al., 1986). This

Apostolos P. Geogopoulos

88

Ly.

Popula t i on v e c t o r

A)

90

D i r e c t i o n of m o v e m e n t

B)

M o v e m e n t trajectories

Vector hypothesis

D)

99% c o n f i d e n c e i n t e r v a l for p o p u l a t i o n v e c t o r

Figure 11. Neuronal population coding of movement direction illustrated for a motor cortical population (N = 241 cells) and one movement direction. A, movement direction; B, family of trajectories made by a well trained monkey; C, vectorial contributions of single cells (continuous lines) add to yield the population vector (interrupted line) which is in the direction of the movement; D, 99% confidence interval for the population vector. (From Georgopoulos et al., 1984; reproduced with permission.)

is illustrated in Figure 12 for eight 2-D reaching directions. Some findings regarding the neuronal population vector are summarized below. The neuronal population vector predicts the direction of reaching during the reaction time. In the paradigms used in the studies described above, a reaction time of approximately 300 ms intervened from the onset of the stimulus to the beginning of the movement. Given that the changes in cell activity in the motor cortex precede the onset of movement by approximately 160-180 ms, on the average (Georgopoulos et al., 1982), it

Cerebral Correlates of Reaching

89

Figure 12. Neuronal population vector (heavy dashed lines with arrow) calculated for 8 movement directions. All clusters represent the same neuronal population composed of individual cell vectors (thin lines, N = 241 cells). The dotted lines in the center indicate the direction of movement. (From Georgopoulos et al., 1983; reproduced with permission.) is of interest to know whether the population vector predicts the direction of the upcoming movement during the reaction time. Indeed, this was found to be the case both for 2-D and 3-D reaching movements (Georgopoulos, Kalaska, Crutcher, Caminiti, & Massey, 1984; Georgopoulos et al., 1988). An example from two movement directions is illustrated in Figure 13. The neuronul population vector predicts the direction of reaching during an instructed delay period. In the experiments yielding this finding, monkeys were trained to withhold a visually cued movement for a period of time after the onset of the visual (i.e., directional) cue and to move later in response to a "go" signal. During this instructed delay period the population vector in the motor cortex computed every 20 ms gave a reliable signal concerning the direction of the movement that was to be

90

Apostolos P. Georgopoulos

90

:I

DIRECTION

OF MOVEMENT

.

*

A5O

/.

P O P U LATlO N VECTOR

Figure 13. Population vectors computed every 20 ms for movements in two different directions. Instantaneous velocity vectors are also shown. Notice that the population vector lengthens well before the movement begins and points in the approximate direction of the upcoming movement. (From Georgopoulos et al., 1984a; reproduced with permission.)

triggered later for execution (Georgopoulos, Crutcher, & Schwartz, 1989). This finding suggests that the motor cortex is involved in processing information about the direction of the upcoming movement in space even in the absence of an immediate movement. The neuronal population vector predicts the direction of reaching for movements of different origin. In the experiments yielding this finding, monkeys made movements that started from different points, were in the same direction but described parallel trajectories in 3-D space. Under these conditions, the population vector in the motor cortex predicted well the direction of the reaching movement (Kettner et al., 1988). The neuronul population coding of the direction of reaching is resistant to loss of cells. The population coding described above is a distributed code and as such does not depend exclusively on any particular cell. This robustness was evaluated by calculating the population vector from progressively smaller samples of cells randomly selected from the original

91

Cerebral Correlates of Reaching

.I] N E

;

20

N 10

0

, 0

I

I

I

I

I

I

I

I

I

I

50

100

150

200

250

300

350

400

450

500

POPULATION SIZE (Number o f c e l l s )

Figure 14. The variability of the estimated 3-D direction of the population vector is plotted as a function of the number of cells in the population. The units on the ordinate are half-angles (in degrees) at the apex of a directional confidence cone computed for the population vector using statistical bootstrapping techniques. (From Georgopoulos et al., 1988; reproduced with permission.) population (Georgopoulos et al., 1988). It can be seen in Figure 14 that the direction of the population vector can be reliably estimated from as few as 100-150 cells. The neuronal population vector transmits directional information comparable to that transmitted by the direction of movement. The information transmitted by the direction of the population vector was calculated using an information-theoretical analysis and compared to the information transmitted by the direction of 2-D reaching movements (Georgopoulos & Massey, 1988). It was found that both the neuronal population vector and the reaching movement transmitted comparable amounts of direction-

92

Apostolos P. Georgopoulos

0

1

2 3 4 5 6 I N P U T INFORMATION ( b i t s )

7

Figure 15. Comparison of information transmitted by the direction of the neuronal population vector in the motor cortex ("Neural") and the direction of 2-D reaching movements (best human performance, "Human"). (From Georgopoulos & Massey, 1988; reproduced with permission.) a1 information at various levels of input information but the information transmitted by the population vector was consistently higher than that transmitted by the movement by a constant amount of approximately 0.5 bits (see Figure 15). These results, of the metric-free informationtheoretical analysis, reinforce the usefulness of the population vector as a meaningful measure of the directional tendency of neuronal ensembles and suggest that there is a loss of information following the processing by the motor cortex. The neuronal population vector can provide insight into the brain mechanisms underlying mental transformations. The fact that the population vector calculated post hoc during the reaction time or during an instructed delay period (see above) points in the direction of the upcoming movement has important implications and potentially significant applications for tasks that require spatial transformations because it provides an accurate and robust monitor of the directional tendency of

Cerebral Correlates of Reaching

93

a neuronal ensemble as this tendency evolves and changes in time. We utilized this feature to gain an insight into the brain correlates of a mental transformation. The task required the making of 2-D reaching movements in a direction that was at an angle from a stimulus direction. Under these conditions the reaction time of human subjects increases in a linear fashion with the angle (Georgopoulos & Massey, 1987) suggesting that a mental rotation of the stimulus direction to the movement direction (like the hand of a clock) might underlie performance in this task. This hypothesis could be tested because the directions above could be visualized as the neuronal population vector. Indeed, a monkey was trained to perform in a conditional 2-D reaching task which required that the monkey move towards a light when the light came on dim, or move in a direction perpendicular and counterclockwise from the light when it came on bright. The position of the light on a circle and the d i m r i g h t condition were combined (8 positions, at 45" intervals, X 2 conditions = 16 combinations) and presented in a pseudo-random sequence. Recordings in the motor cortex revealed that, during the reaction time, the neuronal population vector pointed in the direction of the movement when the monkey was moving towards the light (Figure 16, left panel). However, when the monkey moved in a direction perpendicular and counterclockwise from the light, the neuronal population vector pointed first in the direction of the stimulus and then rotated counterclockwise for approximately 90", and stabilized pointing in the direction of the movement. These findings provide evidence for the mental rotation hypothesis above and underscore the usefulness of the population vector as a meaningful tool for analysis and interpretation of brain events related to cognitive motor transformations.

Parietal Cortex (Area 5 ) Two-dimensional reaching tasks have been used to study the activity of cells recorded in area 5 (Kalaska, Caminiti, & Georgopoulos, 1983; Kalaska, 1988). The results obtained in these studies were very similar to those obtained in the motor cortex. An example of a directionally tuned cell in area 5 is shown in Figure 17. Moreover, the neuronal population vector in area 5 predicted well the direction of reaching (Kalaska et al., 1983; Georgopoulos, 1987). The similar directional properties of area 5 cells and the population coding of the direction of reaching by these cells are important because these two areas may relate to different aspects of the movement, namely the motor cortex to its initiation and area 5 to its

94

Apostolos P. Georgopoulos

Figure 16. Left panel: A, direction of required movement (M) was in the direction of the dim stimulus ( S , open circle). Direction is indicated in polar angles. B, neuronal population vector calculated every 10 ms from stimulus onset ( S ) points in the direction of upcoming movement (180") which started at M. C, population vectors during the reaction time from the moment it is lengthening until the onset of movement. D, direction of population vectors identified in C. above, is plotted against time elapsed from stimulus onset. (Direction is in polar angles). Right requiredA,movement panel: direction(M) of

A

Direct

Rotrtlon

s SM O l- .

-

0

&-'M

t

90'

i I

C

v D fOO]

,

,

?",. :j ,;:I, ,

was in a direction per100 pendicular and counter? 100 200 300 0 100 200 300 clockwise from the direc; mm(nu) s nm(ms) tion of the bright stimulus ( S , filled circle). B, neuronal population vector points initially in the direction of the stimulus (90") and then rotates gradually counterclockwise and stabilizes, pointing in the direction of the upcoming movement (180"). C, population vectors during the reaction time from the moment it is lengthening until the onset of movement. Notice the counterclockwise rotation. D, direction of population vectors identified in C, above, is plotted against time elapsed from stimulus onset. (From Georgopoulos, Lurito, Petrides, Schwartz, & Massey, 1989; reproduced with permission. Copyright by AAAS, 1989)

Cerebral Correlates of Reaching

J , , l , l , , l l l ,

,

1

A

I

1

,

,

,

1

1

1

,

1

95

' I

I

A

I

1

I II

l I I l l / , I l l , 1 1 1 1 1 1 1 1 1 IIIII

~~~~

l

I

II

I

I

I

I

I

PAD120.006 ORIENTATION: R.T

Figure 17. Impulse activity of a cell recorded in parietal area 5 during 2-D reaching movements. Conventions as in Figure 9. (From Kalaska, 1983; reproduced with permission.)

monitoring (Kalaska et al., 1983; Kalaska, 1988). Therefore, the similar relations observed between cell activity in these structures and the direction of movement may reflect a common "language" of communication in the spatial domain, given that these areas are anatomically interconnected (Strick & Kim, 1978; Caminiti et al., 1985). The main difference found between the two areas related to onset times of the changes in cell activity, with motor cortical cells being engaged approximately 60 ms before those of area 5 (Figure 18). This finding is in accordance with the different roles postulated above for these two areas.

96

Apostolos P. Georgopoulos

U

0 Z W 0 LT

w U

0

I

-4m

I

I

-am

-341

t

1

I

-roo

I

o

r

I

m

p

I

o

x

I

a

t M

T I M I N G OF INCREASE OF DISCHARGE CMSECI AT THE PREFERRED D I R E C T I O N

Figure 18. Cumulative frequency distributions of latency of change in cell activity in motor cortex and area 5. (Adapted from Kalaska et al., 1983; reproduced with permission.)

Cerebellum A broad directional tuning has also been observed in the cerebellum (Fortier, Kalaska, & Smith, in press). Figure 19 shows an example of a directionally tuned Purkinje cell in the cerebellar cortex, and Figures 20 and 21 show examples of directionally tuned cells in the cerebellar nuclei, interpositus (Figure 20) and dentate (Figure 21). Similarly to the motor cortex, the neuronal population vector in the cerebellar structures mentioned above predicted well the direction of reaching (Fortier et al., in press). This is significant because a major part of the cerebellar output is directed to the motor cortex via the thalamus; therefore, the similarity in the directional properties of the motor cortex and cerebellum may

Cerebral Correlates of Reaching

97

reflect the cooperation of these structures in the specification of the direction of reaching, and the use of a common "language" for communication.

Spinal Cord and Reaching Reaching involves motion at the shoulder and elbow joints. Behaviorally, there is little doubt that these two joints are controlled as one functional unit (Soechting & Lacqaniti, 1981), and that this control is separate from that of the wrist (Lacquaniti & Soechting, 1982; Soechting, 1984). Descending motor commands from the motor cortex and other brain areas influence the proximal arm motoneurons, i.e., those innervating muscles acting on the elbow and/or shoulder, through a set of interneurons located at the C3-C4 spinal segments, that is above the segments of the proximal motor nuclei. These interneurons ("C3-C4 propriospinal neurons", Lundberg, 1979) have been studied extensively in the cat. They receive monosynaptic inputs from several supraspinal sources (Illert, Lundberg, Padel, & Tanaka, 1978) including the pyramidal (i.e., corticospinal), rubrospinal, reticulospinal and tectospinal tracts; they distribute their axons to several proximal motoneuronal pools (Alstermark, Kummel, Pinter, & Tantisira, 1987); and, they also send an ascending collateral to the lateral reticular nucleus (Alstermark, Lindstrom, Lundberg, & Sybirska, 1981). Selective section of the output from these propriospinal neurons to their target motoneurons results in abnormal reaching with normal grasping, and similar effects are observed when the corticospinal input to the propriospinal neurons is removed (Alstermark, Lundberg, Norrsell, & Sybirska, 1981). Moreover, propriospinal neurons seem to be selectively engaged during reaching movements (Alstermark & Kummel, 1986). These results indicate that the C3-C4 propriospinal system is concerned with the neural integration of the reaching movement at the spinal level and that the motor cortex and other areas control reaching most probably through that system. This motor cortical control is also exerted at other levels within the propriospinal system; for example, there is direct corticospinal input on a key inhibitory interneuron which mediates inhibition from afferent fibers to propriospinal neurons (Alstermark, Lundberg, & Sasaki, 1984). This peripherally initiated inhibition of the propriospinal neurons is important in limiting the reaching movement, for lack of peripheral input results in consistent hypermetria in reaching (Alstermark, Gorska, Johannisson, & Lundberg, 1986).

98

Apostolos P. Georgopoulos

I

,

.

.

.

,

I

-

.

-1OM)-500

, M

.

,

,

500 1000 1.500

TIME(ms)

Figure 19. Impulse activity (upper) and peri-event time histograms (lower) of a Purkinje cell recorded in cerebellar cortex during 2-D reaching movements. Rasters and histograms are aligned to movement onset (M). (Adapted from Fortier et al., in press; reproduced with permission.)

:w.o:-lOM)-wO

M

5M)

loo0 1500

TIME (ma)

0

10

SPIKESlSEC

20

Cerebral Correlates of Reaching

99

INTERPOSITUS

PECO34.001

"'

-1000 -500

M 500 lo00 1500 TIME (ma)

I

1.

I

I

r.o nrn-

2 5

o. PEAO61.002

rn

-1000 -500

M

UT 500 1000 15W

TIME (ms)

Figure 20. Peri-event time histograms of the activity of cells recorded in the interpositus nucleus during 2-D reaching movements. Conventions as in Figure 19. (Adapted from Fortier et al., in press; reproduced with permission.)

100

Apostolos P. Georgopoulos

DENTATE

D

w

n

a a

0

Y -a

m-

$ 5 Y

PBA113.001

v,

-1000 -500

M

500 1000 1500

TIME (ms)

Y

PBEl39.001

-1000 -500

M

500

1000 1500

TIME (ms)

Figure 21. Peri-event time histograms of the activity of cells recorded in the dentate nucleus during 2-D reaching movements. Conventions as in Figure 19. (Adapted from Fortier et al., in press; reproduced with permission.)

Cerebral Correlates of Reaching

101

The results summarized above indicate that a large part of neural integration of the reaching movement is accomplished in the spinal cord. In a way, this is qualitatively similar t o the sophisticated integration observed in spinal circuits underlying locomotion (Grillner, 1981): both cases involve the production of complicated motor outputs, complicated in the sense of involving the time-varying control of several muscles ahd of more than one joint. It is possible, and even probable, that the detailed organization and the neural integration of the reaching movement need not be the concern, or the burden, of the motor cortex or other motor areas. These various areas could be concerned, instead, with the initiation and ongoing control of reaching according to internally generated goals, as, for example, in drawing, or according to information from exteroceptors, as, for example, in reaching towards a visual or auditory target. These functions would then be accomplished by the activation of neuronal populations in different brain areas, including the motor cortex, which, in turn, would engage the spinal "reaching" circuits.

Acknowledgement: This work was supported by USPHS Grant NS17413.

References Alstermark, B., Gorska, T., Johannisson, T., & Lundberg, A. (1986). Hypermetria in forelimb target-reaching after interruption of the inhibitory pathway from forelimb afferents to C3-C4 propriospinal neurones. Neuroscience Research, 3, 457-461. Alstermark, B., & Kummel, H. (1986). Transneuronal labelling of neurones projecting to forelimb motoneurones in cats performing different movements. Brain Research, 376, 387-391. Alstermark, B., Kummel, H., Pinter, M. J., & Tantisira, B. (1987). Branching and termination of C3-C4 propriospinal neurons in the cervical spinal cord of the cat. Neuroscience Letters, 74, 291-296. Alstermark, B., Lindstrom, S., Lundberg, A., & Sybirska, E. (1981). Integration in descending motor pathways controlling the forelimb in the cat. 8. Ascending projection to the lateral reticular nucleus from C3-C4 propriospinal neurones also projecting to forelimb motoneurones. Experimental Brain Research, 42, 282-298. Alstermark, B., Lundberg, A., Norrsell, U., & Sybirska, E. (1981). Integration in descending motor pathways controlling the forelimb in the cat. 9. Differential behavioral defects after spinal cord lesions interrupting defined pathways from higher centres to motoneurones. Experimental Brain Research, 42, 299-318.

102

Apostolos

P. Georgopoulos

Alstermark, B., Lundberg, A., & Sasaki, S. (1984). Integration in descending motor pathways controlling the forelimb in the cat. 11. InhibitoIy pathways from higher motor centres and forelimb afferents to C3-C4 propriospinal neurones. Expenmental Brain Research, 56, 293-307. Barbas, H., & Pandya, D. N. (1987). Architecture and frontal cortical connections of the premotor cortex (area 6) in the rhesus monkey. Journal of Comparative Neurology, 256, 211-228. Caminiti, R., Zeger, S., Johnson, P. B., Urbano, A., & Georgopoulos, A. P. (1985). Cortico-cortical efferent systems in the monkey: a quantitative spatial analysis of the tangential distribution of cells of origin. Journal of Comparative Neurology, 241, 405-419. Ferro, J. M., Bravo-Marques, J. M., Castro-Caldas, A., & Antunes, L. (1983). Crossed optic ataxia: possible role of the dorsal splenium. Journal of Neurology, Neurosurgety, and Psychiatry, 46, 533-539. Fortier, P. A., Kalaska, J. F., & Smith, A. M. (in press). Cerebellar neuronal activity related to whole-arm reaching movements in the monkey. Journal of Neurophysiology. Georgopoulos, A. P. (1986). On reaching.-Annual Review of Neuroscience, 9, 147-170. Georgopoulos, A. P. (1987). Cortical mechanisms subserving reaching. In: Motor areas of the cerebral cortex, CIBA Foundation Symposium No. 132 (pp. 125-132). New York John Wiley. Georgopoulos, A. P. (1988). Neural integration of movement: role of motor cortex in reaching. The FASEB Journal, 2, 2849-2857. Georgopoulos, A. P. (in press). Visual control of reaching. In G. M. Edelman, W. E. Gall, & W. M. Cowan (Eds.), Signal and sense: local and global order in perceptual maps. New York: John Wiley. Georgopoulos, A. P., Caminiti, R., Kalaska, J. F., & Massey, J. T. (1983). Spatial coding of movement: a hypothesis concerning the coding of movement direction by motor cortical populations. Experimental Brain Research Supplement, 7, 327-336. Georgopoulos, A. P., Crutcher, M. D., & Schwartz, A. B. (1989). Cognitive spatial-motor processes. 3. Motor cortical prediction of movement direction during an instructed delay period. Experimental Brain Research, 75, 183-194. Georgopoulos, A. P., Kalaska, J. F., & Caminiti, R. (1985). Relations between two-dimensional arm movements and single cell discharge in motor cortex and area 5: movement direction versus movemeRt endpoint. Expenmental Brain Research Supplement, 10, 176-183.

Cerebral Correlates of Reaching

103

Georgopoulos, A. P., Kalaska, J. F., Caminiti, R., & Massey, J. T. (1982). On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. Journal of Neuroscience, 2, 1527-1537. Georgopoulos, A. P., Kalaska, J. F., Crutcher, M. D., Caminiti, R., & Massey, J. T. (1984). The representation of movement direction in the motor cortex: Single cell and population studies. In G. M. Edelman, W. E. Gall, & W. M. Cowan (Eds.), Dynamic aspects of neocortical function (pp. 501-524). New York: John Wiley. Georgopoulos, A. P., Kalaska, J. F., & Massey, J. T. (1981). Spatial trajectories and reaction times of aimed movements: effects of practice, uncertainty, and change in target location. Journal of Neurophysiology, 46, 725-743. Georgopoulos, A. P., Kettner, R. E., & Schwartz, A. B. (1988). Primate motor cortex and free arm movements to visual targets in three-dimensional space. 11. Coding of the direction of movement by a neuronal population. Journal of Neuroscience, 8, 2928-2947. Georgopoulos, A. P., Lurito, J. T., Petrides, M., Schwartz, A. B., & Massey, J. T. (1989). Mental rotation of the neuronal population vector. Science, 243, 234-236. Georgopoulos, A. P., & Massey, J. T. (1987). Cognitive spatial-motor processes. 1. The making of movements at various angles from a stimulus direction. Experimental Brain Research, 65, 361-370. Georgopoulos, A. P., & Massey, J. T. (1988). Cognitive spatial-motor processes. 2. Information transmitted by the direction of two-dimensional arm movements and by neuronal populations in primate motor cortex and area 5. Experimental Brain Research, 69, 315-326. Georgopoulos, A. P., Schwartz, A. B., & Kettner, R. E. (1986). Neuronal population coding of movement direction. Science, 233, 1416-1419. Godschalk, M., Lemon, R. N., Kuypers, H. G. J. M., & van der Steen, J. (1985). The involvement of monkey premotor cortex neurones in preparation of visually cued arm movements. Behavioural Brain Research, 18, 143-157. Grillner, S. (1981). Control of locomotion in bipeds, tetrapods, and fish. In J. M. Brookhart, & V. B. Mountcastle (Eds.), Handbook of Physiology. The Nervous System IZ (pp. 1179-1236). Bethesda, MD: American Physiological Society. Haaxma, R., & Kuypers, H. G. J. M. (1975). Intrahemispheric cortical connexions and visual guidance of hand and finger movements in the rhesus monkey. Bruin, 98, 239-260.

104

Apostolos P. Georgopoulos

Humphrey, D. R. (1979). On the cortical control of visually directed reaching: contributions by nonprecentral motor areas. In R. E. Talbott & D. R. Humphrey (Eds.), Posture and movement (pp. 51-112). New York Raven Press. Humphrey, D. R., Gold, R., & Reed, D. J. (1984). Size, laminar and topographic origin of cortical projections to the major divisions of the red nucleus in the monkey. Journal of Comparative Neurology, 225, 75-94. Hutchins, K. D., & Strick, P. L. (1987). Origin of corticospinal projections to the ipsilateral spinal cord. Society for Neuroscience Abstracts, 13, 243. Hyvarinen, J., & Poramen, A. (1974). Function of the parietal associative area 7 as revealed from cellular discharges in alert monkeys. Bruin, 97, 673-692. Illert, M., Lundberg, A., Padel, Y., & Tanaka, R. (1978). Integration in descending motor pathways controlling the forelimb in the cat. 5. Properties of and monosynaptic excitatory convergence on C3-C4 propriospinal neurones. Experimental Brain Research, 33, 101-130. Kalaska, J. F. (1988). The representation of arm movements in postcentral and parietal cortex. Canadian Journal of Physiology and Phumcolo&v, 66, 455-463. Kdaska, J. F., Caminiti, R., & Georgopoulos,A. P. (1983). Cortical mechanisms related to the direction of two-dimensional arm movements: relations in parietal area 5 and comparison with motor cortex. Experimental Brain Research, 51, 247-260. Kalaska, J. F., Cohen, D. A. D., Hyde, M. L., & Prud’homme, M. (in press). A comparison of movement direction-related vs. load direction-related activity in primate motor cortex, using a two-dimensional reaching task. Journal of Neuroscience. Kettner, R. E., Schwartz, A. B., & Georgopoulos, A. P. (1988). Primate motor cortex and free arm movements to visual targets in three-dimensional space. 111. Positional gradients and population coding of movement direction from various movement origins. Journal of Neuroscience, 8, 2938-2947. Lacquaniti, F., & Soechting, J. F. (1982). Coordination of arm and wrist motion during a reaching task. Journal of Neuroscience, 2, 399-408. Lemon, R. N. (1984). Methods for neuronal recording in conscious animals. Chisester: John Wiley & Sons. Lundberg, A. (1979). Integration in a propriospinal motor centre controlling the forelimb in the cat. In H. Asanuma & V. J. Wilson (Eds.), Integration in the Nervous System (pp. 47-69). Tokyo: Igaku-Shoin.

Cerebral Correlates of Reaching

105

Mauritz, K. -H., & Wise, S. P. (1986). Premotor cortex of the rhesus monkey: neuronal activity in anticipation of predictable environmental events. @enmental Brain Research, 61, 229-244. Mountcastle, V. B., Lynch, J. C., Georgopoulos, A. P., Sakata, H., & Acuna, C.(1975). Posterior parietal association cortex of the monkey: command functions for operations within extrapersonal space. Journal of Neurophysiology, 38, 871-908. Mountcastle, V. B., Motter, B. C., & Andersen, R. A. (1980). Some further observations on the functional properties of neurons in the parietal lobe of the waking monkey. Behavioral and Brain Sciences, 3, 520-522. Muakkassa, K. F., & Strick, P. L. (1979). Frontal lobe inputs to primate motor cortex: evidence for four somatotopically organized "premotor" areas. Brain Research, 177, 176-182. Murphy, J. T., Kwan, H. C., MacKay, W. A., & Wong, Y. C. (1982). Precentral unit activity correlated with angular components of a compound arm movement. Brain Research, 246, 141-145. Murphy, J. T., Wong, Y . C., & Kwan, A. C. (1985). Sequential activation of neurons in primate motor cortex during unrestrained forelimb movement. Journal of Neurophysiology, 53, 435-445. Perenin, M. -T., & Vighetto, A. (1988). Optic ataxia: a specific disruption in visuomotor mechanisms. Brain, 111, 643-674. Porter, R., & Lewis, M.Mc. (1975). Relationship of neuronal discharges in the precentral gyrus of monkeys to the performance of arm movements. Brain Research, 98, 21-36. Schwartz, A. B., Kettner, R. E., & Georgopoulos, A. P. (1988). Primate motor cortex and free arm movements to visual targets in three-dimensional space. I. Relations between single cell discharge and direction of movement. Journal of Neurocsience, 8, 2913-2927. Soechting, J. F., & Lacquaniti, F. (1981). Invariant characteristics of a pointing movement in man. Journal of Neuroscience, 1, 710-720. Soechting, J. F. (1984). Effect of target size on spatial and temporal characteristics of a pointing movement in man. Experimental Brain Research, 54, 121-132. Strick, P. L., & Kim, C. C. (1978). Input to primate motor cortex from posterior parietal cortex (area 5). I. Demonstration by retrograde transport. Brain Research, 157, 325-330. Tanji, J. (1984). The neuronal activity in the supplementary motor area of primates. Trends in Neuroscience, 7, 282-285. Vaadia, E., Benson, D. A., Hienz, R. D., & Goldstein, M. H. Jr. (1986). Unit study of monkey frontal cortex: active localization of auditory and of visual stimuli. Journal of Neurophysiology, 56, 934-952.

106

Apostolos P. Georgopoulos

Weinrich, M., & Wise, S. P. (1982). The premotor cortex of the monkey. Journal of Neuroscience, 2, 1329-1345 Wise, S. P. (1985). The primate premotor cortex: past, present, and preparatory. Annual Review of Neuroscience, 8, 1-19. Wise, S. P., Weinrich, M., & Mauritz, K. -H. (1986). Movement-related activity in the premotor cortex of rhesus macaques. Progress in Brain Research, 64, 117-131.

VOLITIONAL ACTION, W.A . Hershberger (Editor) B. V. (North-Holland), 1989

0 Elsevier Science Publishers

107

CHAPTER 6 WILL, VOLITIONAL ACTION, ATTENTION AND CEREBRAL POTENTIALS IN MAN: BEREZTSCHAFTSPOTENTL4L, PERFORMANCE-RELATED POTENTIALS, DIRECTED ATTENTION POTENTIAL, EEG SPECTRUM CHANGES H. H. Kornhuber, L. Deecke, W. Lang, M. Lang and A. Kornhuber This chapter is dedicated to Vernon B. Mountcastle

Contents 1. Introduction 2. Methods 2.1. Performance-related DC shifts 2.2. Event-related EEG spectrum 2.3. Regional cerebral blood flow by Tc-99m-HMPAO brain SPECT 2.4. Magnetoencephalography 2.5. Methodological considerations 3. Results and Comments 3.1. The initiation of volitional actions: When to do 3.1.1. Human pathophysiology 3.1.2. Human voluntary movement physiology: The Bereitschaftspotentid 3.1.3. Centralization of the starting function in volitional actions 3.1.4. Selecting the "right" moment to start a movement 3.1.5. After the starting signal has been released 3.1.6. Basal ganglia and cerebellum in the initiation of volitional actions 3.1.7. SMA and the temporal organization of motor sequences 3.2. Physiological signs of anticipatory, task-specific planning 3.3. Resoluteness: ability to maintain goals and adjust behavior 3.4. Cognitive information processing 3.5. Directing attention toward forthcoming, relevant, sensory events 3.5.1 Directed attention potential (DAP) 3.5.2 Anticipation of the right moment to act

108 3.6. 4. 4.1. 4.2. 4.3. 4.4.

H. H. Komhuber et al. Volition in the sense of setting priorities General discussion The utility of a centralized starting function for movements Frontal lobe and volition Attention, the parietal lobe, and the directed attention potential Will and freedom

1. Introduction The concept of will did not originate within the science of neurophysiology. Neither did it come from comparative research on the behavior of animals and men, although it could have, because man and beast differ in the automaticity of their actions. Man acts more deliberately. Man usually considers various needs, duties, and stimuli, makes reasoned decisions, and even engages in creative planning, before acting, whereas in most animal species there is a more direct transfer from drives and stimuli into behavior. But the concept of will did not originate within the science of behavior. Indeed, it came from the forerunner of science, from ancient philosophy. It is an old observation that man has mind, different cultures, and a behavior more influenced by culture than by instinct. In the history of philosophical thought, it gradually became clear that without will, reasoning can not be transduced into behavior. It is only by means of a will with goals beyond our ego, that reason, conscience and good plans are able to determine priorities among needs and actions. Man does not become humane without possessing reason and good will -- at least this is the belief of most people who have thought profoundly about such matters since the time of Plato, including Aristotle, Zen0 from Kition, Thomas Aquinas, Duns Scotus, Descartes, Erasmus, Leibniz, Kant, and in our century Wundt, Pfander, Jaspers, H. Reiner, Nicolai Hartmann, Kurt Schneider, and Ricoeur. A regressive step towards a belief in the impotence of will came from determinism which entered physics from the apocalyptic branch of theology. Then, in the present century, because of the moral degeneration of the under-challenged upper class in the Europe of 1900, the teaching of Schopenhauer (who, perhaps for personal reasons, did not believe in free will) began to have an impact. Schopenhauer tried to eliminate the concept of will by overextending it in such a way that will became identical with drive. Freud then elaborated and explained what drive is supposed to be all about: He claimed that man is driven by a

Cerebral Correlates of Volition and Attention

109

search for pleasure, with will being a narcissistic illusion. Subsequently, the theoretical significance of the will steadily declined, and, by the midsixties of our century, the term will no longer appeared in the Psychologicul Abstracts. Neither will nor volition is to be found in the 1983 edition of the Encyclopaedia Britunnica -- except, of course, for will in the sense of the last will (Le., testament). The impact of hedonism was not confined to academic abstractions. In 1968, there was a "cultural revolution" in Europe, with hedonism ("Freudo-Marxism") as the leading ideology. This new hedonism was marked by a sharp increase in the consumption of alcohol, cigarettes and drugs by the young European generation. As a consequence, alcoholic embryopathy, having been so rare that it was unknown to the medical world, became so common within a few years that it is now twice as common as the usually commonest inborn CNS disturbance, Down's syndrome (H. H. Kornhuber, 1984~). Thus, when our investigation of volition started (H. H. Kornhuber & Deecke, 1964, 1965) it was in opposition both to the morality and the academic mentality of that time. Our research was in part a consequence of the experience that will is important for human behavior under difficult circumstances (H. H. Kornhuber, 1961). We began our investigation of volition with experiments on voluntary movements. At that time nearly nothing was known about the mechanisms of will underlying voluntary movement (Eccles & Zeier, 1980).

2. Methods In our research, four parameters of brain activation have been measured: (a) performance-related shifts of the cortical steady (DC) potential, (b) event-related EEG-spectrum, (c) regional cerebral blood flow (rCBF), and (d) event-related changes of neuromagnetic fields (magnetoencephalography, MEG).

2.1. Performance-related DC shifls Changes of cortical activation are associated with changes of the cortical steady (DC) potential. Increasing levels of cortical activation (changes from sleep to wakefulness, increased awareness, seizure activity, etc.) lead to a rise of surface negativity that is caused by the augmentation of excitatory postsynaptic events at dendrites of superficial cortical layers (for reviews see Caspers, Speckmann, & Lehmenkuhler 1980; Creutzfeldt, 1983). This is true for both global and regional changes:

110

H. H. Kornhuber ef al.

Activations of circumscribed cortical areas cause local shifts of the cortical DC potential of those areas as demonstrated by (a) recordings of field potentials between surface and depth in animals (e.g., Sasaki & Gemba, 1982), or (b) by recordings from subdural electrodes paired with a distant reference electrode (Neshige, Luders, & Shibasaki, 1988). The conductivity of the skull for currents enables the recording of cortical DC shifts even through scalp electrodes. Since conductivity of the skull is low as compared to brain tissue, there is a filtering effect, in that spatial resolution is diminished in scalp recordings as compared to epicortical recordings. In all our experiments, brain potentials have been measured from 10 to 13 positions of the scalp with linked-ears serving for reference. Performance-related cortical DC shifts, as measured by scalp recordings, are usually small, although amplitudes exceeding 30 pV are possible. In order to achieve the required signal to noise ratio, many trials have to be collected and averaged (48 to 128 trials or more) in a time-locked manner, with each trial triggered by an experimental event. In our experiments, electrodes have been placed according to the 10-20 system recommended by Jasper (1958). The following spatial relations between electrode positions and cortical gyri have been found by Homan, Herman, and Purdy (1987): Fp l F p2 (rostra1 limit of superior frontal gyrus); F3F4 (middle frontal gyrus); C3/C4 (precentral gyrus, shoulder to wrist area); P3P4 (superior parietal lobule near intraparietal sulcus, superior to posterior portion of supramarginal gyrus); T3/T4 (overlapping middle and superior temporal gyri); 0 1 / 0 2 (occipital lobe). Some additional electrode positions have been used: FCz (midway between Fz and Cz), C1 (midway between Cz and C3), and C2 (midway between Cz and C4). The odd-numbered sites lie above the left hemisphere, the even numbered sites lie above the right; the z sites are located along the midline. In all experiments, great care has been taken to eliminate artifacts caused by eye movements: During the analysis period, subjects had to fiiate on a point straight ahead and had to prevent blinks and other eye movements. Trials contaminated by eye movements or blinks were excluded from the average. For this purpose, the electro-oculogram (EOG) has been recorded. Recent experiments introduced true DC recordings (M. Lang, Lang, Uhl, et al., 1987; W. Lang, Lang, Podreka, et al., 1988; W. Lang, Lang, Uhl, Koska, et al., 1988; W. Lang, Zilch, Koska, Lindinger, & Deecke 1989). In one experiment, the radial current density into the scalp was calculated and mapped (Lindinger, Lang, Obrig, Kristeva, & Deecke, in

Cerebral Correlates of Volition and Attention

111

press). This procedure, as suggested by Hjorth (1975), Nunez (1981), and Perrin, Bertrand, and Pernier (1987), has the advantage of being reference-free and of decreasing the effect of volume conduction. To estimate radial current density, the scalp distribution of DC potentials was interpolated between electrode positions by use of polynomial cubic splines. Based on this continuous surface, the radial current density was estimated by applying the 2-dimensional Laplacian operator. The influence of the amplitude at an electrode position to the shape of the potential-surface can be weighted by spline functions, a procedure that is important for estimating potentials at the boundaries of the map (see also Koles, Kasmia, Paranjape, & McLean, 1989). For calculations and display of images, a simple head model is used: The scalp is assumed to be a square with left/right boundaries at T 3 n4 and anterior/posterior boundaries at Fpl,Fp2/01,02. Each side of the square is subdivided in 20 sections; this results in a spatial map-resolution of 400 points. For boundary estimation it is assumed that potentials decrease continuously to zero outside the square.

2.2. Event-related EEG spectrum In our experiments, EEG data have been filtered (32 Hz, 48db/octave low pass) and digitized at a sampling rate of 100 data points per second. Event-related EEG spectra have been calculated in periods of 1 to 1.28 s each using the Fast Fourier Transformation (FFT). For each spectrum, 30-50 sweeps have been averaged and Bartlett’s smoothing method has been used to reduce the cut off errors caused by the short epoch length (Diekmann, 1985). Data have been strictly edited off-line; eye, lid, and muscle artifacts have been rejected. Mean power density (MPD) has been computed for the classical frequency bands: (Delta, Theta [e; 3-7 Hz], Alpha [a; 7-14 Hz], and Beta) and for the total spectrum. Two parameters are of particular interest in our experiments, (a) the performance-related increase of the e-MPD, called IN-e-MPD and (b) the performance-related attenuation of the a-MPD, called AT-a-MPD. IN-e-MPD and AT-a-MPD have been calculated relative to the resting state, which is taken from a period that precedes volitional task initiation by 3 to 4 s. MPDs of performance-related epochs show considerable inter-individual variation. The calculation of changes between resting state and performance-related epochs minimizes the effects of this interindividual variability. So far, the a-rhythm has been widely investigated in humans, but little is known about the origin and physiological significance of the

112

H. H. Komhuber et al.

e-rhythm. In lower primates, as well as in humans, the "cortical" e-rhythm (as recorded directly from the cortex or by scalp electrodes) is related to effort in information processing and learning (Doyle, Ornstein, & Galin, 1974; Ahern & Schwartz, 1985; Dolce & Waldeier, 1974; Haslum & Gale, 1973; Ishihara & Yoshi, 1972; Gale, Christie, & Penfold 1971; Otto, Gruner, & Weber, 1983a, 1983b; Rugg & Dickens, 1982; M. Lang, Lang, Diekmann, & Kornhuber, 1987; W. Lang, Lang, Kornhuber, Diekmann, & Kornhuber, 1988). Little is known about the relationship between the "cortical" e-rhythm and the hippocampal e-rhythm; the latter is well investigated in primates and has also been found in humans (Arnolds, Lopes da Silva, Aitinik, Kamp, & Boeijinga 1980; Lesse, Heath, Mickle, Monroe, & Miller, 1955). In primates, hippocampal e-activity has been found to be linked to cognitive behavior such as learning (Crowne, Konow, Drake, & Pribram 1972; Crowne & Radcliffe 1975). Indirect evidence for a relation between limbic e-rhythm and "cortical" e-rhythm has been found in epileptic patients (Talairach, et al., 1973): Stimulation of the anterior cingulate was associated with a spread of excitation to the adjacent frontomedial cortex (including the SMA), and a steady 3-8 Hz rhythm developed in the EEG during stimulation, which was maximum at the vertex. Topographical analyses of the e-rhythm in scalp recordings point to an origin in the fronto-mesial cortex (Koles et al. 1989). For all these reasons, a functional relationship between the limbic system and the frontomesial e-rhythm in humans is assumed.

2.3. Regional cerebral blood flow by Tc-99m-HMPAO brain SPECT Single Photon Emission Computerized Tomography (SPECT) using

Tc-99m-Hexamethylpropyleneamineoxime(Tc-99m-HMPAO) makes it possible to visualize the distribution of regional cerebral blood flow (rCBF) of the brain. Tc-99m labelled HMPAO crosses the blood brain barrier with a high first pass extraction fraction. The tracer is deposited in brain tissue within the first two minutes after intravenous injection and is distributed there proportional to the cerebral blood flow. Within the immediately subsequent 2-hour period, redistribution of the tracer has not been measurable (Neirinckx et al., 1987; Podreka et al., 1987; W. Lang, Lang, Podreka, et al., 1988). Dosages of 0.2 mCi/ kg body weight have been applied in our studies. Local radioactive count rates (gamma-ray) have been measured by means of a dual-head, rotating, scintillation (gamma) camera. A parallel-hole collimator of high resolution (FWHM: 12 mm in the horizontal plane) has been used and 60 (2 x 30) projections have been achieved within 30 min (60 s per angle) with a linear sampling

Cerebral Correlates of Volition and Attention

113

distance of 3.125 mm. Projections have been filtered before reconstruction and corrected for tissue absorption. Seven 3.125 mm thick slices have been summed up consecutively to achieve a set of transverse slices 21.9 mm thick covering the entire organ for final evaluation. Guided by anatomical templates, regions of interest (ROI) have been defined within five consecutive 21.9 mm thick transverse slices. ROIs are not necessarily congruent with anatomical boundaries, but it is assumed that the proportion of Tc-99m-HMPAO deposition in the respective anatomical structure is a major source for variation of the count rate of the whole region (Goldenberg, Podreka, Steiner, & Willmes, 1987). An absolute quantification of Tc-99m-HMPAO SPECT studies is not yet possible. For that reason, a relative local count rate (RI, regional index) has been obtained by referring the tracer concentration within each ROI to the mean count rate across all regions. It is important to note that only relative patterns of perfusion and their task-induced changes can be described. No information is available about absolute level of perfusion and task-induced changes of this parameter.

2.4. Magnetoencephalography Moving charges are associated with magnetic fields. In the nervous system, excitations of neurons cause intracellular and extracellular currents. Assuming a spherical model of the brain, magnetic fields as caused by extracellular currents cancel each other because of the symmetrical spreading of currents within the surrounding brain tissue. It is, therefore, the intracellular current that contributes to magnetic fields at the scalp surface. If particular areas of the cortex become activated, for example, by afferents from the periphery, intracellular currents of activated cortical neurons have an orientation that is perpendicular to the cortical surface. The reason for that orientation of intracellular currents lies in the fact that neurons together with most of their dendrites and neurites are oriented parallel to each other, and perpendicular to the cortical surface. The summation of intracellular currents within a certain cortical area is described by "current dipoles." Since the cortex is folded into gyri, dipoles have either tangential (two tangential directions in the three-dimensional space) or radial orientations with respect to the scalp surface. Dipoles having oblique directions can be decomposed geometrically into tangential and radial vectors. To sum up, magnetoencephalography (MEG) measures changes of the magnetic flux caused by tangential components of current dipoles which vary its strength over time.

114

H. H. Komhuber et al.

At the neurological clinics of Ulm and Vienna multi-channel MEG devices have been built. MEG is recorded in a magnetically and electrically shielded room, in Vienna, with two seven-channel, secondorder gradiometers using dc-SQUIDS (Superconducting Quantum Interference Device) with coils having a diameter of 1.8 cm (coil baseline 4.0 cm; Biomagnetic Technologies, Inc., San Diego, CA), whereas in Ulm a 24-channel system is used. MEG system noise is reducible by shielding to 10 femtoTesla/Hz’’2 .

2.5. Methodological considerations Physiology and informative content differ between electrophysiologic and circulatory-metabolictechniques: Electrophysiologic techniques (EEG, MEG) measure neuronal activity with high temporal resolution that is only limited by the setup of the data acquisition system. The ability to localize neuronal activity in the brain is different in EEG and MEG. In MEG, the present use of gradiometers and limitations of sensitivity restrict this method to investigations of neuronal activities which are close to the surface. Magnetic flux diminishes rapidly when increasing the distance of the sensing system. But this neuronal activity closely related to the surface can be localized with remarkable precision. For example, it is possible to demonstrate in humans the somatotopic organization of the primary motor cortex (MI; Cheyne, Kristeva, Lang, Lindinger, & Deecke, in press) or of the somatosensory cortex (Okada, Tanenbaum, Williamson, & Kaufman, 1984). Even the tonotopic organization of the primary acoustic cortex has been demonstrated with this method (Romani, Williamson, Kaufman, & Brenner, 1982). In the EEG, the ability to localize neuronal activity is limited. This is because extracerebral tissues (in particular the skull) have remarkable smearing effects on the scalp potentials generated by extracellular currents. There is no smearing effect of extracerebral tissue on magnetic fields. The EEG and the MEG delineate two aspects of neuronal activity and complement each other. For both EEG and MEG, considerable progress is going on in two directions, one localizing dipoles, and the other decomposing dipoles which become active simultaneously in different cortical areas. The combination of multi-channel EEG and MEG devices, coupled with realistic models for dipole decomposition and dipole fitting, along with magnetic resonance imaging (MRI), provides the possibility of assessing neuronal activity with high resolution in time and space. Circulatory-metabolic methods like SPECT do not measure neuronal activity directly. Rather, these techniques measure regional cerebral blood

Cerebral Correlates of Volition and Attention

115

flow or metabolic parameters such as the regional glucose utilization in the brain. Close connections between neuronal activity and blood flow and metabolism are assumed. The temporal resolution of these techniques is limited; 20 s seems to be the best resolution at present for PET (in the Tc-99m-HMPAO SPECT technique between 90 and 120 s). The spatial resolution is high, in PET (positron emission tomography) resolution of 4 to 6 mm in the horizontal plane have been achieved (resolution of the Tc-99m-HMPAO SPECT technique is about 10 mm, but could be improved).

3. Results and Comments 3.1. The initiation of volitional actions: When to do 3.1.1. Human pathophysiokiigy. Real volitional action is spontaneous, that is, it is triggered by internal events. This is in contrast to re-actions triggered by external cues from the environment. In Parkinsonian patients, the clinical phenomenon of kinesia paradoxica points towards an impairment of the internal trigger and a dependency on external cues: the otherwise frozen individual may move briskly when a loud verbal command is given or when an accompanying person makes a step forward. The symptom of kinesia paradoxica, therefore, suggests that, while motor plans and motor programs are maintained, the trouble is a matter of access: the internal access is disrupted, the external one is partially preserved. Acute, unilateral lesions of the supplementary motor area (SMA) produce a transient l'akinesia'' or a lack of spontaneous activity on the contralateral side (Penfield & Welch, 1949; Laplane, Talairach, Meininger, Bancaud, & Orgogozo, 1977) or a transient inability to initiate speech (Masdeu, Schoene, & Funkenstein, 1978; Jonas, 1981). Selective impairments of volitional movements with preserved ability to react to external cues have been described in patients with lesions of the frontomesial cortex by Beringer (1944) and by Goldenberg, Wimmer, Holzner, and Wessely (1985). A patient with an infarction in the territory of the right anterior cerebral artery was unable to voluntarily move his left arm, but did so when grasping objects with his left hand which were moved towards him (authors' observations). The so-called "alien hand sign" (Brion and Jedynak 1972) has also been observed in patients with lesions of the SMA (Goldberg, 1985). The "alien hand sign" describes the phenomenon in which patients start to perform apparently purposeful actions but experience them as not being started by their own will. This

116

H. H. Kornhuber et aL

symptom seems to be compatible with a role of the SMA in the initiation of volitional movements. 3.1.2. Human voluntary movement physiology: The Bereitschaftspotential. Research on movement-related brain potentials in humans started in 1964 by investigating brain potentials preceding self-initiated rapid flexion of the index finger (H. H. Kornhuber & Deecke, 1964, 1965). A slow, surface negative potential shift started about one second before the initiation of the movement, as defined by the first EMG response (Figure 1). It has been called Bereitschafrspotential (BP) or readiness potential. The BP has a rather consistent temporo-spatial distribution over the scalp: It starts in leads of the frontocentral midline, namely in Cz or FCz (Deecke, Grozinger, & Kornhuber 1976; Kristeva, Keller, Deecke, & Kornhuber 1979; Grozinger, Kornhuber, & Kriebel 1979; Boschert, Hink, & Deecke 1983; Deecke, Heise, Kornhuber, Lang, & Lang 1984; W. Lang, Lang, Heise, Deecke, & Kornhuber, 1984). It is important to note that even in unilateral movement, the BP starts bilaterally symmetrical in central and parietal recordings. In finger movements, BP becomes lateralized in central regions about 500 ms prior to movement onset, with the larger amplitudes contralateral to the performing hand (Deecke, Scheid, & Kornhuber, 1969; Deecke et al. 1976). BP maxima at movement onset have a characteristic pattern as well: maxima have been found in FCz and Cz, or C1 for right sided finger movements (C2 for left). Changes of BP topography associated with movements of different parts of the body (fingers, toes, hip, etc.) have demonstrated that the primary motor cortex (MI) is activated, starting between 200 and 500 ms prior to movement onset. Fingers and toes, for instance, have different representation areas in MI, with toes being located in mesial parts and fingers in lateral parts of the precentral gyrus. Comparing finger and toe movements, changes of BP topography at the scalp surface can be predicted by current dipoles arising in the particular representation area of the MI cortex (Boschert & Deecke, 1986). With magnetoencephalographic recordings (MEG) the current dipole in the MI cortex prior to self-paced movements could be localized (Deecke, Weinberg, & Brickett 1982, Deecke, Boschert, Weinberg, & Brickett 1983; Cheyne & Weinberg, 1989; Cheyne et al., in press). Recent MEG measurements in unilateral finger movements have provided evidence that not only the contralateral MI cortex becomes active prior to movement onset but also the ipsilateral one (Cheyne & Weinberg, 1989), confirming what Kornhuber and Deecke

Cerebral Correlates of Volition and Attention

R

L/R

117

p r e c

prec

Figure 1. Slow shifts of the cortical DC potential (Bereitschaftspotential, BP) preceding volitional, rapid flexions of the right index finger a t time t = 0 s (vertical line). Recording positions were precentral left (L prec, C3), precentral right (R prec, C4), mid-parietal (Pz). Recordings were unipolar, with linked ears as reference. The difference between the BP in C3 and in C4 is displayed in the lowest graph (UR prec). Superimposed are the results of eight experiments with the same subject on different days. From Deecke et al., 1976.

118

H. H. Kornhuber et al.

had already reported in 1965. But the signs of MI activation appear late (within the last 500 ms before movement onset) in the chain of cortical activation preceding self-paced movements. What then generates the early BP? BP is already in evidence 1 to 2 s prior to movement onset in recordings over the fronto-central, mesial cortex (FCz and Cz) and has its clear maximum there. These two characteristic features are present in various movements such as speech production (Grozinger et al. 1979; Deecke, Engel, Lang, & Kornhuber 1986), saccadic eye movements (Becker, Hoehne, Iwase, & Kornhuber, 1972) or limb movements, and thus point to a common cortical structure which seems to be involved in the initiation of all kinds of volitional movements. In 1978, recordings of the BP in patients with Parkinson’s disease (Deecke & Kornhuber, 1978) and measurements of the regional cerebral blood flow (Lassen, Ingvar, & Skinhoj, 1978; Roland, Larsen, Lassen, & Skinhoj, 1980) led to the following hypotheses: (a) The mesial, fronto-central cortex, including the supplementary motor area (SMA) becomes active in volitional movements. (b) It is this cortical area whose activation causes the early component of the BP in FCz and Cz. (c) BP maximum in the midline (Cz and FCz) is not due to a summation from more lateral potentials on both sides. MEG studies have supported the concept that at least two cortical areas become activated prior to volitional finger movements: Initial activity can be picked up from the SMA, subsequently the MI cortex is additionally activated (Deecke, Boschert, Brickett, & Weinberg, 1985). 3.1-3. Centralization of the starting function in volitional actions. Studies of the Bereitschafspotential, and observations in patients, give consistent evidence that the supplementary motor area (SMA) is the central key structure transducing the will-to-move into effective actions. In other words, the Sh4A has a common starting function for the various kinds of volitional actions such as movements of eyes, limbs or tongue (H. H. Kornhuber & Deecke, 1985). The centralization of the starting function is all the more remarkable for the fact that execution and control of different kinds of movements are widely decentralized in the human brain. For example, the musculature of tongue and mouth is innervated when speaking or chewing in a very precise way. But different parts of the cortex are involved when coordinating tongue and mouth during chewing on the one hand and speaking on the other. In speech production, the area of Wernicke in the temporal lobe and the basal ganglia seem to be important in creating and acoustically controlling the various phonemes (Brunner, Kornhuber, Seemiiller, Suger, & Wallesch,

Cerebral Correlates of Volition and Attention

119

1982; H. H. Kornhuber, 1984a,b; Wallesch, Henriksen, Kornhuber, & Paulson, 1985). Motor systems for eating and chewing developed early in evolution, long before the production of speech. In rodents, the primary motor cortex developed along with the necessity of tactually controlling movements of tongue and mouth. In humans, the primary motor cortex not only controls movements of tongue and mouth but also has the function of tactually and proprioceptively controlling movements of limbs and fingers (particularly, in precision movements; H. H. Kornhuber, 1971, 1974, 1984a,b; Aschoff & Kornhuber, 1975). Furthermore, movements of the eyes are not represented in the primary motor cortex. Saccadic eye movements are organized in the pontomesencephalic region, the parieto-occipital cortex and under certain circumstances in the frontal eye field (Bruce & Goldberg, 1985). Although the motor system is decentralized and widely distributed in the brain, the Bereitschaftspotentialpoints towards a common structure, the SMA, which is involved in the volitional initiation of actions. Meanwhile, positron-emission-tomography and animal experiments have substantiated the activation of the SMA in voluntary saccadic eye movements (Fox, Fox, Raichle, & Burde 1985; Schlag & Schlag-Rey, 1985). SPECT (Single Photon Emission Computerized Tomography) data support SMA activation in speech production (Ingvar, 1983). The SMA is hypothesized to transduce the will-to-move into effective actions (H. H. Kornhuber, 1980). This process requires input from the motivational system to the Sh4A which has been demonstrated anatomically. There is direct or indirect (via the thalamus) input from hypothalamus, amygdala, inferior temporal cortex and prefrontal cortex (Jones, 1983; Wiesendanger & Wiesendanger, 1985). Patients with lesions of the SMA have and experience their will to move; they are able to select between motives and drives (what to do), but the transduction of their intentions into actions is disrupted. 3.1.4. Selecting the "right"moment to start a movement. The hypothesis that the starting function of volitional actions is centralized and organized within the SMA has implications for the motor system. For instance, consider a person performing a rapid reaching movement while walking. Movements of this kind have consequences for posture and have to be embedded into the ongoing motor behavior. Therefore, anticipatory mechanisms for postural adjustment and for the temporo-spatial integration of ongoing and intended actions are necessary. The "right" time to start the reaching movement cannot be settled by the person without having the knowledge that anticipatory adjustments are being prepared.

120

H. H. Komhuber et al.

Neuro-anatomical data would substantiate this view: The SMA has input from the sensorimotor cortex, from the cerebellum and basal ganglia via the thalamus, and from the motivational system (from hypothalamus, amygdala, inferior temporal cortex and prefrontal cortex). Because of these circuits linking SMA to other brain structures, the function of the SMA should not be studied in isolation. Indeed, the SMA seems to play a central role (both anatomically and functionally) in the critical decision of when to start a movement. The SMA has been linked to several functions such as anticipatory adjustments of posture (Massion, Viallet, Massarino, & Khalil, in press) instruction-induced preadjustments of sensorimotor behavior (Tanji, 1984), and the "programming" of action sequences which are "projectional" in that they rely on model-based predictions (Goldberg, 1985). According to the present concept these data and hypotheses have to be re-interpreted and integrated: (a) Anticipatory adjustment and integration of ongoing and intended behavior have to be settled before volitional actions can be started. (b) The starting function of volitional actions is centralized in the SMA. (c) In order to select the "right" time to start the movement, the SMA must have, at least, some information about, or some kind of control over, anticipatory mechanisms. (d) It is not reasonable to assume that all anticipatory mechanisms are organized by the SMA itself. Rather, it is likely that other brain structures get involved. However, the heavy interconnection of the SMA with other brain structures is entirely consistent with the hypothesized key role of initiation and coordination. In a recent experiment, movement-related potentials were recorded during an ongoing performance when an additional action, having strong tendencies to interfere with the ongoing performance, was initiated. Four conditions were investigated (W. Lang, Obrig, Lindinger, Cheyne, & Deecke, 1989): (a) Subjects started to flex and extend repetitively their two index fingers. Frequency of finger flexions was about 2/s. Subjects were instructed to start synchronous movements of the left index finger after some time (about 4 to 6 s after having started the right finger). This task has been called RH-SI (the right hand started, and the task was simple in nature). In (b) the left index finger started the simple task (LH-SI). In (c) the right finger started as in RH-SI with flexions at a frequency of 2/s. But now subjects had to perform flexions at a rate of 3/s with the left hand (RH-COM; right hand, complex situation). As in all tasks, subjects were free to determine the onset of left and right finger movements (in RH-COM the left finger usually started with a delay of 4 to 6 s after the right one). Task (d) corresponded to (c) but now the left

Cerebral Correlates of Volition and Attention

121

finger started to move and the right one followed (LH-COM). Only music students participated in these experiments. They were well trained to perform dissociative rhythms with their index fingers but stated that it took a high degree of effort and concentration to start the 3/s rhythm when already moving the other finger at 2/s. The main point of interest was whether brain potentials preceding movement onset of the second finger (at time t*) would differ between complex (COM) and simple (SI) tasks. Consequently, movement-related, negative, DC shifts were averaged, time-locked to t*. Amplitudes were referred to a baseline taken in the resting state (4 to 3 s before initiating the whole task by moving the first finger). When preparing to start the 3/s rhythm with the second hand, negative potential shifts were larger than they were in the simple task (Figure 2). This difference (Nd: difference of negativity; i.e., extra negativity for the complicated rhythm) developed early -- about 4 s before t*-- and rendered large and significant differences over the central midline (Cz, C1, C2) and in Pz. For statistical comparisons, mean negativities across intervals of one second (N,: between t = 4 s and t = 3 s before t*; N3; N,; Nz) were calculated. Effects of the within-subject factors, "complexity" and "performing hand", on mean negativities were calculated by repeated MANOVAs. In C1, C2, Cz, P3, P4, and Pz, task complexity had a significant effect in all 4 intervals, N, to N4 (e.g., in Cz: F = 1 . 8 , ~< .004 for N,; F = 1 1 . 0 , ~< .005 for N3; F = 1 6 . 9 , ~< .001 for N2; and, F = 2 4 . 8 , ~< .0001 for Nz). Studies concerning inter-limb coordination point to limitations when executing movements simultaneously. Simultaneous movements can be performed accurately as long as they are harmonically related, either in phase or in alternation. Movements that are not harmonically related display interference when performed together (e.g., Klapp, 1979). Kinematic analyses indicate a common time scaling when coordinating bimanual movements (Kelso, Putnam, & Goodman 1983). Bimanual sequences at different rhythms, as investigated in the present study, create tendencies of mutual interference. Although subjects were trained musical students, effort was needed to overcome these interference tendencies . One result of the present study was that interference tendencies are effective not only when performing the task but also long before, in clear anticipation of the task (as evidenced by the DC negativity recorded). Any control mechanism that might, hypothetically, offset interference during task performance might also be too late to be fully effective. Effective coordination would appear to require a control mechanism capable of anticipating impending interference, such as that suggested by

122

H. H. Komhuber et al.

Figure 2. Slow DC-potential shifts preceding the time (t*: vertical line) at which the second hand (here, right hand) starts to move. There is already an ongoing performance of the other hand (here, left hand) with finger movements at a frequency of about 3s. Thin lines: The right finger (second hand) is brought in with the 3/s rhythm at t*. Thick lines: The right finger joins in with the 2/s rhythm at t*. Baseline was taken from the resting period before movement onset of either hand. Averages across all subjects, negativity up. Abscissa: time (s). Ordinate: amplitude (scaled in 20 pV).

PI

P4

PI)

1

I

c4

c3

C1

c2

c1

FZ

F4

FS

the early Nd. Nd may indicate increased motivational and intentional involvement when subjects prepare to act in a conflicting task. In other words, in order to get ready to start the movement, information about ongoing and intended actions has to be integrated by sensory input or by memory-based models of movement outcome and consequences. This integration requires effort and motivation. Nd appears over the central midline (Cz, C1, C2) and parietally (Pz). In view of this topography it is reasonable to assume that the mesial cortex, including the SMA, is involved in this process. Interestingly, at times, subjects did not reach the state of readiness in the complex task, and in these instances were not able to initiate the 3/s sequence. 3.1.5. Afer the starting signal has been released. Systematic analysis of the potential’s time course reveals that in approximately 85% of all subjects the BereitschufspotentiaZ reverses to a positivity about 90 ms prior to movement onset (pre-movement positivity, PMP; Deecke et al., 1969; Deecke et al., 1976). About 50 ms prior to movement onset, a sharp

Cerebral Correlates of Volition and Attention

123

negative potential shift arises in the recording over the contralateral MI cortex. This negative potential has been called motor potential (MP; Deecke et al., 1969). It does not reflect the first activation of the MI cortex. As described above, the contralateral and, to a lesser degree, the ipsilateral MI cortex become activated 500 ms before movement onset. MP may correspond to synaptic events in conjunction with the pyramidal cell firing in area 4 (MI cortex) as the activity in the final motor pathway to alpha motoneurons in the spinal cord. 3.1.6. Basal ganglia and cerebellum in the initiation of volitional actions. The latency of about 40 ms between PMP and MP led to the hypothesis that the decision to move is not transferred in a direct way to the MI cortex, but, rather, via a subcortical loop including basal ganglia and the cerebellum (H. H. Kornhuber, 1971, 1974). Recordings of single neurons have supported such a view (Lamarre, Spidalieri, Busby, & Lund 1980; Melnick, Hull, & Buchwald, 1984). Recordings in Parkinsonian patients during stereotactic operations suggested the existence of slow negative potential shifts in the thalamus prior to uncued, self-paced movements (Knapp, Schmid, Ganglberger, & Haider 1980; Straschill & Takahashi, 1980). A recent study in monkeys demonstrated slow negative potential shifts arising about 500 ms before movement onset in several subcortical nuclei such as substantia nigra, red nucleus, midbrain reticular formation and caudate nucleus (Bauer & Rebert, in press). Therefore, regarding volitional action, theories about "preparatory set" should include circuits linking basal ganglia, cerebellum and cortex. 3.1.7. SMA and the temporal organization of motor sequences. In the preceding section, a theory has been presented in which the selection of the time to start a movement (when to start) is made by the SMA, with its multiple afferents from motivation and sensorimotor systems. This theory separates the decision when to do from other activities involved in volitional movements such as what or how to do. In order to make this functional separation clear, it is important to contrast it against an alternative theory of SMA function put forward by Roland et al. (1980) and Goldberg (1985). In his 1985 review article, Goldberg suggested that: the SMA has an important role to play in the intentional process whereby internal context influences the elaboration of action. It may be viewed as phylogenetically older motor cortex, derived from anterior cingulate periarchicortical limbic cortex, which, as a key part of the medial premotor system, is

124

H. H. Komhuber ef al. crucial in the "programming" and fluent execution of extended action sequences which are "projectional" in that they rely on model-based prediction .... the SMA plays an important role in the development of the intention-to-act and the specification and elaboration of action through its mediation between medial limbic cortex and primary motor cortex. (p. 567)

Roland (1984) proposed that the SMA either elaborates or retrieves from memory the necessary information to form a short sequence of motor commands in which the elementary movements to be executed are specified exactly. With an example from the motor sequence test one could say that the SMA specified: (1) the fingers to be moved in the near future, (2) which were the movements of the individual fingers (i.e., opposition, flexion, extension) and (3) the sequence of (1) and (2). (pp. 209-210) Goldberg and Roland propose that the SMA forms and specifies subroutines in motor sequences during preparation and execution. According to their view, the SMA does not serve to transduce the willto-move into effective action, that is, to give the starting signal, but is involved in the process of determining how to perform the movement. In one experiment, subjects were asked to simulate internally the performance of the motor sequences without actually executing them (Roland et al., 1980). It was found that the SMA was still activated, while the primary motor cortex was not. The following three experiments were designed to test our theory against the one of Roland and Goldberg: Experiment I (W. Lang, Lang, Uhl, Koska, et a]., 1988): In this study 20 young, healthy human subjects performed four different kinds of motor sequences. In all tasks, subjects held their index fingers in an intermediate position during the resting period and started to move the fingers in order to repeatedly reach three positions, a flexed, an intermediate and an extended one. In SI-S, movements were performed simultaneously in the same direction. In SE-S, the right index finger started, the left finger followed with a delay of one movement (sequential). In SI-D, the sequence was initiated by flexing the right finger while simultaneously extending the left. Thus, the two index fingers moved simultaneously but in different directions. In SE-D, movements were

Cerebral Correlates of Volition and Attention

125

performed sequentially and in different directions. Since it was our goal to study the planning and execution of learned movement sequences, subjects practiced each sequence thoroughly in pre-experimental sessions. In the experiment, subjects started the motor sequence at their own volition and performed the task for at least 6 s. An epoch of 5 s prior to movement onset and 6 s thereafter was analyzed. Two parameters were taken to describe negative DC shifts: N-BP (mean negativity over the last 250 ms prior to the volitional initiation) in order to describe the Bereitschafs~otentl,and N-P (mean negativity measured between 2 and 4 s after movement onset) to describe performance-related negative DC shifts. In this experiment, effects of the within-subject factors, "temporal organization" (simultaneous or sequential) and "spatial organization" (same or different directions), were tested by MANOVA. The results were that during the two sequential tasks, SE-S and SE-D, there was a large and sustained negative DC shift in recordings over the mesial fronto-central cortex. In contrast, in the two simultaneous tasks, SI-S and SI-D, performance-related negativity (N-P) declined rapidly over the epoch. This difference between SE and SI tasks was restricted to recordings of the mesial fronto-central cortex and had its maximum in Cz (F = 18.9, p < .0001; see Figure 3). Performancerelated DC shifts did not vary as a function of "spatial organization"; that is, there was no difference whether subjects moved their index fingers in the same or in different directions (for Cz: F = 2.2). Using the Wilcoxon, matched-pairs, signed-rank test, the Bereitschafs~otentialin Cz was significantly larger (p < .017, two-tailed) in SE-S as compared to SLS. Topographical differences of N-BP and N-P between SE-S and SI-S are displayed chromatically in Figure 14 on page 152. This color figure demonstrates the additional activation of the mesial frontocentral cortex in SE-S as compared to SI-S when preparing and executing these tasks. According to Roland's view, the way in which the finger had to be moved (same or different directions) should have had an effect on SMA activity and on performance-related DC shifts in recordings of the frontocentral midline. But such an effect was not found in our experiments. Note the large differences between SE and SI tasks, and the localized appearance of Nd (difference of negativity) in recordings from the frontocentral midline (which is likely to pick up the activity of the SMA). In SI-tasks, performance-related negativity decreases, perhaps reflecting automation of the motor sequence when the two index fingers act in phase. The situation is different in SE tasks. There is a sustained, and rather localized negative DC potential in Cz, C1 and C2. The reason

H. H. Komhuber et al.

126

s I-s

S E-S

Figure 3. Grand means averaged across all 20 subjects for each of 4 conditions: SI-S (simultaneous, same direction), SE-S (sequential, same direction), SE-D (sequential, different directions) and SI-D (simultaneous, different directions). Movement onset at t = 0 s (first activity of the right flexor indicis). The Bereitschaftspotential precedes the voluntarily initiated motor performance; task execution is accompanied by a negative DC potential shift (performance-related negativity). Data from W. king, Lang, Uhl, Koska, et al., 1988. may be that in SE-tasks, movement initiation is dissociated between the two index fingers. This situation is demanding for the central structure (SMA) responsible for starting the movement, and it is known that the two SMAs act in accord (i.e., even in unilateral movements the SMA of either hemisphere is activated, Brinkman & Porter, 1979; Roland, Meyer, Shibasaki, Yamamoto, & Thompson, 1982). Thus, the starting signal to move the index finger of one side can only be given when control mechanisms ensure that a movement of the other side is inhibited. However, this does not imply that the SMA also has to organize this inhibition. Rather, it implies that the SMA must have information that contralateral inhibition is settled when giving the starting signal. The increased activation of the SMA in the SE task, as compared to the SI

Cerebral Correlates of Volition and Attention

127

task, has recently been substantiated in an experiment using the Tc-99mHMPAO SPECT technique (W. Lang, Podreka, & Deecke, unpublished data). Experiment II (W. Lang, Oldenkott, Goldenberg, Reisner, & Deecke, 1988): A total of 15 patients with chronic unilateral lesions of the SMA were examined using the SE and SI tasks. Latencies between the acute state of SMA lesion and date of examination ranged between 8 and 84 months (mean: 34). Eight patients had the lesions in their right, 7 in their left hemisphere. They had no paresis and performed at normal rates in a unilateral tapping test. In the study, they performed 64 trials of two different movements: Simultaneous flexions and extensions of the two forefingers (SI task), and sequential movements of the two forefingers (SE task: extension on the right side, extension left, flexion right, flexion left, etc.). Trials were voluntarily initiated and lasted 6 s. Movements were measured using a splint with potentiometers at the proximal finger joint. The following symptoms were observed: (a) Bradykinesia contralateral to the lesion in SI and SE, (b) marked deceleration of initial movement, (c) switching from sequential performance into simultaneity and (d) frequent failure to initiate or inhibit a movement on one side or the other (as demonstrated in Figures 4 and 5). Experim.ent III (M. Lang, Lang, Kornhuber, Groger, & Kornhuber, 1988): Eight right-handed subjects performed flexions of their right index finger in three different conditions. In one situation, subjects had to initiate single movements according to a precise timing structure which was defined by four intervals, for example 3, 7, 5, and 2 s. In this task, the Bereitschufispotentil was significantly larger in fronto-medial electrodes than in conditions involving either rhythmic, repetitive movements (at about 3/s) or movements which were performed at irregular intervals (see Figure 6). In addition to these experiments, clinical symptoms such as kinesia paradoxica in Parkinsonian patients and the inability to initiate speech and limb movements in acute unilateral SMA lesions can also be taken as evidence that the decision when to do is a process per se and has to be separated from other aspects of volitional actions. Recently, experiments were conducted to compare movement related DC shifts during SE and SI tasks as subjects either moved or imagined moving (mental rehearsal). Preliminary results (in 12 subjects) indicate that in the imagery task, retrorolandic areas are activated during SE and SI, but not the frontocentral midline. In Tc-99m-HMPAO SPECT studies on imagery of motor performances, activation of the SMA could not be found (Goldenberg,

128

H. H. Kornhuber et al.

LH

uuvy U

--m vvu

ex

Figure 4. Finger movements of the right (RH) and left (LH) hand were measured by goniometers. Patients had to perform extensions and flexions in a sequential manner starting with an extension of the right index finger. Examples of omissions of movements of the left finger in various patients (disturbance of movement initiation). Omissions are indicated by a bold horizontal line. In the lower traces, the left finger is not moved at all (left side) or stops moving (right side). As shown, the degree of the symptom "omission of movement" varies. Podreka, Steiner, Suess, & Deecke 1988; W. Lang, Lang, Podreka, Goldenberg, & Deecke, 1989). Results of Experiment I1 are supported by other clinical findings: Patients with unilateral lesions of the SMA not only have difficulty voluntarily initiating single movements but also have difficulties initiating motor elements in a sequence. Jonas (1981) examined speech and found the following symptoms: disturbances to the initiation of propositional speech, hesitations, explosive speech, running of words together, variability in rate of speech emission and speech arrest. Forster (1936) in his remarkable clinical handbook of neurology described the following symptomatology when examining 40 patients who had lesions of the SMA (area 6aB) after surgical treatment for partial epilepsy:

129

Cerebral Correlates of Volition and Attention

Figure 5. Finger movements of the right (RH) and left (LH) hand were measured by goniometers. Patients had to perform extensions and flexions in a sequential manner starting with an extension of the right index finger. Examples of the tendency to continue movements on one side (lack of movement inhibition) instead of alternating between left and right are indicated by arrows.

-

'--W

fl.

r650 ex.

VL 0

2s

I have operated area 6aS [corresponding to the supplementary motor area, SMA] in about 40 patients. Immediate consequences consisted in minor weakness and bradykinesia when moving limbs of the contralateral side or when turning and inclining either head or trunk to the contralateral side....Single, proximal and even distal movements can be performed by the limbs of the contralateral side within the normal scale. Composited movements, however, exhibit a loosening of its structure. There is a disruption of the fluent continuity by which single motor elements are spatially and temporally linked into composited movements. Single motor elements are performed separately, there are delays between them, one element can be omitted for a while and sometimes, an additional and "extra" impulse of will must be given for its initiation. (p. 279; translated into English by authors; for original text, see the Appendix.)

130

H. H. Komhuber et al.

3.2. Physiological signs of anticipatory, task-specific planning Intentionality in action is closely linked to the direction of an action toward a goal, that is, the achievement of a particular outcome. A volitional action is guided by abstract representations of its outcome. As described by Bernstein (1984), an individual has a "model of the future" that includes representations and motor plans of the outcome of an action (the knowledge of how to do). This "model of the future" stems from experience which is acquired by continuous interactions between the individual and its environment. Utilization of this memory-based "model of the future" supports the capacity to prospectively control behavior and to successively improve desired outcomes (Bernstein, 1984). Introspectively, task-specific planning prior to movement onset is a self-evident matter. Many observations are compatible with this assumption: Complex sequences of learned movements can be executed at a rate too fast to be guided by sensory feedback or for the individual components to be each under conscious control (Keele, 1968; Sternberg, Monsell, Knoll, & Wright, 1978). In speech production, one realizes that whole sentences are prepared in advance when observing anticipatory phonematic paraphasias in speech (e.g., "I sope so" instead of "I hope SOt1).

An example for a rather elementary kind of anticipatory, taskspecific "planning" has been given in section 3.1: When different parts of the body perform simple, rapid movements BP topography varies. The analysis of this variation revealed two BP components: A first component could be attributed to an activation of the SMA and remained invariant. A second component was specific for the part of the body moving. It developed at about 500 to 200 ms before movement onset and reflected activation of the primary motor cortex (MI). The question is whether the term "planning" is appropriate for this activation. For a long time it has been believed that the MI cortex is able to initiate and program volitional movements by itself. But it now seems evident that the MI cortex needs either direct or indirect input from basal ganglia, limbic systems, cerebellum, somatosensory association areas, and SMA in order to initiate movements (H. H. Kornhuber, 1971, 1974, 1984a,b). Pre-movement activation of the MI cortex may reflect final steps in the translation of motor programs into patterns of muscular activation. Amplitudes and topography of the Bereitschafspotential depend on structure and/or complexity of forthcoming tasks. Complex movements such as writing or drawing have BPs which are large, particularly early and widely distributed, even appearing in fronto-lateral recordings

Cerebral Correhtes of Volition and Attention

+"

131

*+

Figure 6. Left: Topography of the difference in Bereitschafispotential (BP) between "timed" and "irregular" flexion movements (cf. Exp. III in section 3.1.7). Mean values are taken from 130 ms to 30 ms prior to movement onset. Vertical line in each bar indicates k 1 standard error of the mean. The asterisk indicates a p < .05 in the one-tailed, Mann-Whitney U test. Right: The BP (cortical negativity vs. time, averaged over 8 subjects, movement onset at t = 0) in the "timed" (upper trace) and the "irregular" (lower trace) flexion movement over FCz. Since the experimental condition lacks a resting period, the baseIine was taken from 200 rns to 500 ms after movement onset. Recordings were unipolar, with linked ears as reference. (Schreiber et al., 1983). A sequence of two complex movements has a larger BP than one movement alone (Benecke, Dick, Rothwell, Day, & Marsden, 1985). Complex spatio-temporal sequences are preceded by larger BPs than simple and repetitive finger movements (W. Lang, Zilch, et al., 1989). Taylor (1978) studied the BP during the acquisition of a motor skill. A series of six button presses in a specified pattern constituted the motor task. The BP increased steadily at all electrodes as performance improved, that is, as response time decreased. This close correlation between BP increase and the acquisition of motor skill was taken as evidence that BP reflects task-preparation. After performance

132

H. H. Komhuber et al.

time reached asymptote, BP decreased in the frontal recording, Fz, but remained relatively constant in Cz. Large changes of BP topography take place in situations in which a voluntarily initiated simple finger movement triggers the presentation of stimuli (Deecke et al., 1984; W. Lang et al., 1984). Here the preparation of movement initiation is associated with other processes such as directing attention towards the forthcoming stimuli and preparing for its response. In such situations, BP is overlapped by another slow negative potential shift that has been called directed attention potential, DAP (see section 3.5). In summary, physiological signs of anticipatory, task-specific planning can be studied by the BP technique. Variations of BP topography indicate that various parts of the cortex (frontal, central, parietal) are involved in preparatory processes. As already described above, it turns out to be an oversimplification to contribute such planning functions to the SMA, as proposed by Goldberg (1985). Furthermore, variations of task complexity and structure not only require different amounts of programming but also different levels of motivational and intentional involvement, making it difficult to tear apart these preparatory processes of volitional actions. In order to investigate "preparatory set" neurophysiological techniques are necessary. Measurements of regional cerebral blood flow (rCBF) or metabolism do not posses the temporal resolution to separate states of preparation and those of performance. Some rCBF studies claimed that such a separation would be possible by measuring two separate states, (a) during task performance and (b) when "internally" performing the task without movement. But such argumentation may be erroneous. The "preparatory set" of volitional actions includes the transduction of the will to act into effective movements. The will to act includes the tendency towards its realization. When only imagining a task, that aspect of volition concerfied with the realization of intentions is missing and cannot be investigated.

3.3. Resoluteness: ability to maintain goals and adjust behavior Volitional actions are goal-directed. At times, goals can only be achieved in the distant future. But still, long term goals are maintained, a phenomenon that constitutes a basis for self-continuity during one's life. In the following experiments, subjects had instructions to reduce the error of performance in a conflicting-response-selection task across a series of trials. Maintaining this aim, subjects had to adjust their behavior to resist interference from inadequate stimulus-response patterns and to suppress shifts of attention and thought (M. Lang, Lang, Diekmann, & Kornhuber

Cerebral Correlates of Volition and Attention

133

Figure 7. Subjects tracked a visual target (vt: circle) with a stylus held in their right hand. Horizontal and vertical coordinates [ x(t) and TV-SCREEN y(t) ] of tracking movements were transmitted to the TV screen in a linear ratio and I I displayed as a light spot (f: I feedback). In the learning I task, transmitted coordinates, I either of the horizontal (as PHOTO DETECTOR in the SPECT study) or the vertical (as in the DC-potential study) direction, were IOG multiplied by the factor (-1). In the DC-potential study, subjects fixed their gaze on a fixation point (FIX) in order to prevent artifacts in the EEG recordings. caused by eye movements. Thus, the stimuli were given in the lower field of vision. Subjects were not able to watch movement of the right hand. From W. Lang, et al., 1986.

1987; W. Lang, Lang, Kornhuber, Deecke, & Kornhuber, 1983; W. Lang, Lang, Kornhuber, & Kornhuber, 1986; W. Lang, Lang, Podreka, et al., 1988). In a recent experiment, two parameters of brain activity, performance-related D C shifts and Tc-99m-HMPAO uptake (SPECT, see 2.3) were measured. A total of 17 subjects participated in the SPECT study; in 16 of them, performance-related D C shifts were measured as well. Each subject performed two tasks. The order of tasks was balanced across subjects. Subjects held a stylus equipped with a pressure contact in their right hand (Figure 7). When a subject voluntarily lowered the stylus onto a pressure plate, a visual target (small circle) started moving across a CRT screen at constant speed in three successive steps of 1.5 s each; the direction of each step was randomly determined. Thereafter, the target jumped back to the center of the screen. Subjects had t o track

134

H. H. Komhuber et al.

I

Figure 8. Topographical distribution of performance-related DC-potential shifts, averaged across all subjects. The vertical line indicates the onset of the stimulus program (t = 0). Stimulus directions changed at t = 1.5 and t = 3 s, with the end of the stimulus program at t = 4.5 s. Upper row: Inverted Tracking (IT). Lower row: Tracking (T). Negativity up. The volitional initiation of stimulus onset is preceded by a Bereitschaftspotential (BP); task performance is accompanied by a slow negative potential shift, the performance-related negativity. From W. Lang, Lang, Podreka, et al., 1988. the target by moving the stylus with their right hand. This experimental design provided a continuous tracking performance. The position of the moving right hand was coupled back as a light spot on the TV screen. Accuracy of tracking was determined by the difference between target and light spot. In a visuomotor learning task (Inverted tracking, IT), subjects had to track horizontal movements of the stimulus in an inverted manner, that is, movements of the target to the right required hand movements to the left and vice versa; movements up and down were not inverted. In a control task (Tracking, T), subjects had to track the target in a normal, non-inverted manner. For further details of experimental procedure see W. Lang, Lang, Podreka, et al., 1988. Performance-related DC shifts: Tasks were performed four months after the rCBF measurements. It was assumed that the temporal delay

Cerebral Correlates of Volition and Attention

umns above the horizontal line

c3

135

cz

would reduce memory effects and would therefore provide comparability between the two tasks. In order to avoid familiarity, the IT-task was modified in the brain potential study: now vertical movements of the target had to be tracked in an inverted manner, whereas left and right remained unchanged. After the 4.5 s of tracking, subjects moved the stylus to the starting position and had a resting period before starting the task again. Inter-trial intervals ranged between 8 to 12 s. Performance-related DC shifts are shown in Figure 8. The volitional initiation of the stimulus program was preceded by a Bereitschaflspatential (BP); visuomotor performance was associated with a slow negative DC-potential shift with larger amplitudes in IT as compared to T. Differences of amplitudes (Nd) had a clear fronto-central distribution (Figure 9) and were significant in frontal recordings, C3 and Cz. In these recordings, Nd was correlated with the success of a subject’s visuomotor learning; the coefficients of correlation, r, ranged between .6 and .8. The electrophysiological findings of previous experiments (W. Lang et al., 1983, 1986) were replicated. The conclusion of the DC-potential study, that frontal lobes are critically involved in visuomotor learning, was confirmed by the results of the SPECT study: In IT, as compared to T, there was an increased relative tracer uptake in frontal areas (in particular middle frontal gyrus of both sides, and fronto-mesial cortex) as compared to T dorsolateral parts. In addition, the SPECT study showed

136

H. H. Kornhuber et al.

increased relative tracer uptake in basal ganglia and cerebellum (see color Figure 15 on page 152). Performance-related DC shifts and EEG spectra were analyzed in another experiment on visuomotor learning (M. Lang, Lang, Diekmann, & Kornhuber 1987). In agreement with findings reported above, performance-related, negative DC shifts in frontal recordings were larger in IT as compared to T. In visuomotor learning, e-MPD (mean power density of the e frequency band) was significantly larger during task performance than it was in the resting state. This IN-e-MPD (increase of e-MPD, see 2.2) was not present in T. In frontal recordings, differences of IN-e-MPD between IT and T were larger in good learners than in those who performed less efficiently. But differences were significant for the whole group of subjects (M. Lang, Lang, Diekmann, & Kornhuber 1987). In another visuomotor learning task (W. Lang et al., 1986) the feedback signal to the TV screen was distorted by imposing a sine wave when the subject started to track the target (DT; distorted tracking). This manipulation created a rather complex distortion which had to be compensated during tracking. In contrast to the other experiments, a simple cognitive strategy such as the inversion of horizontal and/or vertical direction could not be used. Subjects were still able to significantly reduce the error of tracking although they could not verbalize, and were not consciously aware of, the strategy they used. Amplitudes of negative DC shift were larger in DT as compared to a simple tracking control (T). This difference (Nd) was again significant in frontal recordings and correlated with the success of learning in Fz, FCz and F4 (r ranging between .5 and .6). A correlation between cortical negative shifts and success in learning could not be found in F3 (as it had been in the inverted learning tasks; see Figure 10). In another experiment, the question was raised whether there are differences between mentally and actually performing a visuomotor learning task using inversion of horizontal directions (Lang, Uhl, Koska, Lindinger, & Deecke unpublished data). The following significant differences were found: (a) When actually tracking, cortical DC shifts had their maxima in recordings from the fronto-central midline, whereas in the imagery situation, maxima were found parietally. (b) In general, amplitudes of performance-related DC shifts were larger during actual tracking as compared to mere imagining. (c) In the imagery task, DC shifts were significantly lateralized with the larger amplitudes occurring over the left hemisphere (in frontal, central and parietal recordings). The interpretation of this lateralization was ambiguous since it could be

Cerebral Correlates of Volition and Attention

137

I

A

Figure 10. Topography of the correlation of subjects’ Nd (the difference, i.e., extra negativity, between the amplitudes of the performance-related negativity during the visuomotor learning task and the visuomotor control task) with the success of subjects’ visuomotor learning. The diameters of the circles are proportional to the correlation coefficients, r, for the respective electrodes. Left side: correlations in an inverted-tracking task (inversions of horizontal directions). Right side: correlations in a distorted-tracking task. due to the fact that subjects imagined acting with the contralateral right hand. Therefore, in a control experiment, 12 subjects were instructed to track the target with their left hand either mentally or actually. The lateralization of negativity towards the left hemisphere remained only in the imagery situation and was significant in frontal recordings. In conclusion, the frontal lobes are critically involved in visuomotor learning tasks. Such tasks require subjects to rely on an inner representation of the goal, to maintain it against interference of other thoughts and to develop adequate response patterns in order to achieve the goal. When developing these new stimulus-response patterns, interference from old associations has to be overcome. Possibilities of using cognitive

138

H. H. Komhuber et al.

concepts and predictive strategies varied between the tasks: Visuomotor learning involving horizontal and/or vertical inversions enabled subjects to use a verbal strategy in their prediction of trajectories, whereas learning in the distorted tracking situation was mainly based on feedback recognition and utilization (which could not be verbalized and experienced). This is why distorted tracking was associated with right frontal activation, whereas inverted tracking caused a more bilateral frontal activation. When imagining tracking in an inverted manner without actually moving, the left frontal lobe, but not the right one, became activated (in this context see section 3.5). Patients with frontal lesions become distractable (Luria, 1966). They still have long term goals but have problems in maintaining them. Susceptibility to interfering Stimuli has often been described; instead of continuing an action they tend to compulsively utilize presented objects or to imitate actions of the examiner (Lhermitte, Pillon, & Serdaru, 1986a,b). These patients have a loss of active and intentive elements in their behavior which are essential for pursuing prospective goals (Kleist, 1934; Fuster, 1980). Similar disorders have been found in animals after lesions of the prefrontal cortex: Those of the prearcuate cortex disturb performance in tasks that require a choice of action depending on the special attribute (including temporal or spatial discontinuities) of a recent stimulus (for review see Brody & Pribram, 1978; Fuster, 1980; H. H. Kornhuber, 1987). The importance of the frontal cortex for flexible, goaldirected behavior is also supported by anatomical data demonstrating a convergence in the frontal cortex of sensory inputs from the outer world, via the sensory projection and association areas of the posterior cortex, and of motivational impulses from the limbic system and the hypothalamus (Nauta, 1972; Kievit & Kuypers, 1975). Negative DC-potential shifts and Tc-99m-HMPAO uptake of the frontal cortex during visuomotor learning may indicate cortical activation. Increased performance-related e - W D may indicate activation of the frontal limbic system.

3.4. Cognitive information processing Verbal cognitive learning in associative learning or concept formation tasks has consistently been found to be associated with negative DC-potential shifts in left frontal recordings, indicating an activation of this area (M. Lang, Lang, Uhl, et al., 1987; W. Lang, Lang, Uhl, Kornhuber, et al., 1988; Uhl, Lang, Lindinger, & Deecke, 1989; Uhl, Franzen, et al., 1989). This left frontal activation remains rather constant when varying material (nonsense syllables, meaningful words) and learning

Cerebral Correlates of Volition and Attention

139

strategies (creating a meaningful sentence with two items, imagery; W. Lang, Lang, Uhl, Kornhuber et al., 1988; Uhl, Lang, et al., 1989). When presenting pairs of words with pre-established, habitual associations (control tasks), the left frontal activation is not present. In a recent experiment two different conditions of verbal cognitive learning were investigated (Uhl, Franzen, et al., 1989): In one condition each trial contained a new pair of words that had to be associated by creating a meaningful sentence. In the second condition 8 pairs of words had to be associated in a first run. The following 7 runs (each run consisting of 8 trials) used the same words but each time in a different combination. This design caused proactive interference; that is, subjects had to prevent interference by prior habits when learning new ones. In both conditions, the negative DC-potential shift in left frontal recordings appeared. However, in the task with proactive interference an additional negativity occurred, particularly in frontopolar recordings (Fpl and Fp2; overlying the rostra1 limit of the superior frontal gyrus). During concept formation and associative learning, subjects perform creative operations by using cognitive strategies. It is interesting to note that associative learning with imagery techniques causes activation of the left fronto-lateral cortex, whereas imagery of non-verbal materials (colors, maps or faces), without the necessity of transforming, processing, or encoding the items, causes an activation not of the frontal lobes but of the occipito-parieto-temporal regions, with a significant lateralization towards the left hemisphere (Uhl, Goldenberg, et al., 1989). EEG spectra have been calculated in the concept formation task and in one of the associative learning experiments (M. Lang, Lang, Diekmann, & Kornhuber 1987, W. Lang, Lang, Kornhuber, et al., 1988). In these experiments, IN-e-MPD was significantly larger in frontal recordings (F3, FCz) during cognitive learning than in the control task. The frontal e-rhythm differed not only between learning task and control but also between successful learners and those who performed less efficiently. The a-rhythm did not differ between tasks or groups of subjects. In frontal recordings, the difference of IN-e-MPD between learning task and control was correlated with the difference of the performance-related negativity, Nd, between the two tasks. In summary, the left frontal lobe, and the limbic and paralimbic systems are activated when subjects perform cognitive learning tasks in which verbal material has to be processed. Performance is guided by a goal, namely, to learn the associations in order to perform efficiently in subsequent retrieval tasks. Direction of behavior toward a future goal

140

H. H. Komhuber et al.

requires drives, motives and will. Parallels to the visuomotor learning tasks are obvious and, again, we found similar physiological signs, increased frontal negativity and e-MPD. Measurements of the regional cerebral blood flow have also given evidence for an activation of the frontal cortex in verbal cognitive learning (Goldenberg et al., 1987; W. Lang, Lang, Goldenberg, Podreka, & Deecke, 1987; Maximilian, Prohovnik, Risberg, & Hakansson, 1978).

3.5. Directing attention toward forthcoming, relevant, sensory events 3.5.1. Directed attention potential (DAP). In this section evidence will be given that the parietal lobe is involved in the anticipatory process of directing attention towards stimuli having relevance for a forthcoming action. In the following experiments (for details see Deecke et al., 1984 and W. Lang et al., 1984) cortical DC shifts in self-initiated, sensory-guided tracking movements were investigated. Subjects fixed their gaze on a fixation point straight ahead. When lowering a stylus with their right hand onto a plate, a light spot in the left hemi-field of vision started moving for 1 s in a first random direction, and subsequently for 1 s in another random direction. Thus, sensory information was primarily projected to the sensory areas of the right hemisphere. Predominant events such as the start of a stimulus program and the change of stimulus direction were predictable in time, but unpredictable in direction. Subjects had to track the light spot that moved at a constant speed. Similarly, tactile stimuli were applied to the subject’s left palm by a modified XY-plotter and had to be tracked in a similar manner. Figure 11 compares corresponding recordings of the two hemispheres in the visual and the tactile task. As can be seen, the volitional initiation of the stimulus program at t = 0 was preceded by a BereitschafspotentiaE. Change of stimulus direction (t = 1 s) and the end of the stimulus program (t = 2 s) were preceded by slow negative potential shifts of an expectancy wave (CNV) kind. The BP contained a very striking feature: As predictable, in unilateral (right-sided) self-paced movements, BP had larger amplitudes over the contralateral (left-sided) primary motor cortex (C3) as compared to the ipsilateral one (C4). But in parietal recordings, an opposite lateralization effect occurred; that is, BP amplitudes were larger in P4 than in P3. In studies of unilaterally performed rapid finger movements, such an ipsilateral preponderance of the BP has never been observed. But the situation in the present experiment was different, because either a visual or a tactile stimulus of

Cerebral Correlates of Volition and Attention

a)

141

TRACKING VISUAL

Figure 11. Hemispheric differences for the visual tracking experiment. Grand means averaged across 16 subjects. Dotted lines indicate recordings from the left, solid lines indicate recordings from the right hemisphere. Intervals with significant differences between single pairs of data points are marked (two-tailed t tests, p c .05). Modified, from Lang W et al., 1984.

task-relevance was triggered by the initiating movement, and appeared in the left hemi-field of vision or the left palm. The additional activation of the right parietal cortex can thus be interpreted as sign of directing attention towards a forthcoming task-relevant sensory cue which was predictable in time. It has, therefore, been called "directed attention potential, DAP" (Deecke et al., 1984; Kornhuber,l984a; W. Lang et al., 1984). In an additional control experiment naive subjects were instructed to initiate the same stimulus program but without tracking the target. Thus, movement initiation with the right hand was again associated with the occurrence of a visual target on the left side, but now the target had

142

H. H. Kornhuber et al.

no behavioral relevance for the subject; a directed attention potential was absent. Therefore, directed attention can be characterized as an active and selective process that is based on the goals of an action. We propose that intentionality in actions includes not only representations of desired goals and motor planning but also the anticipation of behavior-relevant sensory events. To the extent that (a) unanticipated events are startling experiences for us, and (b) we are startled only infrequently, it appears that we actually anticipate virtually every voluntary act and its related sensory messages, at times without being conscious of it. In another experiment, the effect of pre-information on the parietal directed attention potential was tested (M. Lang, Lang, Kornhuber, Bunz, & Kornhuber, in press). Pre-information was given by a signal, S1. Subjects had to track a visual target that moved for 3 s, starting at signal S2, 3.5 s after S1. Eight experimental conditions were given in random order (A,B,C,D,E,F,G,H). A,C,E,G conditions required normal right handed tracking of a polygon. A warning stimulus Sw was given 0.5 s prior to S2. In the B,D,F,H conditions, subjects had to track the visual target in an inverted manner (horizontal and vertical inversion of subject’s hand tracking coupled back to the screen). In the sequence of conditions A,C,E,G (normal tracking, T) and B,D,F,H (inverted tracking, IT) the amount of pre-information provided by S1 increased; that is, A and B included no pre-information, whereas G and H contained maximum preinformation (presentation of the target and the information about normawinverted tracking). These two parameters, pre-information and tracking mode were treated as factors for statistical analysis. ANOVA revealed a significant influence of the pre-information factor on the negative amplitude prior to Sw at parietal recordings and on the difference between left and right hemisphere. Negative DC shifts increased slowly towards the presentation of S2. Amplitudes increased with increasing amounts of pre-information delivered by S1. Lateralization towards the left parietal (contralateral to the moving right hand) electrode was found, although the stimulus was projected in the subject’s central field of vision. In the tracking epoch, the frontal e-rhythm was significantly larger in the IT than in the T conditions. In the epoch between S1 and Sw, the a-rhythm in parietal and occipital recordings was attenuated, with this attenuation decreasing with increasing pre-information. Changes of the e-MPD in the tracking epoch were consistent with previous studies (section 3.3): IN-e-MPD was larger in IT than in T. Performance-related a-attenuation and DC-potential shifts are considered to be signs of

Cerebral Correlates of Volition and Attention

143

cortical activation. However, in the epoch between S1 and Sw, "preinformation'' had contrary effects on these two parameters: Amplitudes of DC shifts in parieto-occipital recordings increased with increasing preinformation, whereas a-attenuation decreased in the same recordings. DC-potential data may indicate that the process of directing attention towards S2, increases with increasing pre-information delivered by S l , which renders the possibility of information processing in anticipation. a-attenuation may be more susceptible to arousal: In a situation of uncertainty, as created by little pre-information, arousal may be larger than in situations with more pre-information. In their animal studies, Mountcastle, Lynch, Georgopoulos, Sakata, and Acuna (1975) and Lynch (1981) showed that neurons in the parietal lobe become active when monkeys detect, look at or reach towards a motivationally relevant object or event within their extra-personal space. Input from the limbic system to the parietal lobe has been demonstrated by Mesulam (1983). Mountcastle et al. (1975) propose that "several abnormalities of function that occur in humans and monkeys after lesions of the parietal lobe can be understood as deficits of volition, of the will to explore with hand and eye the contralateral half-field of space, a deficit caused by the loss of the command operations which exist in the parietal association cortex" (p. 905). Clinical symptoms of parietal lesions have been described since B a h t (1909) and are called neglect to the contralatera1 half-field of space (Brain, 1941; Semmes, Weinstein, Genth, & Teuber, 1963). 3.5.2. Anticipation of the right moment to act. In the sensory-guided tracking task described above (see 3.5.1), the two dominant events, the onsets of the first and the second trajectories of the target at t = 0 s and t = 1 s, respectively, are predictable in time. The right moment to start the tracking movement is not selected by the subject but is given by external cues. This gives rise to the following considerations: The frontomedial cortex including the SMA has been described as playing a responsible role when subjects have to select and start a movement on their own (see section 3.1). Is this process "switched off' if no longer needed? Figure 12 demonstrates that fronto-mesial cortex and sensory areas have different temporal patterns of cortical activation: In FCz (over the mesial, fronto-central cortex including the SMA), the Bereitschaftspotential decreases about 150 ms before movement onset (PMP) whereas recordings from the occipital right cortex, 0 2 , show a sustained negativity which declines only 200 ms after the onset of movement. Even more striking is

144

H. H. Komhuber et al.

v i s . stim. midline, mOl. suppl. a r e a

Figure 12. Recordings in FCz (midline, supplementary motor area) and 0 2 (occipital right) are compared. Grand means averaged across 16 subjects. Dotted lines bounding the average indicate 2 1 standard error of the mean. Onset of voluntarily initiated stimulus occurred at t = 0 s, change of stimulus direction at t = 1 s, end of the stimulus program at t = 2s.

@

DCCIP~IOI

&

~~

-I

0

I

2

sec

the difference at the onset of the second trajectory. Now, the preceding negative potential shift in FCz ends already about 300 ms before this event, whereas in the sensory areas the negative potential shift continues for 200 ms after the onset of the trajectory. It is to be concluded that the sensory cortical areas remain activated in order to attend to and analyze the trajectories of the targets that have to be tracked. Primary sensory and association areas (including the parietal lobe) may now trigger the tracking performance themselves. Fronto-medial structures show an early decline of surface-negativity, probably delegating the function of movement initiation to the posterior primary sensory and association areas (H. H. Kornhuber, 1984a).

3.6. Volition in the sense of setting priorities In volitional actions, a stimulus or a drive is not immediately transferred into action. Rather, there is a moment of consideration before acting in which the present and forthcoming situations are evaluated in the context of duties, long term goals, etc. These reflections

Cerebral Correlates of Volition and Attention

145

are associated with decisions about what to do, and are fundamental for the experience of volition in our actions. Any direct response to an internal (drive) or external stimulus is inhibited. If certain motives and drives become overwhelming, selectivity of behavior and volition in action is reduced, as in patients suffering from obsessive and compulsive disorders. These patients experience and complain that their behavior is not under control of their will any more. Lesions of the orbito-frontal cortex lead to disinhibition of primitive drives and to moral depravation as originally described by Kleist (1934; for review see Fuster, 1980; Stuss & Benson, 1986; H. H. Kornhuber, 1987). The self-perception regarding their behavioral disturbance is obviously reduced in these patients. An experiment has been performed in which subjects were asked to withstand self-administered aversive stimulation (pain and fatigue), by pressing on a button for as long as possible (A. Kornhuber, Lang, Lang, Kure, & Kornhuber, in press). Painful electrical stimuli with intensities of 25% or 75% of the subjective maximum pain level were applied either to the left or the right index finger. Pain was applied by self-paced button pressing, and disappeared when the index finger was lifted from the button. In another set of 4 conditions the same subjects had to press down on the button with 25% or 75% of a predetermined maximum of force, either with the left or the right index finger. Payment depended on the level of task performance, that is, how long subjects were able to tolerate the pain or the fatigue of pressing the button. This experimental design led to a conflict of motives: a conflict between a desire to earn wages (by withstanding pain or fatigue) on the one hand, and a desire to escape or relax on the other. The following main results were found when analyzing the EEG spectrum: During task performance frontal and central medial e-MPD was larger than during the preceding resting state (IN-e-MPD). When subjects stopped the performance, e-MPD decreased again (DE-e-MPD). Task-induced IN-e-MPD and relaxation-induced DE-e-MPD were significantly larger in the conditions with higher loads (with pain and force at the 75% level) as compared to those with lower ones. IN-e-MPD and DE-e-MPD (Figure 13) had a clear frontal distribution with a maximum in recordings from the fronto-central midline. Will is needed to select between motives and drives. Selection and maintenance of this choice during the action are important in these tasks. The choice in the present experiments involved the decision what to do and not (as in previous studies on visuomotor and cognitive learning) on the decision how to do. In spite of these differences one physiological parameter remained invariant: The frontomedial e-MPD increases with

146

H. H. Komhuber et al.

increasing difficulty of choice of what or how to do. We suggest that all deliberately and consciously-controlled decisions on aspects of voluntary actions require motivational input and will. The frontal e-rhythm may reflect input from the limbic system to the frontal convexity.

4. General discussion

4.1. The utility of a centralized startingfinction for movements There is an important feature in voluntary movements as opposed to voluntary thoughts: the movements interfere with the external world and might endanger the safe posture of the body in the field of gravity. Motion interferes with other living beings and is a potential signal for enemies. Therefore, the initiation of a voluntary movement needs more control than the initiation of a thought. Before a voluntary action is started, the position of body and limbs, as well as ongoing movements and signals from the environment, need to be considered. That is, signals from all sensory and motor systems and those from the motivation system have to be coordinated, in order to select the best starting time. The numerous afferents to the SMA are in accord with such a function. The SMA receives information from the precentral motor cortex, from the somatosensory and parietal areas, from the cerebellum and basal ganglia via thalamic nuclei, from the claustrum, and from various stages of the motivational system such as the hypothalamus (via the mediodorsal thalamic nucleus), the amygdala, the prefrontal cortex, etc. (Jones, 1983; Wiesendanger & Wiesendanger, 1985). The SMA is active prior to all kinds of voluntary movements including those which are not represented in the precentral motor cortex such as eye movements (Deecke et al., 1969; Deecke et al. 1976, Deecke & Kornhuber 1978, Becker et al., 1972, Grozinger et al. 1979). Regional cerebral blood flow measurements (Roland et al., 1980) are in agreement with our conclusion, although the time resolution of this method by itself is insufficient to even distinguish between changes before and after movement onset. The principle of centralization as realized in the SMA is particularly interesting in view of the fact that the motor system is much more decentralized in the brain than our present textbooks are advertising. For instance, the motor cortex for speech is in the temporal lobe (H. H. Kornhuber, 1984b). The same tongue movements which during chewing are controlled by the motor cortex and its tactile afferents (the motor cortex having features of a somatosensory association area; H. H. Kornhuber, 1974) need an entirely different control during speech production in which the upper

Cerebral Correlates of Volition and Attention

Pain

147

Force

Figure 13. Topography of the difference between 75% and 25% of maximum tolerance in the "pain" and the "force" tasks (cf. section 3.6) in terms of the relative decrease of Theta mean power density (9-MPD) due to relaxation. There is a larger relative decrease over the frontal areas. temporal lobe and its auditory afferents are taking part. The relationship of the SMA with the motor cortex and the basal ganglia are in agreement with its proposed starting function. Neurons of the pallidum (Melnick et al., 1984) often start to discharge several hundred milliseconds prior to the onset of a voluntary movement, similar to the early beginning of the Bereitschuftspotential (H. H. Kornhuber & Deecke, 1965; Deecke et al., 1969, 1976; Deecke, Kornhuber, Lang, Lang, and Schreiber, 1985; Deecke and Kornhuber, 1977, 1978; W. Lang, Lang, Uhl, Koska, et al., 1988; W. Lang, Zilch, et al. 1989). Both the basal ganglia and the cerebellum are necessary for the preparation of an aimed voluntary movement (H. H. Kornhuber, 1974; Lamarre et al., 1980). The coordination of serial movements is just a special case of starting movements at the right moment (timing). As our data show, the SMA is likewise involved in single as in serial voluntary movements (H. H. Kornhuber & Deecke, 1965; Deecke et al. 1976; Deecke and Kornhuber, 1978; M. Lang et al., 1988; W. Lang, Lang, Uhl, Koska, et al., 1988; W. Lang, Zilch, et al., 1989).

148

H. H. Kornhuber et al.

4.2. Frontal lobe and volition The first theory of voluntary movements was Liepmann's explanation of apraxia (Liepmann, 1900). In his theory he postulated an "imagination" (Vorstellung) between sensation and motion, but no volition. According to Liepmann, the posterior cerebral lobes and the motor cortex were involved, but not the frontal lobe, Since Walter Rudolf Hess found mechanisms of different drives, such as hunger, thirst, aggression, etc., within the diencephalon and limbic system (Hess, 1949), motivation was usually thought to be localized in subcortical systems, whereas the cortex was considered to supply merely cognitive information. The views of Feuchtwanger's (1923) and Kleist's (1934), who treated frontal lobe symptoms in terms of a psychology of will, were usually neglected by most psychologists. Even Teuber (1964), who knew the old German literature, did not seriously consider this tradition. His interpretation of frontal lobe function restricted itself to the concept of corollary discharge (i.e., feedback information from motor to sensory systems), without realizing the fundamental importance of volition in human motor action and behavior. In the psychological abstracts, will and volition had disappeared by 1965. Attempts to compensate for this deficit came from neurophysiology (H. H. Kornhuber, 1984a,c, 1987, 1988). It was Karl Kleist (1918, 1934), who distinguished between two different frontal-lobe syndromes in man: The syndrome of the lateral convexity of the frontal lobe, on the one hand, characterized by a reduction of spontaneity, indifference and apathy, and the syndrome of the orbito-frontal cortex, on the other hand, consisting of lack of endurance, deficits in moral behavior and disinhibition of drives. In our opinion, the inability of the patient with frontolateral lesions to develop new sorting strategies in the Wisconsin card sorting test as described by Brenda Milner (1963, 1964) and confirmed by us (Bechinger, Kornhuber, Jung, & Sauer, 1986) agrees well with Karl Kleist's statement. As was pointed out earlier (H. H. Kornhuber, 1973), the deficits in the delayed response task of monkeys having fronto-lateral lesions (Jacobsen, 1935, 1936; Jacobsen & Nissen, 1937) is also in good agreement with this dimension of frontal lobe function: There has to be a selection of the important events during the transfer from short term to long term memory prior to long term storage, and the motivation system is crucially involved in this evaluation. A classification of volitional functions into three stages has been proposed by Kornhuber (1984a, 1987): The first stage involves setting priorities among the needs, that is to say, to determine what shall be done

Cerebral Correlates of Volition and Attention

149

(what to do). The second stage involves developing abilities, making plans and decisions about the way to do it (how to do), and the third involves the selection of the right moment when the action is to be started (when to do). As Karl Kleist (1934) had concluded earlier, the orbital cortex seems to be important for the first stage: The single unit data of Rolls (1983) and Thorpe, Rolls, and Maddison (1983) agree well with Kleist’s concept. For the second stage, the fronto-lateral cortex seems to be important. It is only the cortex of the frontal lobe convexity that exhibits significant correlations between an increase in the cortical DC potential and subjects’ success in learning a new performance, as demonstrated by human learning experiments (W. Lang et al., 1983, 1986, W. Lang, Lang, Podreka, et al., 1988). Furthermore, the observations of Kleist (1934), Jacobson (1935), Freeman and Watts (1942), Milner (1963, 1964), and Fuster (1973) are compatible with this interpretation. Additionally, the fact that the monoaminergic neuronal systems project mainly to the frontal lobe is in line with the lobe’s putative role in the second-stage process determining how to do (Lindvall & Bjorklund, 1983). For the third stage, mesial cortical areas of the frontal lobe, more specifically the SMA, seem to play the main role as pointed out at the beginning of this discussion. The function of acting at the right moment had previously escaped the attention of scientific research. That is, the right time for action (which is so important in education, sports, hunting, therapy, policy, business etc.) became a subject of neurophysiological interest only recently (Kornhuber, 1984a). This was probably because psychologists and philosophers had regarded psychological time as being a personal experience which was either too subjective or too abstract to afford the possibility of scientific investigation. Of course, the motionand-time studies of industrial psychology and engineering were realistic, but they were tied into the stimulus-response paradigm of machine-maninteractions. The problem of the right time, in our context, becomes sharper when the right time is not predetermined by a machine (or where the question is just whether a man is able to respond quickly enough), but, rather, when the individual is free to choose the right moment to start. The ancient Greeks intuitively knew the importance of the right moment; they had a word for it, Kairos. (Kairos, the youngest son of Zeus, was also the god of the right measure, in the sense of moderation.) The data on theta-enhancement over near-midline cortex under conditions of effort generally agree with the enhancement of the cortical DC potential. There is a significant correlation of both, theta-power and

150

H. H. Komhuber et al.

negative DC-potential shifts in the learning of a performance (M. Lang, Lang, Diekmann, & Kornhuber 1987; W. Lang, Lang, Kornhuber, et al., 1988). On the other hand, the relationship between the DC-potential shift and attenuation of the alpha-rhythm is less clear.

4.3. Attention, the parietal lobe and the directed attention potential At the same time that will and volition fell into disrepute and disappeared from psychology, attention gained a favorable reputation. Attention has a function in perception that is similar to the one volition has in action, but it was easier to think of attention as being stimulusdependent; the thought of volition appeared to imply freedom, which was a banished idea. In contrast to the importance of the frontal lobe for voluntary action, the maximum attention potential is over the parietal lobe, contralateral to the side of the visual or tactile stimuli (H. H. Kornhuber, 1984a; W. Lang et al., 1984). This is in agreement with the convergence of the sensory afferents from different senses (vision, audition and touch) in the posterior parietal area (Jones & Powell, 1970; H. H. Kornhuber, 1983), with clinical findings of disturbed attention (syndrome of neglect) following lesions of the parietal lobe (Brain, 1941; Semmes et al., 1963) and, most of all, with the single-unit (monkey) data of Mountcastle and his group (Mountcastle, Andersen, & Motter, 1983; Mountcastle, 1979, 1989).

4.4. Will and fieedorn Let us, finally, look at the reason why scientists repressed the concept of the will. It was probably because will belongs to freedom in the sense of free will and good will. Many scientists of today consider the idea of freedom an illusion. However, we must realize that there are two kinds of freedom: (a) freedom porn something (independence), and (b) freedom to or for something (ability, performance). This was clearly expressed by Nietzsche 1883; the roots of this distinction came from Nicolaus Cusanus, Pic0 della Mirandola, Leibniz, and Kant. Unfortunately, this distinction had little impact on the mentality of our civilization, for the causal connection of the two kinds of freedom was not considered -- perhaps because of the prejudice that freedom should not have a cause. Freedom from, however, is invariably based on freedom to (H. H. Kornhuber, 1984~). For instance, freedom from robbery is based on the ability of the state to maintain order, on parents’ ability to educate their children, etc.; freedom from illusion is based on the ability to reason, on the ability of the brain to function normally, etc. This positive freedom,

Cerebral Correlates of Volition and Attention

151

freedom to or for, is a relative freedom; it is not contrary to nature. Many abilities contribute to it, but will has a central position among them. This relative freedom of will is reflected in the various ways we maintain our psychic balance, for example, avoiding dehydration by drinking water, avoiding exhaustion by sleeping, avoiding disinhibition by abstaining from alcohol, etc. When we become ill, we may try to maintain our freedom by seeking help from physicians. For example, physicians are obliged to prevent cretinism by early diagnosis of hypothyroidism in newborn infants and by treatment with thyroxine. Parents, teachers, psychologists, and doctors take on the task of helping others to become more free; indeed, that task is a challenge we all face. As Aristotle asserted in his Nicomachean Ethics, "virtues arise in us neither by nature nor contrary to nature; but by our nature we can receive them and perfect them by habituation ...We become just by doing what is just, temperate by doing what is temperate, brave by doing brave deeds" (Apostle, 1984, p. 21). What Aristotle called virtue is about the same as what we call positive freedom, or freedom for. Chance events in our brain may contribute to freedom via phantasy when creativity matters. Chance events alone, however, do not constitute freedom. It is by higher mental functions, by intelligence, reasoning, conscience, authenticity, by learning, practice, creativity, constructive cooperation and training of will that we become more free (H. H. Kornhuber, 1984c, 1987, 1988). If we compare the brains of different animals, it is obvious that the older parts of the motivation system, the hypothalamus and the limbic system are rather conservative. What makes the difference between a chimpanzee and a rat, and again between homo sapiens and a chimpanzee, is the larger development of the cortical association areas, almost half of which (the frontal, orbital, and anterior medial cortex) subserve volitional functions. This corresponds to the importance of consideration, planning, reasoning and associated volitional functions involved in responsible human behavior. While hedonism tries to make the whole brain serve a minor diencephalic function, a more humane philosophy (which reminds us that our reasonable will has higher goals, goals beyond our ego) is obviously in better agreement with human brain physiology. Human cerebral anatomy and physiology also leave open vast possibilities for the development of different modes of conduct, through cultural evolution, education, and learning. The more mind, the more good will is necessary to make mind effective for constructive and responsible behavior. In this endless task of mental and volitional development, more than intelligence matters, for mankind is always in

152

H. H. Komhuber et al.

Figure 14, upper photograph on facing page: (see Experiment I of section 3.1.7.) Comparison between SE-S and SI-S. Differences of amplitudes have been calculated for the Bereifschafspotentiual (Nd-BP, left side) and the performance-related negativity (Nd-P, right side). Based on Nd-BP and Nd-P, topographical distributions of current density at the scalp surface were calculated and displayed. Current flowing into the scalp is indicated in colors of yellow and red, current flowing out of the scalp in colors of green and blue. Note, current flowing into the scalp in the area of the central midline (overlying the SMA) indicating increased negativity of the mesial central cortex in SE-S as compared to SI-S.

Figure 15, lower photograph on facing page. (Compare with Figure 9 of secion 3.3.) Tc-99m-HMPAO brain SPECT of one subject. Four axial slices from cranial (left) to caudal (right). Upper row: Inverted Tracking (IT). Middle row: Tracking (T). Lower row: subtraction (IT minus T) after normalization of count rates. The relative tracer distribution is displayed in colors ranging from blue (low) to white (high concentration). In the lower row, colors of red and white display areas having a higher relative tracer concentration in IT as compared to T. This is particularly true in the mesial, fronto-central cortex, the dorso-lateral cortex of both hemispheres (mainly congruent with the middle frontal gyrus), basal ganglia and cerebellum. From W. Lang, Lang, Podreka, et al., 1988.

Cerebral Correlates of Volition and Attention

153

154

H. H. Komhuber et al.

danger of drifting away from those ideals that hold the greatest promise for humanity. Because of human creativity, man needs moral education by educated persons.

Cerebral Correlates of Volition and Attention

155

References Ahern, G. L., & Schwartz, G. E. (1985). Differential lateralization for positive and negative emotion in the human brain: EEG spectral analysis. Neuropsychologia, 23, 745-755. Apostle, H. G. (1984). Aristotle's Nicomachean Ethics . Grinnell, Iowa: The Peripatetic Press. h o l d s , D. E., Lopes da Silva F. H., Aitinik J. W., Kamp A., & Boeijinga, P. (1980). The spectral properties of hippocampal EEG related to behavior in man. In G. Pfurtscheller, P. Buser, & F. H. Lopes da Silva (Eds.), Rhythmic EEG activities and cortical functioning (p. 91-102). Amsterdam: Elsevier. Aschoff, J. C., & Kornhuber, H. H. (1975). Functional interpretation of somatic afferents in cerebellum, basal ganglia, and motor cortex. In H. H. Kornhuber (Ed.), The somatosensory system (pp. 145-157). Stuttgart: Georg-Thieme-Publishers. Balint, R. (1909). Seelenlahmung des "Schauens", optische Ataxie, raumliche storung der Aufmerksamkeit. Monatsschrifi fiir Psychiatrie und Neurologie, 25, 5 1-81. Bauer, H., & Rebert, C. (in press), Preliminary study on subcortical slow potentials related to the readiness potential in monkey. International Journal of Psychophysiology. Bechinger, D., Kornhuber, H. H., Jung, H., 8z Sauer, E. (1986). Higher density of mental capacities in the left cerebral hemisphere of man: A quantitative investigation in children with localized cerebral lesions. Archives Italiennes de Biologie, 124, 83-93. Becker, W., Hoehne, O., Iwase, K., & Kornhuber, H. H. (1972). Bereitschaftspotential, pramotorische Positivierung und andere Hirnpotentiale bei sakkadischen Augenbewegungen. Vision Research, 12, 421-436. Benecke, R., Dick, J. P. R., Rothwell, J. C., Day, B. D., & Marsden, C. D. (1985). Increase of the Bereitschafispotential in simultaneous and sequential movements. Neuroscience Letters, 62, 347-352. Beringer, K. (1944). Antriebsschwund mit erhaltener Fremdanregbarkeit bei beidseitiger frontaler Markschadigung. Zeitschrifi f i r Neurologie, 176, 10-30. Bernstein, N. (1984). Trends in physiology and their relation to cybernetics. In H. T. A. Whiting (Ed.), Human motor actions: Bemtein reassessed (pp. 531-544). Amsterdam: North-Holland,, Boschert, J., & Deecke, L. (1986). Cerebral potentials preceding voluntary toe, knee and hip movements and their vectors in human precentral gyrus. Brain Research, 376, 175-179.

156

H. H. Komhuber et al.

Boschert, J., Hink R. F., & Deecke, L. (1983). Finger movement versus toe movement-related potentials: Further evidence for supplementary motor area (SMA) participation prior to voluntary action. Human Neurobiology, 52, 73-80. Brain, W. R. (1941). Visual disorientation with special reference to lesions of the right cerebral hemisphere. Brain, 64, 244-272. Brinkman, C., & Porter, R. (1979). Supplementary motor area of the monkey: Activity of neurons during performance of a learned motor task. Journal of Neurophysiology, 42, 681-709. Brion, S., & Jedynak, C. P. (1972). Trouble du transfer interhemispherique a propos de trois observations de tumeurs du corps calleux: Le signe de la main etrangere. Revue Neurologique, 126, 257-266. Brody, B. A., & Pribram, K. H. (1978). The role of frontal and parietal cortex in cognitive processing. Bruin, 101, 607-633. Bruce, C. J., & Goldberg, M. E. (1985). Primate frontal eye fields. I. Single neurons discharge before saccades. Journal of Neurophysiology, 53, 603-635. Brunner, R. J., Kornhuber, H. H., Seemueller, E., Suger, G., & Wallesch, C. W. (1982). Basal ganglia participation in language pathology. Brain and Language, 16, 281-299. Caspers, H., Speckmann, E. J., & Lehmenkuhler, A. (1980). Electrogenesis of cortical DC potentials. In H.H. Kornhuber & L. Deecke (Eds.), Motivation, motor and sensory processes of the brain: Electrical

potentials, behavior and clinical use. Progress in Brain Research (Bd 54) (3-16). Amsterdam: Elsevier. Cheyne, D., Kristiva, R., Lang, W., Lindinger, G., & Deecke, L. (in press). Functional organization of sensori-motor cortex as revealed by magnetoencephalography. Proceedings of the Ninth International Conference on Event-Related Potentials of the Brain, EPIC IX, June, 1989. Cheyne, D., Weinberg, H. (1989). Neuromagnetic fields accompanying

unilateral finger movements: pre-movement and movement-evoked fields. Manuscript submitted for publication. Creutzfeldt, 0. (1983). Cortex cerebri. Berlin: Springer. Crowne, D. P., Konow, A., Drake, K. J., & Pribram, K. H. (1972). Hippocampal electrical activity in the monkey during delayed alternation problems.

Electroencephalography and Clinical Neurophysiology, 33, 567-517. Crowne, D. P., & Radcliffe, D. D. (1975).Some characteristics and functional relations of the electrical activity of the primate hippocampus and a hypothesis of hippocampal function. In R. L. Isaacson & K. H. Pribram (Eds.), The hippocampus; neurophysiologyand behavior ( 185-206). New York Plenum Press.

Cerebral Correlates of Volition and Attention

157

Deecke, L., Boschert, J., Brickett, P., Weinberg, H. (1985). Magnetoencephalographic evidence for possible supplementary motor area participation in human voluntary movement. In H. Weinberg, G. Stroink & T. Katila (Eds.), Biomagnetkm: applications and theory (369-372). New York Pergamon. Deecke, L., Boschert, J., Weinberg, H., & Brickett P. (1983). Magnetic fields of the human brain (Bereitschafsmagnegeld) preceding voluntary foot and toe movements. Experimental Brain Research, 52, 81-86. Deecke, L., Engel, M., Lang, W., Kornhuber, H. H. (1986). Bereitschafspotential preceding speech after holding breath. Experimental Brain Research, 65, 219-223. Deecke, L., Grozinger, B., & Kornhuber, H. H. (1976). Voluntary finger movement in man: Cerebral potentials and theory. Biological Cybernetics, 23, 99-119. Deecke, L., Heise, B., Kornhuber, H. H., Lang, M., & Lang, W. (1984). Brain potentials associated with voluntary manual tracking. Annals New York Academy of Science, 374, 361-372. Deecke, L., & Kornhuber, H. H. (1977). Cerebral potentials and the initiation of voluntary movement. In J.E. Desmedt (Ed.), Attention, voluntary

contraction and event-related cerebral potentials. Progress in clinical neurophysiologv (Vol. I ) . Basel: Karger. Deecke, L., & Kornhuber, H. H. (1978). An electrical sign of participation of the mesial "supplementary" motor cortex in human voluntary finger movement. Brain Research, 159, 473-476. Deecke, L., Kornhuber, H. H., Lang, W., Lang, M., & Schreiber, H. (1985). Timing function of the frontal cortex in sequential motor and learning tasks. Human Neurobiology, 4, 143-154. Deecke, L., Scheid, P., & Kornhuber, H. H. (1969). Distribution of readiness potential, pre-motion positivity, and motor potential of the human cerebral cortex preceding voluntary finger movements. Experimental Brain Research, 7, 158-168. Deecke, L., Weinberg, H., & Brickett, P. (1982). Magnetic fields of the human brain accompanying voluntary movements: Bereitschafsmagne@eld. Experimental Brain Research, 48,144-148. Diekmann, V. (1985). Kurzzeit-Spektral-Analyse von ereignisbezogenen EEGAbschnitten. In H. Ganshirt, P. Berlit, & G. Haack (Eds.), Verhandlungen der Deutschen Gesellschafrfir Neurologie (881-885). Heidelberg: Springer. Dolce, G., & Waldeier, H. (1974). Spectral and multivariate analysis of EEG changes during mental activity in man. Electroencephalography and Clinical Neurophysiology, 36, 577-584.

158

H. H. Komhuber et al.

Doyle, J. C., Ornstein, R., & Galin, D. (1974). Lateral specialization of cognitive mode: II EEG frequency analysis. Psychophysiology, 11, 567-578. Eccles, J. C., & Zeier, H. (1980). Gehim und Geist. Miinchen, Zurich: Kindler. Feuchtwanger, E. (1923). Die Funktionen des Stirnhirns, ihre Pathologie und Psychologie. In 0. Forster & K. Williams (Eds.), Monographien aus Gesamtgebiete der Neurolgie und Psychiatrie, Heft 38 (4-194). Berlin: Springer-Verlag. Forster, 0. (1936). Motorische Felder und Bahnen. In 0. Bumke & 0. Forster (Eds.), Handbuch der Neurologie IV (1-448). Berlin: Springer. Fox, P. T., Fox, J. M., Raichle, M. E. & Burde, R. M. (1985). The role of the cerebral cortex in the generation of voluntary saccades: a positron emission tomographic study. Journal of Neurophysiology, 54, 384-369. Freeman, W., & Watts, J. W. (1942). Psychosurgery. Springfield, IL: Thomas. Fuster, J. M. (1973). Unit activity in prefrontal cortex during delayed-response performance: Neuronal correlates of transient memory. Journal of Neurophysiology, 36, 61-78. Fuster, J. M. (1980). The prefrontal cortex. New York: Raven. Gale, A., Christie, B., & Penfold,V. (1971). Stimulus complexity and the occipital EEG. Br. J. Psychol. 62, 527-531. Goldberg, G. (1985). Supplementary motor area structure and function: review and hypotheses. The Behavioral and Brain Sciences, 8, 567-616. Goldenberg, G., Podreka, I., Steiner, M., Suess, E., & Deecke, L. (1988). Pattern of regional cerebral blood flow related to visual and motor imagery: Results of emission computerized tomography. In M. Denis, J. Englekamp, & J. T. E. Richardsen (Eds.), Cognitive and neuropsychological approaches to mental imagey (363-373). Dordrecht: Martinus Nijhoff Publishers. Goldenberg, G., Podreka, I., Steiner, M., & Willmes, K. (1987). Patterns of regional cerebral blood flow related to memorizing of high and low imagery words--an emission computer tomography study. Neuropsychologia, 25, 473-485. Goldenberg, G., Wimmer, A., Holzner, F., & Wessely, P. (1985). Apraxia of the left limbs in a case of callosal disconnection: The contribution of medial frontal lobe damage. Cortex, 21, 135-148. Grozinger, B., Kornhuber, H. H., & Kriebel, J. (1979). Participation of mesial cortex in speech: Evidence from cerebral potentials preceding speech production in man. In 0. Creutzfeld, H. Scheich, & C. Schreiner (Eds), Hearing mechanisms and speech (189-192). Berlin Heidelberg NY: Springer-Verlag. Haslum, M. N., & Gale, A. (1973). Inter-modal and intra-subject consistency in EEG correlates of vigilance. Biological Psychology, 1, 139-150.

Cerebral Correlates of Volition and Attention

159

Hess, W. R. (1949). Das Zwischenhim. Syndrome, Lokalkationen, Funktionen. Basel: Schwabe. Hjorth, B. (1975). An on-line transformation of EEG scalp potentials into orthogonal source derivations. Electroencephalography and Clinical Neurophysiology, 39, 526-530. Homan, R. W., Herman, J., & Purdy, P. (1987). Cerebral location of international 10-20 system electrode placement. Electroencephalography and Clinical Neurophysiology, 66, 376-382. Ingvar, D. H. (1983). Serial aspects of language and speech related to prefrontal cortical activity. Human Neurobiology, 2, 177-189. Ishihara, T., & Yoshi, N. (1972). Multivariate analytic study of EEG and mental activity in juvenile delinquents. Electroencephalography and Clinical Neurophysiology, 33, 71-80. Jacobsen, C. F. (1935). Functions of the frontal association area in primates. Archives of Neurology and Psychiatry, 33, 558-569. Jacobsen, C. F. (1936). Studies of cerebral function in primates: I. The functions of the frontal association areas in monkeys. Comparative Psychology Monographs, 13, 3-60. Jacobsen, C. F., & Nissen, H. W. (1937). Studies of cerebral function in primates. IV. The effects of frontal lobe lesions on the delayed alternation habit in monkeys. Journal of Comparative and Physiological Psychology) 23, 101-112. Jasper, H. H. (1958). The ten twenty electrode system of the international federation: Appendix to the Proceedings of the General Assembly Held on the Occasion of the IVth International EEG Congress. Electroencephalography and Clinical Neurophysiology, 10, 371-375. Jonas, S. (1981). The supplementary motor region and speech emission. Journal of Communication Disorders, 14, 349-373. Jones, E. G. (1983). The nature of afferent pathways conveying short-latency inputs to primate motor cortex. In J.E. Desmedt (Ed.), Motor control mechanisms in health and disease. New York: Raven Press. Jones, E. G., & Powell, T. P. S. (1970). An anatomical study of converging sensory pathways within the cerebral cortex of the monkey. Brain, 93, 793-820. Keele, S. W. (1968). Movement control in skilled motor performance. Psychological Bulletin, 70, 387-403. Kelso, J. A. S., Putnam, C. A., & Goodman, D. (1983). On the space-time structure of human interlimb coordination. Quarterly Journal of Experimental Psychology, 35A, 347-375.

160

H. H. Komhuber et al.

Kievit, J., & Kuypers, H. G. J. M. (1975). Subcortical afferents to the frontal lobe in the rhesus monkey studied by means of retrograde horseradish peroxidase transport. Brain Research, 85, 261-266. Klapp, S. T. (1979). Doing two things at once: The role of temporal compatibility. Memoty and Cognition, 7, 375-381. Kleist, K.(1918). Die Hirnverletzungen in ihrer Bedeutung fur die Lokalisation der Hirnfunktionen. Allgemeine Zeitschrifi fiir Psychiatrie, 74, 542-553. Kleist, K. (1934). Gehimpathologie. Leipzig: Barth. Knapp, E., Schmid, H., Ganglberger, J. A., Haider, M. (1980). Cortical and subcortical potentials during goal-directed and serial goal-directed movements in humans. Progress in Brain Research, 54, 66-69. Koles, Z. J., Kasmia, A., Paranjape, R. B., & McLean, D. R. (1989). Computed radial-current topography of the brain: Patterns associated with the normal and abnormal EEG. Electroencephalographyand Clinical Neurophysiology, 72, 41-47. Kornhuber, A. W., Lang, M., Lang, W., Kure, W., & Kornhuber, H. H. (in press). Frontal Theta activity as an electrophysiological correlate of will. Proceedings of the Ninth International Conference on Event-Related Potentials of the Brain, EPIC IX, June, 1989. Kornhuber, H. H. (1961). Psychologie und Psychiatrie der Kriegsgefangenschaft. In H.W. Gruhle, R. Jung, W. Mayer-Gross & M. Muller (Eds.), Psychiatrie der Gegenwart, Forschung und Praxis, Band IZI (631-742). Berlin, Gottingen, Heidelberg: Springer. Kornhuber, H. H. (1971). Motor functions of the cerebellum and basal ganglia: The cerebello-corticalsaccadic (ballistic) clock, the cerebello-nuclear hold regulator, and the basal ganglia ramp (voluntary speed smooth movement) generator. Kybemetik, 8, 157-162. Kornhuber, H. H. (1973). Neural control of input into long-term memory: Limbic system and amnestic syndrome in man. In H.P. Zippel (Ed.), Mernoty and transfer of information. New York: Plenum. Kornhuber, H. H. (1974). Cerebral cortex, cerebellum and basal ganglia: An introduction to their motor functions. In F.O. Schmitt & F.G. Worden (Eds.), The neurosciences, third study program. Cambridge, Mass: Massachusetts Institute of Technology Press. Kornhuber, H. H. (1980). Introduction. In H. H. Kornhuber & L. Deecke (Eds.), Motivation, motor and sensory processes of the brain: Electrical potentials, behavior and clinical use. Progress in Brain Research (Vol. 54). Amsterdam: Elsevier. Kornhuber, H. H. (1983). Functional interpretation of the multimodal convergence in the central nervous system of vertebrates. In E. Horn (Ed.), Multimodal convergences in sensory systems. Fortschritt der Zoologie (Bd 28). Stuttgart: Gustav Fischer.

Cerebral Correlates of Volition and Attention

161

Kornhuber, H. H. (1984a). Attention, readiness for action, and stages of voluntary decision -- Some electrophysiological correlates in man. Experimental Brain Research, 9, 420-429. Kornhuber, H. H. (1984b). Mechanisms of voluntary movement. In W. Prinz & A.F. Sanders (Eds.), Cognition and motorprocesses (164-173). Berlin, Heidelberg: Springer-Verlag. Kornhuber, H. H. (1984~).Von der Freiheit. In M. Lincauer & A. Schopf (Eds.), Wie erkennt der Mensch die Welt? Grundlagen des Erkennens,

Fuhlens und Handelns. Geistes- und Naturwissenschaftler im Dialog. Stuttgart: Klett. Kornhuber, H. H. (1987). Handlungsentschluss, Aufmerksamkeit und Lernmotivation im Spiegel menschlicher Hirnpotentiale, mit Bemerkungen zu Wille und Freiheit. In Heckhausen (Ed.), Jenseits des Rubikon: Der Wille in den Humanwissenschaften (377-401). Berlin: Springer-Verlag. Kornhuber, H. H. (1988). The human brain: From dream and cognition to fantasy, will, conscience, and freedom. In H. J. Markowitsch (Ed.), Information processing by the brain (241-258). Bern: Hans Huber Publishers. Kornhuber, H. H., & Deecke, L. (1964). Hirnpotentialanderungen beim Menschen vor und nach Willkurbewegungen,dargestellt mit Magnetbandspeicherung und Ruckwartsanalyse. Pflugers Archiv Physiologie, 281, 52. Kornhuber, H. H., & Deecke, L. (1965). Hirnpotentialaenderungen bei Willkuerbewegungen und passiven Bewegungen des Menschen: Bereitschaftspotential und reafferente Potentiale. Pfliigers Archiv., 284, 1-17. Kornhuber, H. H., & Deecke, L. (1985). The starting function of the SMA. The Behavioral and Brain Sciences, 8, 591-592. Kristeva, R., Keller, E., Deecke, L., & Kornhuber, H. H. (1979). Cerebral Potentials Preceding Unilateral and SimultaneousBilateral Finger Movements. Electroencephalography and Clinical Neurophysiology, 47, 229-238. Lamarre, Y., Spidalieri, G., Busby, L., & Lund, J. P. (1980). Programming of initiation and execution of ballistic arm movements in the monkey. In H. H. Kornhuber & L. Deecke (Eds.), Motivation, motor and sensory

processes of the brain: Electrical potentials, behavior and clinical use. Progress in Brain Research (Vol 54). Amsterdam: ElsevierMorthHolland. Lang, M., Lang, W., Diekmann, V., & Kornhuber, H. H. (1987). The frontal Theta rhythm indicating motor and cognitive learning. In R. Johnson, J. W. Rohrbaugh, & R. Parasuraman (Eds.), Current trends in event-related potential research. EEG Suppl. 40 (322-327). Amsterdam: Elsevier Science Publishers.

162

H. H. Komhuber et al.

Lang, M., Lang, W., Kornhuber, A. W., Bunz, J., & Kornhuber, H. H. (in press). Electrophysiological signs of directed attention prior visuomotor tracking. Proceedings of the Ninth International Conference on EventRelated Potentials of the Brain, EPIC IX, June, 1989. Lang, M., Lang W., Kornhuber, A., Groger, P., & Kornhuber, H. H. (1988). Frontomedial increase of the Bereitschafispotential prior to precisely timed movements.Electroencephalographyand ClinicalNeurophysiology, 69, 17 P. Lang, M., Lang, W., Uhl, F., Kornhuber, A., Deecke, L., & Kornhuber, H. H. (1987). Slow negative potential shifts indicating verbal cognitive learning in a concept formation task. Human Neurobiology, 6, 183-190. Lang, W., Lang, M., Goldenberg, G., Podreka, I., & Deecke, L. (1987). EEG and rCBF evidence for left frontocortical activation when memorizing verbal material. In R. Johnson, J.W. Rohrbaugh, & R. Parasuraman (Eds.), Current research in event-related brain potentials. EEG Journal, Suppl. 40, 328-334. Lang, W., Lang, M., Heise, B., Deecke, L., & Kornhuber, H. H. (1984). Brain potentials related to voluntary hand tracking, motivation and attention. Human Neurobiology, 3, 235-240. Lang, W., Lang, M., Kornhuber, A., Deecke, L., & Kornhuber, H. H. (1983). Human cerebral potentials and visuomotor learning. PfliigersArchiv, 399, 342-344. Lang, W., Lang, M., Kornhuber, A., Diekmann, V., & Kornhuber, H. H. (1988). Event-related EEG-spectra in a concept formation task. Human Neurobiology, 6, 295-301. Lang, W., Lang, M., Kornhuber, A., & Kornhuber, H. H. (1986). Electrophysiological evidence for right frontal lobe dominance in spatial visuomotor learning. Archives Italiennes de Biologie, 124, 1-13. Lang, W., Lang, M., Podreka, I., Goldenberg, G., Deecke, L. (1989). Joint Bereitschafspotential and SPECT application in functional mapping of learning-related brain activity and imagery, In G. Pfurtscheller & F. Lopes da Silva (Eds.), Functional brain imaging (251-259). Bern: Hans Huber. Lang, W., Lang, M., Podreka, I., Steiner, M., Uhl, F., Suess, E., Miiller, C., & Deecke, L. (1988). DC-potential shifts and regional cerebral blood flow reveal frontal cortex involvement in human visuomotor learning. Experimental Brain Research, 71, 579-587. Lang, W., Lang, M., Uhl, F., Kornhuber, A., Deecke, L., Kornhuber, H. H. (1988). Left frontal lobe in verbal associative learning -- a slow potential study. Expenmental Brain Research, 70, 99-108.

Cerebral Correlates of Volition and Attention

163

Lang, W., Lang, M., Uhl, F., Koska, C., Kornhuber, A., & Deecke, L. (1988) Negative cortical DC shifts preceding and accompanying simultaneous and sequential finger movements. Experimental Brain Research, 7I, 579-587. Lang, W., Obrig, H., Lindinger, G., Cheyne, D., Deecke, L. (1989). Cortical

activation patterns when initiating and coordinating bimanual motor sequences with different temporal patterns. Manuscript submitted for publication. Lang, W., Oldenkott, B., Goldenberg, G., Reisner, T., & Deecke, L. (1988). Disturbed performance of motor sequences in patients with chronic lesions of the supplementary motor area (SMA). Congress of the international Medical Society of Motor Disturbances (iSMD), June, 1988. Conference proceedings (34). Basel: AKM Congress Service. Lang, W., Podreka, I., Suess, E., Miiller, C., Zeitlhofer, J., & Deecke. L. (1988). Single photon emission computerized tomography during and between seizures. Journal of Neurology, 235, 277-284. Lang, W., Zilch, O., Koska, C., Lindinger, G., Deecke, L. (1989). Negative cortical DC shifts preceding and accompanying simple and complex sequential movements. Experimental Brain Research, 74, 99-104. Laplane, D., Talairach, J., Meininger, V., Bancaud, J., & Orgogozo, J. M. (1977). Clinical consequences of corticectomies involving the supplementary motor area in man. Journal of the Neurological Sciences, 34, 301-314. Lassen, N. A., Ingvar, D. H., & Skinhoj, E. (1978). Brain Function and Blood Flow. Scientific American, 239, 50-59. Lhermitte, F., Pillon, B., & Serdaru, M. (1986a). Human autonomy and the frontal lobes. Part I: Imitation and utilization behavior: A neuropsychological study of 15 patients. Annals of Neurology, 19, 326-334. Lhermitte, F., Pillon, B., & Serdaru, M. (1986b). Human autonomy and the frontal lobes. Part 11: Patient behavior in complex and social situations: The "Environmental Dependency Syndrome". Annals of Neurology, 19, 335-343. Lesse, H., Heath, R. G., Mickle, W. A., Monroe, R. R., & Miller, W. H. (1955). Rhinencephalic activity during thought. Journal of Nervous and Mental Disease, 122, 433-440. Liepmann, H. (1900). Das Krankheitsbild der Apraxie ("motorischen Asymbolie"). Monatsschri'fif i r Psychiatrie und Neurologie, 8, 14-44, 102-132, 182-197.

164

H. H. Komhuber et al.

Lindinger, G., Lang, W., Obrig, H., Kristeva, R., Deecke, L. (in press). Topographical analyses of radial current density -- Distribution of cortical DC-activation in motor tasks. Proceedings of the Ninth International Conference on Event-Related Potentials of the Brain, EPIC IX, June, 1989. Lindvall, O., & Bjorklund, A. (1983). Dopamine- and norepinephrine containing neuron systems: Their anatomy in the rat brain. In P.C. Emson (Ed.), Chemical neuroanatomy. New York: Raven Press. Luria, A. R. (1966). Higher cortical functions in man. London: Tavistock. Lynch, J. C. (1981). The functional organization of the posterior parietal association cortex. The Behavioral and Brain Sciences, 4, 485-534. Masdeu, J. C., Schoene, W. C., & Funkenstein, H. (1978). Aphasia following infarction of the left supplementary motor area. Neurology, 28, 1220-1223. Massion, J., Viallet, F., Massarino, R., & Khalil, R. (in press). Coordination between posture and movement in Parkinsonism and SMA lesion. In L. Deecke, J. C. Eccles & V. B. Mountcastle (Eds.), From neuron to action. Heidelberg: Springer. Maximilian, V. A., Prohovnik, I., Risberg, J., & Hakansson, K. (1978). Regional cerebral blood flow in the left cerebral hemisphere during word pair learning and recall. Brain and Language, 6, 22-31. Melnick, S. A., Hull, C. D., & Buchwald, N. A. (1984). Activity of forebrain neurons during alternating movements in cats. Electroencephalography and Clinical Neurophysiology, 57, 51-68 Mesulam, M. M. (1983). The functional anatomy and hemispheric specialization for directed attention. Trends in Neurosciences, 6, 384-387. Milner, B. (1963). Effects of different brain lesions on card sorting. Archives of Neurology, 9, 90-100. Milner, B. (1964). Some effects of frontal lobectomy in man. In K. Akert & J.M. Warren (Eds.), The frantal granular cortex and behavior. New York: MacGraw-Hill. Mountcastle, V. B. (1979). An organizing principle for cerebral function: The unit module and the distributed system. In F.O. Schmitt & F.G. Worden (Eds.), The neurosciences; N t h study program (21-42). Cambridge: MIT Press. Mountcastle, V. B. (1989). The parietal lobe system and certain aspects of spatial perception. In L. Deecke, J. C. Eccles & V. B. Mountcastle (Eds.), From neuron to action. Heidelberg: Springer. Mountcastle, V. B., Andersen, R. A., & Motter, B. C. (1983). The influence of attentive fixation upon the excitability of the light-sensitive neurons of the posterior parietal cortex. Journal of Neuroscience, 1, 1218-1235.

Cerebral Correlates of Volition and Attention

165

Mountcastle, V. B., Lynch, J. C., Georgopoulos, A., Sakata, H., & Acuna, C. (1975). Posterior parietal association cortex of the monkey: Command functions for operations within extrapersonal space. Journal of Neurophysiology, 38, 871-908. Nauta, W. J. H. (1972). Neural associations of the frontal cortex. Acta Neurobiologiae Experimentalis, 32, 125-140. Neirinckx, R. D., Canning, L. R., Piper, I. M., Nowotnik, D. P., Pickett, R. D., Holmes, R. A., Volkert, W. A., Forster, A. M., Weisner, P. S., Marriot, J. A., & Chaplin, S. B. (1987). Technetium-99m d,l-HM-PAO: a new radiopharmaceutical for SPECT imaging of regional cerebral blood perfusion. Journal of Nuclear Medicine, 28, 191-202. Neshige, R., Liiders, H., Shibasaki, H. (1988). Recording of movement-related potentials from scalp and cortex in man. Brain, 111, 719-736. Nietzsche, F. (1883). Abo sprach Zaratustra. Leipzig: Fritzsche, Nunez, P. (1981). Electric fields of the brain. The neurophysics of EEG. Oxford: Oxford University Press. Okada, Y. C., Tanenbaum, R., Williamson, S. J., Kaufman, L. (1984). Somatotopic organization of the human somatosensory cortex revealed by neuromagnetic measurements.Experimental Brain Research, 56,197-205. Otto, E., Gruner, P., & Weber, H. (1983a). Das Elektroenzephalogramm bei standardisierter geistiger Belastung: Variablentrends im Theta-, Alphaund Beta-Band. Psychiatrie, Neurologie und Medizinische Psychologie. Leipzig, 35, 720-726. Otto, E., Gruner, P., & Weber, H. (1983b). Perioden- und Amplitudenverteilungsparameter des EEGs bei geistiger Belastungzur Konkordanz maschineller und visuomanueller Auswertungsmethoden. Psychiatrie, Neurologie und Medizinische Psychologie, Leipzig, 35, 727-735. Penfield, W., Welch, K. (1949). The supplementary motor area in the cerebral cortex of man. Archives of Neurology and Psychiatty, 66, 289-317. Perrin, F., Bertrand, O., Pernier, J. (1987). Scalp current density mapping: Value and estimation from potential data. IEEE Transactions of Biomedical Engineering, BME-34, 238-288. Podreka, I., Suess, E., Goldenberg, G., Steiner, M., Brucke, T., Miiller, C., Lang, W., Neirinckx, R. D., & Deecke, L. (1987). Initial experience with Tc-99m-Hexamethylpropyleneamineoxime (Tc-99m-HMPAO) brain SPECT. Journal of Nuclear Medicine, 28, 1657-1666. Roland, P. E. (1984). Organization of motor control by the human brain. Human Neurobiology, 2, 205-216. Roland, P. E., Larsen, B., Lassen, N. A., & Skinhoj, E. (1980). Supplementary motor area and other cortical areas in organization of voluntary movements in man. Journal of Neurophysiology, 43, 118-136.

166

H. H. Kornhuber et al.

Roland, P. E., Meyer E., Shibasaki, T., Yamamoto Y. L., & Thompson, C. J. (1982). Regional cerebral blood flow changes in cortex and basal ganglia during voluntary movements in normal human volunteers. Journal of Neurophysiology, 48, 467-480. Rolls, E. T. (1983). The initiation of movements. In J. Massion, J. Paillard, W. Schultz, & M. Wiesenddnger (Eds.), Neural coding of motor perfomumce. Experimental Brain Research (Suppl 7). Heidelberg: Springer. Romani, G. L., Williamson, S. J., Kaufman, L., & Brenner, D. (1982). Characterization of the human auditory cortex by the neuromagnetic method. Experimental Brain Research, 47, 381-393. R u g , M. D., & Dickens, A. M. (1982). Dissociation of alpha and theta activity as a function of verbal and visuospatial tasks. Electroencephalography and Clinical Neurophysiology, 53, 201-207. Sasaki, K., & Gemba, H. (1982). Development and change of cortical field potentials during learning processes of visually initiated hand movements in the monkey. Experimental Brain Research, 48, 429-437. Schlag, J., & Schlag-Rey, M. (1985). Eye fixation units in the supplementary eye field of the monkey. Society for Neuroscience, (Abstracts) 1111, 82. Schreiber, H., Lang, M., Lang, W., Kornhuber, A., Heise, B., Keidel, M., Deecke, L., & Kornhuber, H. H. (1983). Frontal hemispheric differences in the Bereitschaftpotential associated with writing and drawing. Human Neurobiology, 2, 197-202. Semmes, S., Weinstein, S., Genth, L., & Teuber, H. L. (1963). Correlates of impaired orientation in personal and extrapersonal space. Brain, 86, 747-772. Sternberg, S., Monsell, S., Knoll, R. L., & Wright, C. E. (1978). The latency and duration of rapid sequences: Comparisons of speech and typewriting. In G. Stelmach (Ed.), Information processing in motor control and learning. New York: Academic Press. Straschill, M., & Takahashi, I. (1980). Slow potentials in the human subthalamus associated with rapid arm movements. Progress in Brain research, 46, 29-36. Stuss, D. T., & Benson, D. F. (1986). The frontal lobes. New York: Raven Press. Talairach, J., Bancaud, J., Geier, S., Bordas-Ferrer, M., Bonis, A., Szikla, G., & RUSU,M. (1973). The cingulate gyrus and human behavior. Electroencephalography and Clinical Neurophysiology, 34, 45-52. Tanji, J. (1984). The neuronal activity in the supplementary motor area of primates. Trends in Neurosciences, 7, 282-285.

Cerebral Correlates of Volition and Attention

167

Taylor, M. J. (1978). Bereitschaftpotential during the acquisition of a skilled motor task. Electroencephalography and Clinical Neurophysiology, 45, 568-576. Teuber, H. L. (1964). The riddle of frontal lobe function in man. In J. M. Warren & K. Akert (Eds.), The frontal granular cortex and behavior (410-477). New York: McGraw-Hill. Thorpe, S. J., Rolls, E. T., & Maddison, S. (1983). The orbitofrontal cortex: Neuronal activity in the behaving monkey. Experimental Brain Research, 49, 93-115. Uhl, F., Franzen, P., Serles, W., Lang, W., Lindinger, G., & Deecke, L. (1989).

The effect of proactive integerence in verbal cognitive learning -neurophysiological evidencefor frontal lobe activation in normal human subjects. Manuscript submitted for publication. Uhl, F., Goldenberg, G., Lang, W., Lindinger, G., Steiner, M., & Deecke, L. (1989). Imagining faces, colours and spatial maps II. Negative cortical DC potentials. Manuscript submitted for publication. Uhl, F., Lang, W., Lindinger, G., & Deecke, L. (1989a). Elaborative Strategies in word pair learning. A DC-potential study. Manuscript submitted for

publication. Wallesch, C. W., Henriksen, L., Kornhuber H. H., & Paulson, 0. B. (1985). Observations on regional cerebral blood flow in cortical and subcortical structures during language production in normal man. Brain and Language, 25, 224-233. Wiesendanger, R., & Wiesendanger, M. (1985). Afferent connections of the supplementary motor area. Experimental Brain Research, 58, Al.

Appendix Excerpt from Forster (1936): Ich habe das Feld 6aD in nahezu 40 Fallen excidiert. Die unmittelbaren Folgen bestehen erstens in einer geringen Schwache und Verlangsamung aller Bewegungen der kontralateralen Extremitat und der Neigung und Wendung des Kopfes und Rumpfes nach der Gegenseite. ... Alle Einzelbewegungen der einzelnen Extremitatenabschnitte einschliefilich der Bewegungen der Zehen, der einzelnen Finger und des Daumens bleiben aber in vollem Umfang erhalten. Hingegen erleiden die zusammengesetzten Bewegungen ... eine mehr oder weniger deutliche Lockerung des Gefiiges. Die fliefiende Kontinuitat, in welcher die einzelnen Komponenten der zusammengesetzten Bewegungen

168

H. H. Komhuber et al. raumlich und zeitlich aneinandergekoppelt sind, ist durchbrochen.

Die einzelnen Komponenten laufen mehr getrennt ab, eine hinkt der anderen her, eine Einzelkomponente kann zunlichst ganz ausbleiben und bedarf unter Umstanden zu ihrem Zustandekommen eines ad hoe erteilten besonderen Willensimpulses. (p. 279)

VOLITIONAL ACTION, W.A. Hershberger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

169

CHAPTER 7 CORTICAL MODIFICATION OF SENSORIMOTOR LINKAGES IN RELATION TO INTENDED ACTION William A. MacKay and Donald J. Crammond The extent to which volition may be found in "voluntary" movement is a moot point, nowhere better discussed than in the writings of William James (1890), the source of the quotations which follow. Volition implies conscious thought, which in normal habitual action is needed only to prepare a chain of automatic processes, what we now call a motor program. In James' scheme, sensory signals generated by execution of a movement (reafference) play a major role in triggering each element of a motor sequence. This hardly constitutes volitional action if "a strictly voluntary act has to be guided by idea, perception, and volition, throughout its whole course". But "these immediate antecedents of each movement of the chain are at any rate accompanied by consciousness of some kind. They are sensations to which we are usually inattentive, but which immediately call our attention if they go wrong." In other words, the habitual level of motor control is subconscious: the neural signals eliciting motor responses are not strictly volitional. '"The will, if any will be present, limits itself to apemission that they [the neural signals] exert their motor effects." Indeed the central idea of the motor program is that a plan of action can unfold automatically as a computer program, in sequential flows of information without conscious intervention. Libet (1985) has ingeniously demonstrated that even the neuronal initiation of a self-willed movement is an unconscious event: awareness comes after a delay of about 300 ms. In the global scheme of things, however, the conscious element always comes first, even if it does not actually trigger action. Just as the computer's programmer must set up an algorithm to process information, consciousness prepares the CNS to produce a specific action when the appropriate inputs arise. Marcel (1980) has eloquently argued that consciousness provides the link between perception and action. Its purpose is to select between possible alternatives to give action a unitary goal. The links between

170

JElliam A. MacKizy and Donald J. Crammond

perception and action can occur, of course, at many levels from the reflexive to the highly considered. These could correspond to the different levels of consciousness if consciousness is an ubiquitous property of voluntarily active systems. Concomitant levels of volition or "permission" would solve the problem encountered by James. In neurophysiology the closest that one normally ventures to the concept of consciousness is the recognition of motor "set", the state of preparedness for a particular action. But if Marcel is right, perhaps this is close to the mark. Our thesis here will be that the processes of selection of sensorimotor linkages which underlie production of intended movements, are a general model of the characteristic operations of the nervous system, at all levels, to effect volitional action in the broadest sense. The definition of a goal sets the volitional drive to act, a will which is essentially permissive, as James argued. That is, it enables a family of action sequences, any one member of which could achieve the goal depending on current peripheral conditions. The last point is both significant and ironic. Volitional actions can only be given general plans a priori within the CNS. Ultimate determination of detail originates from outside. The facilitation of interaction between the inside and outside such that required information is available when needed, is the key to successful voluntary action. We will establish this thesis in broad outline by reviewing pertinent neurological, physiological and anatomical studies, including our own work on modulation of somatosensory responses in cerebral cortex during volitional arm movement.

Frontal Syndromes Cortical lesions anterior to the motor region result in the release of reflexes, or more complex naturally conditioned responses, from their normal constraints. The sensory trigger signal, whenever it may occur, drives the motor response whether the response is behaviorally appropriate or not. Hence there is a loss of volitional control such that the program fails to direct incoming signals to serve the intended goal. One classic sign of premotor damage is forced grasping: tactile stimulation of the volar surface of the hand automatically elicits grasping (Denny-Brown & Botterell 1947). Smith, Bourbonnais, & Blanchette, (1981) have shown that ablations limited to the supplementary motor area (SMA), the medial end of the premotor strip (see Figure 2), are sufficient to produce this syndrome in monkeys. Similarly, surgical ablation of the frontal eye field region (FEF) prevents patients from making eye saccades

Cortical Sensorimotor Linkages

171

away from a visual stimulus (Guitton, Buchtel, & Douglas, 1985). In effect, the patients lose the choice of alternatives to the '"visual grasp" response, and are compelled to foveate. Denny-Brown's (1958) hypothesis of frontal lobe inhibitory regulation of parietal lobe (sensory) input to motor centers, has been extended by the recent work of Lhermitte. He has demonstrated characteristic "'utilization" or "imitation behavior'' (Lhermitte, Pillon, & Serdaru, 1986) in patients with frontal lobe lesions, generally in the inferior half of the frontal cortex. Tactile, visuotactile or visual presentation of a useable object compels the patient to grasp and use that object. Gestures made by the examiner or postures assumed are also spontaneously mimicked by the patient. Perhaps the most dramatic examples of utilization behavior have been reported by Goldberg, Mayer, and Toglia (1981). Some of their patients with medial frontal lesions have involuntarily reached for door knobs, pencils, etc., with the arm contralateral to the lesion. The patients complained, that the arm would "not do what I want it to do". They could not voluntarily initiate movements with the affected arm, although a verbal command from someone else instantly set it off, correctly. Evidently the frontal lobe damage in these cases is revealing a sensorimotor connection which always exists, but normally is suppressed until conditions are appropriate for it to be expressed. The missing neuronal tissue helped to implement goals by selecting those sensorimotor linkages which would facilitate the intended action, and by inhibiting the rest. Without such an interface, motor output tends to be both dictated by randomly occurring external events, and limited to the most common patterns. The clinical evidence forces the additional (and very Jamesian) conclusion that the internal initiation of movement uses an anatomical base which is substantially distinct from that used by external triggering processes, and also distinct from that mediating the automatic information flow which sustains a motor program once started. On a general level frontal lobe dysfunction results in an inability to organize and plan a series of acts (Milner & Petrides 1984). Sequential ordering, both retrospectively of events in recent memory and prospectively for the preparation of future events, is badly muddled. Thus there is a tendency to perseveration, repeating the same stereotyped act again and again although it does not accomplish the goal. Alternative strategies cannot be worked out because of the inability to keep track of how individual elements relate to one another. Possibly prefrontal areas serve

172

WilliamA. Mac-

and Donald J. Crammond

CONSTRUCT r e p r e s e n t a t i o n

to a c t

to p e r c e i v e

STRATEQY

INTERMODAL

SELECTION

SCHEMAS

ACTION PLAN

SYNERGY ASSEMBLY

EFFECTORS

SENSORY MAPS

R E F L E X LINKS

PERIPHERAL INPUT

Figure 1. Schematic diagram of the interactions between the frontal and parietal lobes. A anatomical outline of monkey cerebrum which summarizes some of the connections discussed by Goldman-Rakic (1988). B: the hierarchical cascade emphasizes the frontal function of selecting appropriate sensorimotor links to effect a desired action. Abbreviations: cs, central sulcus; ips, intra-parietal sulcus; ps, principal sulcus; sts, superior temporal sulcus. to construct associations between specific action elements. When one element occurs, the next in order is appropriately facilitated (and all others are suppressed). Strokes which disable the motor and premotor cortex result in paresis for obvious reasons and spasticity from unregulated spinal reflexes. Furthermore, motor sequencing within a program can be affected, as noted in Alf Brodal’s (1973) account of his own convalescence after a stroke (right internal capsule infarct producing pure motor hemiplegia). He could readily start to tie a bow but then “his fingers did not know the next move”. Breaks in “the succession of movements (due to pareses and

Cortical Sensorimotor Linkages

173

spasticity) interrupted a chain of more or less automatic movements. Consciously directing attention to the finger movements did not improve the performance; on the contrary it made it quite impossible.” Thus precentral cortex may facilitate the tight sensorimotor links which keep a motor sequence proceeding to completion, The extensive anatomical studies summarized by Goldman-Rakic (1988) have revealed strong and precisely organized interconnections between the prefrontal and posterior parietal cortex, and a large number of common target areas, some of which are depicted in Figure 1A. This framework provides an example of a physical substrate for the selection processes described above. Goldman-Rakic (1988) postulates that the parietal lobe is instrumental in constructing representations of the periphery and environment, whilst the frontal lobe puts these representations to use in order to effect actions. The selection of sensory constructions for motor use is depicted as a repeating cascade in Figure 1B to emphasize our belief that essentially the same types of interaction occur at all levels of the neuronal hierarchy. Frontal cortex may facilitate some feedforward paths and inhibit others, possibly through extensive connections to the parietal and temporal lobes. Without this anticipatory activity there is a failure to adequately prepare the next elements in a sequence, so that they are not triggered by the relevant cues.

Frontal Metabolism During Voluntary Behavior More insight into the role of frontal zones has been provided by studies of cerebral blood flow in humans (summarized in Roland 1985). Roland has distinguished 17 functional zones anterior to motor cortex. It will suffice here to describe only a few because a common theme permeates the entire lobe. Roland divides the superior frontal region (anterior to SMA) into three general zones, anterior, middle and posterior superior prefrontal cortex (Figure 2). For the anterior zone, blood flow increases in all tasks in which a primary instruction is given containing conditional directives for future processing. This area appears to participate in the recruitment of other cortical fields necessary to implement the actions required by the instruction. Simple direct commands (“open fist”) do not elicit metabolic changes in the anterior zone. The posterior portion is metabolically activated by tasks that require the analysis of sensory information or information retrieved from memory as a prior condition for further processing. The most intense increases

174

William A. MacKay and Donald J. Crarnmond

Figure 2. Outline of human cerebrum showing some of the functional zones described by Roland. Abbreviations: cs, central sulcus; M1, primary motor cortex; FEF, frontal eye field; PM, premotor cortex; SMA, supplementary motor area; SPF, superior prefrontal cortex.

occur for communication between different sets of cortical fields in rapid succession, e.g., describing from memory the furniture layout of your living room. Thus frontal zones may select a set of other cortical areas appropriate to handling the demands of the task. Presumably more anterior zones function at a higher, more general level, while posterior zones become increasingly particular. Areas closest to the motor cortex manipulate sensorimotor triggers actually driving movement elements. The region immediately anterior to the motor cortex is divided into two functional zones by Roland, the premotor cortex and more medially the supplementary motor area (SMA). SMA has increased blood flow during the preparation and performance of complex motor sequences, including speech. SMA is not activated during simple repetitive motions (tapping a finger) to which the subject pays no attention. Nor is the premotor cortex significantly activated in such cases. The premotor region is particularly involved in exploratory limb movements carried out under sensory guidance. It must be remembered that the frontal lobe and basal ganglia are interdependent. The output of the basal ganglia is largely directed at the frontal lobe and thus may be used as a critical selector of frontal regions. Indeed, the striatum could provide a key path by which volition at a high level could access specific frontal "organizers." Regardless of the likely involvement of the basal ganglia, Roland (1985) has drawn some general conclusions about the hierarchy of steps underlying the generation of voluntary movement which are worthy of note. Firstly, before the execution of voluntary behavior, the brain ''tunes the cortical fields that are expected to participate in the processing of the

Cortical Sensorimotor Linkages

175

task-related information." Secondly, "when the fields are tuned, they are recruited for the task and subsequently participate in the processing of the task-related information. However, since the brain has to estimate beforehand which cortical areas might be necessary for the task, some areas might be recruited in certain cases which actually do not participate in the processing of the task-related information." In an interactive model of volitional action, the final selection of output processors depends ultimately on current conditions. The preparatory process which Roland calls Yuning" might be likened to the creation of a dynamic attractor toward which subsequent input signals are drawn, just as a ball rolls toward the bottom of a basin (Kwan 1988).

Sensory Transmission During Purposeful Movement To account for the selective use of sensory inputs in motor programs, physiologists have proposed that gating or modulatory processes alter responses to sensory inputs according to a motor "set", specific for the intended movement. The gating can occur at all levels of the neuraxis, from the segmental to the cortical. The mechanisms by which cerebral cortex can regulate its own somatosensory inputs have been studied in some detail. Both via the corticospinal tract and corticobulbar-reticulospinal tracts, the cortex influences transmission within the dorsal horn and dorsal column nuclei (DCN). The dominant effect is one of inhibition of sensory transmission (Towe & Jabbur 1961), an inhibition which arises from somatosensory, motor and premotor cortex (Felix & Wiesendanger 1970). Nevertheless, significant excitatory effects are also present, especially from motor cortex. In general, motor cortex appears to have a net excitatory effect on the dorsal horn, including spinothalamic projection neurons (Yezierski, Gerhart, Schrock, & Willis, 1983). Giuffrida, Sanderson, and Sapienza (3 985) have shown that motor cortical facilitation of DCN transmission occurs only for afferent input from the vicinity of the activated muscle, and only for the duration of the motor cortex outflow. Other DCN cells with receptive fields remote from the activated muscle are inhibited. This work was done in rats, however, and may not be generally applicable to other species. In monkeys, for example, Jiang, Chapman, and Lamarre (1988) found that microstimulation of the motor cortex inhibited air puff responses in primary somatosensory cortex when the receptive field

176

William A. MacKky and Donald J. Crammond

overlapped or was distal to the activated muscle: more proximal fields were not affected. By means of layer VI efferents, each cortical area regulates its own thalamic inflow. The corticothalamic projection from somatosensory cortex appears to have a mainly excitatory effect on thalamocortical neurons (Yuan, Morrow, & Casey, 1986). Thus the thalamocortical circuit offers a mechanism for amplifying particular somatosensory signals. This mechanism may be used, however, only in selected circumstances. Because of the dominant descending inhibition of DCN, voluntary movement of a limb significantly reduces evoked potentials in the medial lemniscus due to cutaneous stimulation of that limb, starting up to 200 ms before the movement (Ghez & Pisa 1972; Coulter 1974). Passive movements have no effect. At the thalamocortical level, Chapman, Jiang, and Lamarre (1988) have found that additional inhibition is imposed. It is, however, strictly limited to the time of actual limb displacement and appears to be associated with the arrival of reafferent input. At the thalamocortical level, passive and active movements produce the same suppressive effect on cutaneous responsiveness. The modulation of sensory transmission appears to be graded in proportion to the level of motor output (Coulter 1974; Ghez & Pisa 1972). It may be misleading, however, to say that sensory inflow is reduced in proportion to movement speed. The gating process is probably a very necessary measure to maintain a meaningful flow of information without saturation. The faster the movement, the greater the barrage of activity bombarding the sensorimotor regions and the greater the necessity to set limits on what gets through so that it can be effectively processed. A general homeostatic rule of sensory regulation could be postulated as follows: sensory responsiveness of the cerebral cortex is reduced in proportion to the total instantaneous afferent activity to that cortical region. For example, responses of motor cortical cells to paw stimulation are suppressed just prior to and during the stance phase of locomotion (Palmer, Marks, & Bak, 1985), i.e., when paw receptors would be most activated. In somatosensory cortex of rats, Chapin and Woodward (1982) found that responses to paw tactile stimulation during the step cycle of locomotion were modulated in one of two patterns. Responsiveness to paw stimulation was either reduced throughout the step cycle except for brief instants during the swing phase, or it was reduced throughout most of the cycle but disinhibited just prior to footfall. The latter case is an example where the general rule was broken and cortical neurons showed

177

Cortical Sensorimotor Linkages

t e s t pulse times

i i i f i r r r r r 1

2

3

4

5

.......

I target

_...,.

........ ............................................

ex

forearm displacement I

I

0.5 s e c

Figure 3. Paradigm to test cortical proprioceptive responsiveness during a cyclic forearm movement of alternating flexion and extension. The monkey tracked a stepping visual target (dotted line). In any one cycle a test pulse perturbation could be delivered at one of the times indicated by the arrows, to stretch either the flexor or extensor muscles. Test pulses were given in random order at 5 different phases of the active flexion and the same 5 phases of extension. The forearm displacement is a grand mean from all unperturbed trials in one monkey and has error bars marking the standard error at the time of test Dukes. relatively high responsiveness at a time when receptors were most active. Moreover, Iwamura, Tanaka, Sakamoto, and Hikosaka (1985) have reported neurons in areas 1 and 2 which were responsive to cutaneous stimulation of the hand only during active manipulation of objects. Such "active touch" cells provide evidence that the CNS, probably the cerebral cortex itself, can and does augment specific signals during the course of voluntary movements. The amplified signals are of potential use in guiding or controlling the movement performed. To demonstrate this point, Hikosaka, Tanaka, Sakamoto, and Iwamura (1985) showed that reversible lesions of discrete regions of the area 2 finger representation led to a loss of finger coordination during natural manipulative behaviors.

Cortical Responsiveness During Movement The studies discussed above were chiefly concerned with cutaneous inputs. Proprioceptive signals, which provide particularly important cues for the control of ongoing limb movements, could be treated differently

178

William A. MacKay and Donald J. Crammond

within the CNS. However, Evarts and Fromm (1977) have shown that some corticospinal tract cells in motor cortex are insensitive to hand perturbations during very fast supination/pronation movements of the hand, whereas comparable perturbations during slow movements elicit strong responses, i.e., the general rule of afferent regulation seems to be obeyed. To examine the gating of proprioceptive input from the moving arm in more detail, we studied the responsiveness of sensorimotor cortex to a test pulse (forearm perturbation) delivered during the course of active movement. We gave test pulses at different phases of a normalspeed voluntary elbow movement, to look for changes which might reveal when the proprioceptive signal was most useful to motor control processes at the cortical level. In two macaque monkeys trained to perform alternating flexions and extensions of the forearm, the responsiveness of cortical neurons in the sensorimotor region (areas 4,3a,1,2,5) to a uniform torque perturbation was tested at 10 phases in the movement cycle (Figure 3), 5 in each direction of movement. When cells were selected such that their receptive field was proprioceptive input from the elbow, i.e., the test stimulus, then specific patterns of modulation were observed over the movement cycle. Responsive neurons with non-elbow receptive fields tended to conform to a standard modulation pattern of minimal responsiveness at peak velocity of movement in either direction. In almost all cases neuronal responsiveness was significantly reduced during the movement cycle compared to the same test stimulus delivered to the "resting" forearm. Thus the data presented here reveal relative changes in responsiveness over the flexion-extension movement cycle. For the elbow proprioceptive cells, the majority of neurons showed a modulated responsiveness in all cortical areas studied. The responsiveness of area 4 (motor cortex) neurons showed the least reduction of all areas at the onset of agonist muscle activity, a phase when responsiveness is most commonly reduced among neurons responding to deep elbow inputs. Of all the cells sampled, area 4 neurons were found to be the most responsive to deep elbow inputs during the agonist half of the movement cycle. That is to say, area 4 cells receiving input from the elbow flexor muscles (and generally evoking elbow flexion when stimulated) were most responsive prior to and during active flexion of the forearm. An example of just such a cell is illustrated in Figure 4A. Perturbation-evoked field potentials from the deep layers of motor cortex showed the same property (Figure 4B). The early negativity was well maintained throughout the agonist (flexion) phase but was greatly

179

Cortical Sensorimotor Linkages

FLEXION

A

EXTENSION

lestretching flexors

5

7

rns

D

0

D

2 ......... ...... .. . :I 7

s. 100

200

0

100

200

.I ?

I

I

0

200

0

2 0 0 rns

Figure 4. Responsiveness of motor cortex to a forearm perturbation stretching the elbow flexor muscles, tested at the 5 phases of active flexion and extension shown in Figure 3. A: post stimulus time histograms from a representative neuron, localized by the filled circle on diagram of electrode track at left. All histograms are scaled to the number of trials which is indicated. Vertical dotted lines mark the onset of significant increases in firing rate. Vertical calibration provides probability of firing. B: local field potential recorded in the deep layers of motor cortex (filled circle). Vertical scale is 0.5 mV, positive upwards. CS: central sulcus. attenuated during the antagonist (extension) phase. Responses of neurons in area 3a were, on the whole, rather similar to those of motor cortical cells, with maximum reduction during the antagonist half of the cycle. But many cells (32%) in this primary sensory region showed no phase-dependent changes. In area 1, however, a remarkable contrast was noted. The responsiveness of area 1 neurons during the antagonist half of the cycle, was consistently greater than that

180

William A. MacKay and Donald J. Crammond

A stretching

FLEXION

extensors

PHASE 3

, ms

B

EXTENSION

0

100

200

0

100

200

ACTIVE FLEXION

41

;L

$j

I

EXTENSION

45

0 sec

.5

1

Figure 5. A: Responsiveness of a neuron in area 1 to test pulses in phase 3 (peak velocity) of flexion and extension. B: Discharge of same cell during active flexion and extension without test pulses. Vertical calibration indicates 10" displacement and 0.5 probability of discharge.

of cells in other cortical areas responding to elbow stimuli. Thus an area 1 cell responding to elbow extensor muscle stretch was most responsive to extensor inputs during active flexion, when the extensor muscle was in fact being stretched (Figure 5A). Consequently this neuron increased its discharge during active flexion and decreased it during extension (Figure 5B). This is an important observation for several reasons. Firstly, it underscores the point that generalizations such as reafferent input being suppressed at times when it is most intense may not be universally applied. Secondly, it shows that a specific region in the somatosensory cortex appears to monitor input from the antagonist muscle as it is stretched. This complements the existing evidence that proprioceptive information from the stretched antagonist is used in regulating movement amplitude (Capaday & Cooke 1983). In posterior parietal area 5, neurons could show a similar pattern to area 1 (e.g., Figure 6), or the area 4-3a pattern. Generally, however, responsiveness was strongest at the end of the agonist phase of movement and prior to the antagonist phase. In other words, cells responding to

181

Cortical Sensorimotor Linkages

EXTENSION

FLEXION

I

2Lkpubn 27

- -

JI

L

:

3 24

4 stretching extensors

1

rns

:

0

100

1

o . : : : - : . *

200

0

100

200

Figure 6. Responsiveness of a neuron in area 5 to the test pulses at 5 phases of flexion and extension. IPS: intraparietal sulcus.

elbow extensor muscle stretch were most responsive in advance of the time when the extensor muscle would be stretched, i.e., prior to active flexion (Figure 6). Many of the studied neurons in all cortical regions, failed to show drops in responsiveness during the peak velocity agonist phase of movement. Again this shows the danger of drawing conclusions from generalizations about reafferent signals. The sensorimotor cortex is clearly responsive to potential inputs from a muscle throughout its period of active contraction. Note that motor cortex cannot be slavishly driven by incoming sensory signals, but carefully selected and controlled sensory signals do get access to the motor apparatus where they perform an important function. For example, transcortical loops could mediate aspects of load compensation (Phillips, 1969) during a movement. In our experiments we were able to monitor the late stretch reflex (latency 50-60 ms) elicited by the test pulses. On the assumption that this reflex is at least partly transcortical (Cheney & Fetz 1984), it was of interest to determine whether changes in reflex amplitude for different testing times during the movement cycle paralleled changes in cortical unit responsiveness. The reflex showed a very consistent pattern: it was largest

182

William A. Mackky and Donald J. Crammond

at the onset of agonist contraction (cf. MacKay et al. 1983). Although most single cortical units did not show the same modulation pattern, the pattern was reasonably similar to that of the entire sampled population of elbow-related motor cortical neurons, which showed maximum responsiveness in phase 1 of agonist contraction, and a significant loss of responsiveness for the remaining 4 phases. The onset of contraction is normally when the greatest inertial load is encountered by the agonist muscle. It is therefore to good purpose that the motor cortex is prepared at this instant to trigger motor reinforcement if muscle spindle input indicates resistance. This is an example of the simplest and most direct (reflexive) triggering of action.

Movements Cued by a Sensory Stimulus When movements are cued by the occurrence of a sensory stimulus, neuronal responsiveness to the stimulus is enhanced in some cortical areas. For example, in a monkey conditioned to rapidly move his hand upon receipt of a vibrotactile stimulus, responses to the stimulus in areas 3a and 1 were increased compared to those when no movement was made (Nelson 1984). Cortical area 3b, however, showed little response modulation. It appears that primary sensory zones may faithfully monitor peripheral conditions regardless of the motor set, while other areas which provide an interface for sensory signals with the motor apparatus are greatly modified by motor preparation. One such interface is somatosensory association area 5, where Chapman, Spidalieri, and Lamarre (1984) have found movement-dependent modulation of responses to forearm perturbations. For many cells, responses to forearm perturbation were small or absent when the associated movements were not subsequently made. Similarly, in motor cortex itself many cells respond to a vibrotactile stimulus if it cues a movement, but the response is usually attenuated or blocked if the movement is to be withheld (Kurata & Tanji 1985). Motor cortical responses to somatic cues are frequently bimodal with a short-latency component coming directly from somatosensory cortex or thalamus and a later component linked to the intended movement. The latter is always suppressed if no movement is made: the former is only partly blocked (Kurata & Tanji 1985). Visual and auditory signals do not share the same tight coupling to motor cortical neurons that somatic inputs possess (Lamarre, Busby, & Spidalieri, 1983). In motor (and premotor) cortex, responses to visual

Cortical Sensorimotor Linkages

183

B

rJ\

elbow anale

I

E

:

:

:

1

:

:

2

:

t

:

3

:

:

4

I ssc

i

5

i

Figure 7. Responses in area 7a of parietal lobe to task-related events. A: Peri-event time histogram with 3 sample trials below showing a cell's firing pattern related to appearance of target on the video monitor (see inset), to the movem,ent of the square cursor indicating hand movement, and to anticipation of the reward given at the end of each trial. Discharge was not correlated to eye movements monitored in EOG records. B: Two trials of the forearm visual tracking task showing discharge of a neuron during the time that the cursor is within the target window. The stepping target is indicated by the broken line, and the arrows mark the times when torque perturbations were given. stimuli are only observed if the stimulus serves as a cue for action (Kwan, MacKay, Murphy, & Wong, 1985). When that action is not fully specified in advance, the visual cue activates a rather diverse population of neurons, many of which are not needed for the subsequent movement. In this regard, Riehle and Requin (1989) have found precentral cells that respond much more strongly to a visual cue when the motor reaction is not known in advance compared to when it is known. In prefrontal cortex (principal sulcus), visual (or other) stimuli excite large numbers of neurons provided the stimulus serves as a behavioral cue, e.g., for a conditioned motor response (Fuster 1984). The frontal eye fields (FEF) may be the only frontal region where visual responses

184

William A. MacKizy and Donald J. Cramrnond

per se can be found. Yet, even here, about half of the visual responses are highly dependent on the occurrence of a saccade which moves gaze onto the neuron’s receptive field (Bruce & Goldberg 1984). In area 7 of parietal lobe, responses to visual stimuli are also highly dependent on motor set (Bushnell, Goldberg, & Robinson, 1981; Mountcastle, Andersen, & Motter, 1981). In monkeys performing the same flexion-extension tracking paradigm mentioned earlier (either wrist or elbow movement), we have found within area 7a some rather complex visual responses (Figure 7). Cells respond to the cue to start (the appearance of the target on a video monitor), and to the moving cursor (on the video monitor) as the tracking movement is performed. Occasional cells respond to the placement of the cursor correctly within the target window (Figure 7B). Such types of responses would be disastrous if they directly commanded a fixed motor output, as in a reflex. But given an appropriately prepared motor pattern, these responses could serve useful functions as temporal triggers. The enhancement of sensory responses when they are to be used to guide or trigger action, provides a selection mechanism. But in the case of visual or auditory triggers especially, specific connections to effectors cannot be made without losing adaptability. Therefore, a process of preparation must precede the trigger volley so that it may be functionally focused onto the appropriate motor output zones. Moreover, any sensory cue to prompt a movement will not actually trigger it unless the movement is specified in advance. Thus when the movement is not known in advance, a level of interpretative processing intercedes between sensory cue and triggering of motor output. The source of the actual trigger is embedded in the CNS.

Sensory Representation Before Movement There is a natural tendency to expect activity in somatosensory cortex after movement onset when the great barrage of reafferent input arrives. Indeed such is the case (Bioulac & Lamarre 1979). But from foregoing arguments regarding the preparatory facilitation of sensorimotor links, a logical outcome would be the occurrence of parietal discharge before movement. In fact there are now many reports of parietal activity preceding movement onset, usually in neurons which are also activated during the movement. Precocious discharge is most conspicuous in posterior parietal cortex, but it can also be found in somatosensory areas

Cortical Sensorimotor Linkages

185

3a, 1 and 2 of the postcentral gyrus (Fromm & Evarts 1982; Nelson 1987; Soso & Fetz 1980). In area 5, Seal, Gross, Doudet, & Bioulac, (1983) identified 2 populations of cells. The first discharged up to 280 ms before movement onset and did not have clear peripheral input: discharge persisted after deafferentation. The other population responded to peripheral stimuli and discharged after the onset of movement, unless the animal was deafferented. One may postulate that the first population functioned to modulate the responsiveness of the second during the movement. However, MacKay, Kwan, Murphy, and Wong (1978) observed area 5 cells with a tonic relationship to wrist angle and clear sensory input from passive bending of the wrist, which changed their discharge rate to correspond to the intended final angle immediately before the commencement of the movement. The results of Favorov, Sakamoto, and Asanuma (1988) indicate that much of the premovement discharge in motor cortex and in area 2 is due to minute changes in muscle tone and consequent afferent feedback to sensory and motor cortex. Furthermore, the corticoperipheral loop is necessary for skilled execution of a task involving picking up food from a revolving wheel. Dorsal column section abolished both cortical premovement discharge and performance skill. In other words, somatosensory inputs can play a major role in the preparatory process for action (cf. Dubrovsky & Garcia-Rill 1973). A peripheral loop, however, does not explain the premovement discharge which survives deafferentation (Seal et al. 1983). Very recently, Crammond and Kalaska (in press) recorded area 5 unit discharge during a preparatory waiting period: the monkey had full information about where he was to move his hand when the GO signal appeared (change of target lamp color). During the preparatory period the discharge rates of over half of the sampled cells changed. Moreover, these changes were predictive of the activity during the subsequent movement. The preferred direction for preparatory discharge was similar to that for movement-related discharge. In area 2 neurons, however, preparatory discharge showed no directional preference. The preparatory discharge of the area 5 cells would appear to be an excellent example of what Roland (1985) has called "tuning" of cortical fields which are to be used to control the motor performance. The tuning is direction-specific, and since the intended movement is guided by sensory inputs, it is likely that premotor cortex may be critically involved in the tuning process. Certainly Riehle and Requin (1989) and Weinrich, Wise, and Mauritz

186

William A. MacKay and Donald J. Crammond

(1984) have found direction-specific preparatory activity within the premotor area. Batuev, Shaefer, and Orlov (1985) have also observed spatially selective preparatory activity in both prefrontal cortex (principal sulcus) and in the intraparietal sulcus. Source localization of scalp potentials in humans during pre-cued preparatory periods in which either force or direction (or both) of an upcoming elbow movement was known, have revealed that the major locus of developing negativity (CNV) was in the parietal lobe, especially when direction was cued in advance (Bonnet & MacKay, 1989). Since many, if not most, parietal neurons have demonstrable receptive fields, it is not unreasonable to consider pre-movement parietal activity a form of sensation, or at least specific sensory facilitation. We have encountered a number of cells in areas 5 and 7 which discharge in anticipation of receptive field contact by the examiner (MacKay & Crammond 1987; cf. Hyvarinen 1982). The discharge is triggered by sight of the examiner's hand approaching the receptive field, and is reinforced at the moment of contact. Such discharge may correspond to anticipatory sensation from the target zone. When contacting one's own body, anticipatory sensation from the target region may involve the cerebellum. Sasaki (1985) reported a case study of a patient who suffered a localized left cerebellar infarct. Whenever the patient attempted to touch some part of his body with the left hand, the intended target dropped out of his body perception: he felt as though he was groping in "a sea of clouds". It appears that afferent inflow from the intended target region was subjected to the same movement-related suppression which affects the moving limb at the level of the cord and brainstem. Sasaki postulated that compensation for loss of input to the target representation may normally be mediated by dentate nucleus facilitation of parts of the prefrontal and premotor cortex. This possibility does fit with Roland's characterization of frontal areas. The preparation for action involves not only regulation of the input channels to be used, but an initial construction of the internal representation of the action. Thus when a voluntary movement is learned, "the idea of the movement's sensory effects will have become an immediately antecedent condition to the production of the movement itself' (James 1890). Such anticipatory representation may prepare the way for motor triggers to activate specific neuronal groups. Anticipation of non-somatic events (e.g., visual cues) occurs in premotor cortex (Mauritz & Wise 1986), and may similarly serve to set up reactions to the anticipated events.

Cortical Sensorimotor Linkages

187

Cortical Preparatory Mechanisms Evarts (1984) postulated that set-dependent premotor cortex activity may alter motor cortical responses to sensory trigger stimuli. The available data are in accord with this idea, although direct evidence is not abundant. Microstimulation of the supplementary motor area (SMA) can suppress motor cortical responses to muscle afferent inputs (Hummelsheim, Wiesendanger, & Bianchetti, 1986). However, the key word in Evarts' postulate was "set-dependent": effects on sensory responses must be tested within a specific action context. Tanji, Kurata, and Okano (1985) locally cooled SMA while monkeys performed a key-press to a sensory cue or withheld responses for a different cue. During cooling frequent errors were made, with motor cortical responses to the nontrigger cue always preceding the erroneous key-presses. A population of SMA cells responded specifically to the nontrigger cue (Kurata & Tanji 1985): silencing these cells by cooling may have allowed the erroneous motor cortical responses. Inhibition seems to be the dominant effect of SMA on cortical sensorimotor connections. The use of delay or preparatory periods in behavioral paradigms for cortical unit recording has revealed the existence of a continuous ordering of neuronal activities from cue-related to delay-related to movementrelated, all within the same functional region. This is true within prefrontal (principal sulcus) cortex (Fuster 1984), the frontal eye fields (Bruce & Goldberg 1984), premotor cortex (Riehle & Requin, 1989; Weinrich et al. 1984), motor cortex (Lecas, Requin, Anger, & Vitton 1986; Riehle & Requin, 1989), and posterior parietal cortex (Batuev et al. 1985; Seal & Commenges 1985). One may therefore postulate that functional zones are structured as internal cascades, with cue responses in some way facilitating delay-related activity which in turn prepares movement-linked discharge (Requin, Riehle, & Seal, 1988). Such local cascades would be in addition to the interactions schematized in Figure lB, and would necessarily be closely interwoven with them. Thus flow of activity along a local cascade would be critically dependent on the receipt of external signals from other regions. This feature of cortical information flow is sketched in Figure 8. Extrinsic preparatory facilitation of a cortical region initiates local interactions between cue-, delay- and actionrelated clusters (labelled 1, 2 and 3, respectively in Figure 8). The net effect is to create a potential well as it were, which serves as an "attractor" for the final motor trigger. Thus the trigger does not require restricted anatomical connections.

188

WUhm A. MacKay and Donald J. Crammond

Figure 8. Highly abstracted model of preparatory facilitation of a cortical area to create an attractor for a trigger signal. Local interactions between adjacent clusters (arrows) receiving preparatory facilitation, intensify the attractor structure. Clusters are classified as 1: cue-related, 2: delay-related, or 3: movement-related. The true anatomical relationships of the functional groups may vary among cortical areas. The different task-related clusters may be layered in specific cortical laminae. In prefrontal cortex, cue responses are found mainly in layer IV, delay activity in layers 11-111 and action-related activity in layers V-VI (Sawaguchi, Matsumura, & Kubota, 1989). Delay-related activity may be in part involved in the computational processing interposed between cue and movement. It continues even when the sensory cue is circumvented. For example, "visual cells" in FEF are silent during saccades made to memorized targets in the dark. Only the "visuomovement" (or delay-related) and "movement" cells are activated (Bruce & Goldberg 1984), the former to a lesser extent than the latter. Thus the information for spatial guidance can come either from the periphery or from cognitive memory. Where the trigger signal comes

Cortical Sensorimotor Linkages

189

from to ignite the movement cells remains a mystery: perhaps it is released by basal ganglia disinhibition. Except for the simplest of reflexes, no movement is performed unless it is prepared in advance. Particular sensory channels are facilitated, others suppressed, and specific internal representations necessary to guide the movement are activated. As a result of volitional setting up, motor triggers, whether they arise from filtered external stimuli or internal signals, initiate a prepared sequence of events. A motor trigger itself need have no motor relevance other than time: it is led inexorably to the correct target.

Acknowledgements Many thanks are due Drs. Alexa Riehle and Hon Kwan for suggesting improvements to drafts of the text. The experimental studies were supported by MRC of Canada.

References Batuev, A.S., Shaefer, V.I., & Orlov, A.A. (1985). Comparative characteristics of unit activity in the prefrontal and parietal areas during delayed performance in monkeys. Behavioral Brain Research, 16, 57-70. Bioulac, B., & Lamarre, Y. (1979). Activity of postcentral cortical neurones of the monkey during conditioned movements of a deafferented limb. Brain Research, 172, 427-437. Bonnet, M., & MacKay, W.A. (1989). Changes in CNV related to preparatory advance information about direction and force of an intended movement. Brain, Behavior and Evolution, 33, 147-152. Brodal, A. (1973). Self-observations and neuro-anatomical considerations after a stroke. Brain, 96, 675-694. Bruce, C.J., & Goldberg, M.E. (1984). Physiology of the frontal eye fields. Trends in Neurosciences, 7, 436-441. Bushnell, M.C., Goldberg, M.E., & Robinson, D.L. (1981). Behavioral enhancement of visual responses in monkey cerebral cortex. I. Modulation in posterior parietal cortex related to selective visual attention. Journal of Neurophysiology, 46, 755-772. Capaday, C., & Cooke, J.D. (1983). Vibration-induced changes in movementrelated EMG activity in humans. Experimental Brain Research, 52, 139-146.

190

WilliamA. M a c k y and Donald J. Crammond

Chapin, J.K., & Woodward, D.J. (1982). Somatic sensory transmission to the cortex during movement. 11. Phasic modulation over the locomotor step cycle. Experimental Neurology, 78, 670-684. Chapman, C.E., Spidalieri, G., & Lamarre, Y. (1984). Discharge properties of area 5 neurones during arm movements triggered by sensory stimuli in the monkey. Brain Research, 309, 63-77. Chapman, C.E., Jiang, W., & Lamarre, Y. (1988). Modulation of lemniscal input during conditioned arm movements in the monkey. Expenmental Brain Research, 72, 316-334. Cheney, P.D., & Fetz, E.E. (1984). Corticomotoneuronal cells contribute to long-latency stretch reflexes in the rhesus monkey. Journal of Physiology, 349, 249-212. Coulter, J.D. (1974). Sensory transmission through lemniscal pathway during voluntary movement in the cat. Journal of Neurophysiology, 37, 831-845. Crammond D.J., & Kalaska, J.F. (In press). Neuronal activity in primate parietal cortex area 5 varies with intended movement direction during an instructed delay period. Experimental Brain Research. Denny-Brown, D. (1958). Nature of apraxia. Journal of Nervous and Mental Disease, 126, 9-32. Denny-Brown, D., & Botterell, E.H. (1947). The motor functions of the agranular frontal cortex. Research Publications Association for Research In Nervous and Mental Disease, 27, 235-345. Dubrovsky, B, & Garcia-Rill, E. (1973). Role of dorsal columns in sequential motor acts requiring precise forelimb projection. Experimental Brain Research, 18, 165-177. Evarts, E.V. (1984). Neurophysiological approaches to brain mechanisms for preparatory set. In S. Kornblum & J. Requin (Eds.), Preparatoy States and Processes (pp. 137-153). Hillsdale NJ: Erlbaum Assoc. Evarts, E.V., & Fromm, C. (1977). Sensory responses in motor cortex neurons during precise motor control. Neuroscience Letters, 5, 267-272. Favorov, O., Sakamoto, T., & Asanuma, H. (1988). Functional role of corticoperipheral loop circuits during voluntary movements in the monkey: a preferential bias theory. Journal of Neuroscience, 8, 3266-3277. Felix, D., & Wiesendanger, M. (1970). Cortically induced inhibition in the dorsal column nuclei of monkeys. Pjliigers Archiv, 320, 285- 288. Fromm, C., & Evarts, E.V. (1982). Pyramidal tract neurons in somatosensory cortex: central and peripheral inputs during voluntary movement. Brain Research, 238, 186-191. Fuster, J.M. (1984). Behavioral electrophysiology of the prefrontal cortex. Trends in Neurosciences, 7, 408-414.

Cortical Sensorimotor Linkages

191

Ghez, C., & Pisa, M. (1972). Inhibition of afferent transmission in cuneate nucleus during voluntary movement in the cat. Brain Research, 40, 145-151. Goldberg, G., Mayer, N.H., & Toglia, J.U. (1981). Medial frontal cortex infarction and the alien hand sign. Archives of NeuroZogy, 38, 683-686. Goldman-Rakic, P.S. (1988). Topography of cognition: parallel distributed networks in primate association cortex. Annual Review of Neuroscience, 11, 137-156. Giuffrida, R., Sanderson, P., & Sapienza, S. (1985). Effect of microstimulation of movement-evokingcortical foci on the activity of neurons in the dorsal column nuclei. Somtosensory Research, 2, 237-247. Guitton, D., Buchtel, H.A., & Douglas, R.M. (1985). Frontal lobe lesions in man cause difficulties in suppressing reflexive glances and in generating goal-directed saccades. Experimental Brain Research, 58, 455-47. Hikosaka, O., Tanaka, M., Sakamoto, M., & Iwamura, Y. (1985). Deficits in manipulative behaviors induced by local injections of muscimol in the first somatosensory cortex of the conscious monkey. Brain Research, 325, 375-380. Hummelsheim, H., Wiesendanger, M., & Bianchetti, M. (1986). The supplementary motor area modulates perturbation-evoked discharges of neurones in the precentral motor cortex. Neuroscience Letters, 67, 119-122. Hyvarinen, J. (1982). The parietal cortex of monkey and m n . Berlin: Springer-Verlag. Iwamura, Y., Tanaka, M., Sakamoto, M., & Hikosaka, 0. (1985). Vertical neuronal arrays in the postcentral gyrus signalling active touch: a receptive field study in the conscious monkey. Experimental Brain Research, 58, 412-420. James, W. Principles of Psychology. 1890, New York: Holt. Jiang, W., Chapman, C.E., & Lamarre, Y. (1988). The topographical modulation of somatosensory transmission to SI cortex produced by microstimulation of MI cortex in the monkey. Society for Neuroscience Abstracts, 14(1), 509. Kurata, K., & Tanji, J. (1985). Contrasting neuronal activity in supplementary and precentral motor cortex of monkeys. 11. Responses to movement triggering vs. nontriggering sensory signals. Journal of Neurophysiology, 53, 142-152. Kwan, H.C. (1988). Network relaxation as behavioral action. Research in Biological and Computational Vision Technical Report, RBCV-TR-8826, Dept. of Computer Science, University of Toronto. Kwan, H.C., MacKay, W.A., Murphy, J.T., & Wong, Y.C. (1985). Properties of visual cue responses in primate precentral cortex. Brain Research, 343, 24-35.

192

William A. MacKay and Donald J. Crammond

Lamarre, Y., Busby, L., & Spidalieri, G. (1983). Fast ballistic arm movements triggered by visual, auditory, and somesthetic stimuli in the monkey. I. Activity of precentral cortical neurons. Journal of Neurophysiology, 50, 1343-1358. Lecas, J.-C., Requin, J., Anger, C., & Vitton, N. (1986). Changes in neuronal activity of the monkey precentral cortex during preparation for movement. Journal of Neurophysiology, 56, 1680-1702. Lhermitte, F., Pillon, B., & Serdaru, M. (1986). Human autonomy and the frontal lobes. Part I: Imitation and utilization behavior: a neuropsychological study of 75 patients. Annals of Neurology, 19, 326-334. Libet, B. (1985). Unconscious cerebral initiative and the role of conscious will in voluntary action. Behavioral and Brain Sciences, 8, 529-566. MacKay, W.k, & Crammond, D.J. (1987). Neural correlates in posterior parietal lobe of the expectation of events. Behavioral Brain Research, 24, 167-179. MacKay, W.A., Kwan, H.C., Murphy, J.T., & Wong, Y.C. (1978). Responses to active and passive wrist rotation in area 5 of awake monkeys. Neuroscience Letters, 10, 235-239. MacKay, W.A., Kwan, H.C., Murphy, J.T., & Wong, Y.C. (1983). Stretch reflex modulation during a cyclic elbow movement. Electroencephalography and Clinical Nerophysiology, 55, 687-698. Marcel, A.J. (1980). Explaining selective effects of prior context on perception: the need to distinguish conscious and pre-conscious processes. In J. Requin (Ed.), Anticipation and Behaviour (pp. 411-432). Paris: C.N.R.S. Mauritz, K. H., & Wise, S.P. (1986). Premotor cortex of the rhesus monkey: neuronal activity in anticipation of predictable environmental events. Experimental Brain Research, 61, 229-244. Milner, B., & Petrides, M. (1984). Behavioural effects of frontal-lobe lesions in man. Trends in Neurosciences, 7, 403-407. Mountcastle, V.B., Andersen, R.A., & Motter, B.C. (1981). The influence of attentive fixation upon the excitability of the light-sensitive neurons of the posterior parietal cortex. Journal of Neuroscience, I, 1218-1235. Nelson, R.J. (1984). Responsiveness of monkey primary somatosensory cortical neurons to peripheral stimulation depends on "motor set." Brain Research, 304, 143-148. Nelson, R.J. (1987). Activity of monkey primary somatosensory cortical neurons changes prior to active movement. Brain Research, 406, 402-407. Palmer, C.J., Marks, W.B., & Bak, M.J. (1985). The responses of cat motor cortical units to electrical cutaneous stimulation during locomotion and during lifting, falling and landing. Experimental Brain Research, 58, 102-1 16.

Cortical Sensorimotor Linkages

193

Phillips, C.G. (1969). Motor apparatus of the baboon’s hand. Proceedings of the Royal Society, Series B, 173, 141-174. Requin, J., Riehle, A., & Seal, J. (1988). Neuronal activity and information processing in motor control: from stages to continuous flow. Biological Pychology, 26, 179-198. Riehle, A., & Requin, J. (1989). Monkey primary motor and premotor cortex: single-cell activity related to prior information about direction and extent of an intended movement. Journal of Neurophysiology, 61, 534-549. Roland, P.E. (1985). Cortical organization of voluntary behavior in man. Human Neurobiology, 4, 155-167. Sasaki, K. (1985). Cerebro-cerebellar interactions and organization of a fast and stable hand movement: cerebellar participation in voluntary movement and motor learning. In J. Bloedel, J. Dichgans & W.Precht (Eds.), Cerebellar Functions (pp. 70-85). Berlin: Springer-Verlag. Sawaguchi, T., Matsumura, M., & Kubota, K. (1989). Depth distribution of neuronal activity related to a visual reaction time task in the monkey prefrontal cortex. Journal of Neurophysiology, 61, 435-446. Seal, J., & Commenges, D. (1985). A quantitative analysis of stimulus- and movement-related responses in the posterior parietal cortex of the monkey. Experimental Brain Research, 58, 144-153. Seal, J., Gross, C., Doudet, D., & Bioulac, B. (1983). Instruction-related changes of neuronal activity in area 5 during a simple forearm movement in the monkey. Neuroscience Letters, 36, 145-150. Smith, A.M., Bourbonnais, D., & Blanchette, G. (1981). Interaction between forced grasping and a learned precision grip after ablation of the supplementary motor area. Brain Research, 222, 395-400. Tanji, J., Kurata, K., & Okano, K. (1985). The effect of cooling of the supplementary motor cortex and adjacent cortical areas. Experimental Brain Research, 60, 423-426. Towe, A.L., & Jabbur, S.J. (1961). Cortical inhibition of neurons in dorsal column nuclei of cat. Journal of Neurophysiology, 24, 488-498. Weinrich, M., Wise, S.P., & Mauritz, K.-H. (1984). A neurophysiological study of the premotor cortex in the rhesus monkey. Brain, 107 385-414. Yezierski, R.P., Gerhart, K.D., Schrock, B.J., & Willis, W.D. (1983). A further examination of effects of cortical stimulation on primate spinothalamic tract cells. Journal of Neurophysiology, 49, 424-441. Yuan, B., Morrow, T.J., & Casey, K.L. (1986). Corticofugal influences of S1 cortex on ventrobasal thalamic neurons in the awake rat. Journal of Neuroscience, 6, 361 1-3617.

This Page Intentionally Left Blank

VOLITIONALACTION, W.A. Hersh berger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

195

CHAPTER 8 CEREBRAL CORRELATES OF AUDITORY ATTENTION R. Naatanen One of the clearest, if not oldest, distinctions made within the attention literature is that between active and passive attention, a distinction proposed by William James a century ago (1890). Active attention refers to a selective state of our information-processing system (and to the ensuing selective processing) which we ourselves develop, according to our momentary interests and motives. Passive attention, in contrast, is initiated by environmental events, and this occurs irrespective of our will. Passive attention is even elicited when it is clearly harmful to the ongoing performance. Here, apparently, we are facing a mechanism of vital significance capable of generating high-priority interrupt signals that secure the eliciting stimulus's access to the limited-capacity system. These two complementary mechanisms central to information processing, conscious perception and experience, as well as to behavior, are currently intensely investigated in cognitive psychology and related fields. Despite this massive work, however, most centraI issues still await a solution. The best-known of these controversies involves the level of information processing at which attentional selection takes place in a typical selective-attention situation such as selective dichotic listening ("the filtering paradigm"; Kahneman & Treisman, 1984). The so-called earlyselection theories hold that in certain conditions, little or no semantic processing occurs to the unattended input, whereas the opposite type of theories, the late-selection theories, maintain that all sensory information is automatically processed to the semantic level. The latter type of theory describes attention only as selecting input to consciousness, memory, and response. Because behavioral research on attention has not been able to provide conclusive answers to central issues such as that mentioned above, despite their utilization of ingenious paradigms, more and more research interest has been directed to the underlying physiology. Physiological processes initiated by stimuli under different attentional instructions have

196

R Naatanen

been studied, the purpose being to map the attentional influences on these processes. It was hoped that this might enable construction of putative mechanisms of attentional selection operating between the stimulus and the response, hidden from behavioral attention research by the "Black Box". This optimism has been reinforced by the rapid development of new technology for cognitive brain research. In the present chapter, my purpose is to review some central lines of physiological research on human auditory attention, both passive and active attention. Most of this work was conducted by recording eventrelated brain potentials (ERP) from the scalp of healthy human subjects but important complementary information was provided by related magnetoencephalographic (MEG) and regional cerebral blood-flow (rCBF) work. ERPs or evoked potentials (EP) are discrete and minute electrical potentials which appear in the electroencephalogram (EEG). They are usually caused by and time-locked to sensory stimuli. These small changes in the EEG are normally obscured by much larger spontaneous brain waves and rhythms. However, if brief EEG epochs are summed and averaged over many presentations of the same stimulus, the EEG activity time-locked to the stimulus is enhanced whereas the randomly occurring spontaneous waves are reduced, leaving distinct ERP waveforms for study. The ERP consists of a sequence of positive and negative waves or peaks. Although these deflections in the waveform provide a convenient point for measurement, they are not necessarily generated by individual cerebral events. At any point in time, multiple cerebral processes may contribute to the ERP waveform. In this chapter, an ERP component will be taken to be "the contribution to the recorded waveform of a particular generator process, such as the activation of a localized area of cerebral cortex by a specific pattern of input .... Whereas the peaks and deflections of an EP can be directly measured from the averaged waveform, the components contributing to these peaks can usually be inferred only from the results of experimental manipulation" (Naatanen & Picton, 1987, p. 376; see also Donchin, Ritter, & McCallum, 1978). Thus, by recording ERPs, it appears that we might be able to follow some aspects of the processing initiated by a stimulus in the brain. Further, if we are able to disentangle the overlapping components of the ERP and to localize their sources, we could trace a substantial part of the spatio-temporal activation pattern associated with stimulus processing. Source localization on the basis of ERPs recorded in parallel from numerous scalp sites is, however, rather problematic, but nevertheless

Correlates of Auditory Attention

197

some relatively safe conclusions can be made when ERP information is interpreted in the light of the known brain anatomy and physiology (see Wood, et al., 1984). Recording stimulus-induced changes in the magnetic fields surrounding the head (produced by the brain's electrical activity) provides, in general, a better means for localizing event-related electrical activity in the brain (for reviews, see Hari, in press; Kaufman & Williamson, 1988), but due to certain limitations of the MEG, a parallel recording of the ERP and the MEG is advisable in many types of studies (see, e.g., Hari, Kaila, Katila, Tuomisto, & Varpula, 1982). Further, complementary information may be obtained by recording regional cerebral blood flow, or rCBF, and regional metabolic changes, the latter usually by means of positron emission tomography (PET). Thus, modem technology has opened ample opportunities for noninvasive research on the physiological basis of information processing as well as on attentional manifestations in, and effects on, these physiological events. Consequently, some of the physiology underlying human attention is already rather well known. The import of these data to cognitive theories of attention is, however, in most cases rather unclear, due to the lack of understanding of what the physiological processes measured mean in terms of actual information extraction, transfer, and use (Naatanen, 1988). In the next section, some central aspects of these physiological data, involving auditory attention, are reviewed and discussed with a view to clarifying not only the mechanism of auditory attention but also that mechanism's role in auditory information processing.

Passive Attention and Its PhysioIogy in Audition Passive attention was described above as a form of attention which is involuntarily caught by stimuli, rather than initiated by ourselves. Two broad classes of stimulus events can be separated here, again as previously recognized by William James (1890): (a) stimuli which elicit attention due to their physical properties, and (b) stimuli which elicit attention due to their psychological or semantic properties (i.e., what James called "derived" properties). As to these physical properties, the onset of a stimulus tends to capture attention, especially when the stimulus is delivered after a long silent interval. According to Newstead and Dennis (1979), it is impossible not to become aware of discrete stimuli occasionally presented to the otherwise "silent" unattended ear while listening to stimuli delivered to the opposite ear. The offset of a long-duration sound also is an attention-

198

R. Nadtanen

capturing event. Similarly, we tend to become aware of the occurrence of occasional changes in a repetitive or continuous stimulus in our acoustic environment. The effectiveness of a stimulus onset in catching our attention apparently depends on various factors such as the loudness and rise time of the stimulus as well as the length of the preceding silent interval. Similarly, a stimulus offset is, apparently, more attentioncatching, the louder the stimulus, the longer the duration, and the shorter the fall time. Now, the question arises whether the auditory ERP includes any components which are generated by mechanisms underlying shifts of attention to physical stimulus events. For instance, with regard to stimulus onsets, such a mechanism would be one which is particularly sensitive to stimulus onsets, especially to abrupt onsets of loud stimuli after long silent periods, but does not belong to neural mechanisms encoding any particular stimulus attribute such as frequency. Consider the various components comprising the auditory ERP. A discrete auditory stimulus, such as a brief tone pip, first elicits cochlear and brainstem potentials recordable from the scalp. These deflections are of very low amplitude, however, and must be averaged over many hundreds or even thousands of stimuli in order to be resolved clearly. The brainstem response consists of seven small deflections, all occurring within the first 10-12 ms from stimulus onset (Picton, Stapells, & Campbell, 1981; Starr & Don, 1988). These deflections probably reflect the arrival of sensory input to the various auditory nuclei in the cochlea and brainstem. The brainstem responses are followed by the middle-latency responses occurring from about 10 to 50 ms from stimulus onset. These comprise a sequence of low-amplitude, fast deflections, some of which are myogenic (muscular) in origin; those of cerebral origin are probably generated at the thalamic and cortical levels (Picton, Hillyard, Krausz, & Galambos, 1974). The arrival of the auditory input to the primary auditory cortex occurs, according to Vaughan and Arezzo (1988), at about 9 ms after stimulus onset and is reflected by the No wave of the middlelatency responses. According to Celesia (1976), auditory input arrives at the primary auditory cortex in 10-12 ms from stimulus onset. The middle-latency responses are followed by a large wave complex Nl-P2, with peak latencies at about 100 and 200 ms, respectively, usually preceded by a small P1 peaking at about 50 ms. These waves reach their maximal amplitudes at, or slightly anterior to, the vertex of the scalp (Picton et al., 1974). The N1 wave can be decomposed into at least three

Correlates of Auditory Attention

199

different "genuine" components which, according to Naatanen and Picton (1987), are as follows: (1) The frontocentral negativity generated by bilateral, vertically-oriented dipole planes in the auditory cortices on the superior aspect of the temporal lobe (Vaughan & Ritter, 1970); (2) the T complex, with a positivity peaking at 90-100 ms and a negativity at 140-150 ms, probably originating in the auditory association cortex on the lateral aspect of the superior temporal gyrus (Wolpaw & Penry, 1975); (3) the nonspecific component, maximal at the vertex, at a latency of 100 ms (Hari et al., 1982), with an unknown locus of origin (which probably is outside the auditory cortex, however). The above-listed ERP components are characterized as exogenous (Donchin et al., 1978) or "obligatory" (Naatanen, 1987) for they are mainly determined by physical and temporal features of stimulation and are, within a very wide range of variation in the organism's state, obligatorily elicited by appropriate stimulation. Most of these components seem to reflect the operations of mechanisms with other than attentional functions, judging from the very short recovery times of these components after a stimulus (e.g., Picton et al., 1981; Starr & Don, 1988). These functions probably belong to the processing of specific stimulus information, and this processing occurs both at the subcortical and cortical levels. The components forming the large N1 wave, however, at least the supratemporal component, might express the operation of a mechanism somehow related to passive attention. The recovery time of this component is of the order of 10 s, judging from MEG recordings (Makela, Hari, & Leinonen, 1988); MEG permits measurement of this component which is difficult to accomplish by means of the ERP. (The recovery time of the nonspecific N1 component is too long for a principal mechanism of passive attention elicited by stimulus onset; see Fruhstorfer et al., 1970.) Naatanen (1986) proposed that the function of the neuronal population that generates the supratemporal component is to summon attention to the eliciting auditory stimulus, whose specific features might be analyzed by neuronal events that are earlier and faster but subjectively "silent", and in part subcortical. It appears impossible that events of this processing level per se could underlie conscious perception or trigger attention in view of the large number of parallel processes occurring in the different sensory systems of different modalities. Moreover, in that case, concentrated task performance, probably even sleep, would be either impossible or very difficult, at least much more difficult than they are in fact. Therefore a separate mechanism is needed which (a) selects stimuli

200

R Naatanen

for entry into the limited-capacity system and thus protects it from sensory overload, and (b) controls the threshold of this entry. The brain mechanism generating the supratemporal N1 component might serve as such a mechanism within the auditory modality. As already mentioned, this suggestion is supported by the relatively long recovery times of the supratemporal component. This component is considerably larger for longer ISIs (Figure 1). It is very large in response to the first stimulus and attenuates thereafter, reaching some constant level by the 5th stimulus or so, as illustrated in Figure 1, depending on the inter-stimulus interval (ISI) used (see Ritter, Vaughan, & Costa, 1968). These facts roughly correspond to the subjective "obtrusiveness" of a stimulus when it is presented as the first stimulus of a stimulus block, or at long ISIs, and might also explain the attention-catching character of stimuli delivered at long ISIs to the ''unattended" ear in dichotic listening (Newstead & Dennis, 1979); this is also consistent with the fact that we usually perceive even repetitive stimuli consciously. Moreover, the supratemporal component does not fully disappear even with very short ISIs, which agrees with the fact that we tend to perceive even these stimuli perfectly well. Consistently with this, experiments conducted at the auditory detection threshold (e.g., Parasuraman, Richter, & Beatty, 1982) have shown that the detection of weak auditory stimuli correlates strongly with the N1 amplitude. Further, whereas the N1 is especially sensitive to transient aspects of stimulation, that is, to energy change per unit time during stimulus onsets and offsets (Naatanen & Picton, 1987; see also Graham, 1979)), it seems to be associated with no more specific aspect of perception than with mere detection. For example, a dissociation between N1 amplitude and subjective loudness has been demonstrated in several ways (for a review, see Naatanen & Picton, 1987). Butler (1972) in turn found a dissociation between the frequency-specificity of the neuronal populations generating the N1 and perceived pitch. Davis and Zerlin (1966) suggested that "the mechanisms that generate the potential and determine its magnitude do not lie on the direct path, so to speak, to psychological sensation but rather on a parallel path with other functions" (Davis & Zerlin, 1966, p. 116). Consistent with this, very similar N1 responses are elicited by a wide variety of acoustic stimuli such as clicks, tones, speech and animal sounds (Gaillard & Lawson, 1984; Woods & Elmasian, 1986; Hari, in press). Finally, the N1 wave is reduced as sleep becomes deeper and finally disappears at Stage 4, recovering to half the waking amplitude during REM sleep (see Figure 13 in Naatanen & Picton, 1987). Judging from the IS1 in that study, this N1

201

Correlates of Auditoy Attention

NON-SPECIFIC COMPONENT SUPRATEMPORAL COMPONENT

>

ST I MULUS

1

AMPLITUDE

2

3

4

5

,

LUl

NON-SPECIFIC

SUPRATEMPORAL COMPONENT I

I

5

10

IS1

A schematic illustration of the effect of the serial posi1.m of a stimulus in a stimulus train on the vertex N1 wave (top row) and on its two components, the nonspecific (second row) and the supratemporal (third row) ones. In the bottom of the figure, the differential IS1 sensitivity of these two components is schematically illustrated.

202

R Naatanen

wave was mainly composed of the supratemporal component. Thus, under the present hypothesis, the obtrusiveness of discrete auditory stimuli is reduced during sleep, which is consistent with behavioral data. Such a reduction may be important for maintaining the sleep state. In the foregoing, I have reviewed evidence that suggests, or is consistent with, the hypothesis that the supratemporal N1-generator process acts as an internal attentional trigger, choosing auditory stimuli for conscious perception. The evidence for the specific relation of N1 to energy onset and offset suggests that the underlying neuronal population detects a stimulus or its termination but does not provide specific perceptual contents. Consequently, the process informs that some stimulus is occurring without indicating what the stimulus is or what its precise features are. Although the generator is organized with at least some stimulus-specificity, the numerous dissociations between the N1 and the specific contents of perception suggest that perception is not based on the stimulus-specificity of this generator, but, rather, probably on that of earlier mechanisms of sensory analysis. The nonspecific component of the N1 wave is associated with a transient arousal burst caused by the stimulus (for a review, see Naatanen & Picton, 1987). Probably, this arousal burst, or the activation of the generator mechanism of the component itself, together with the associated proprioceptive and other internal feedback, also possesses an attentioncatching property. Because of the long refractory period (see Figure l), the component is elicited mainly at the onset of a stimulus sequence, and by subsequent stimuli only when ISIs are long. Hence this generator process cannot provide the principal mechanism for conscious perception of auditory stimuli. In reality, acoustic stimulation (e.g., speech) is often continuous rather than discrete, with changes occurring in the absence of any "empty" ISIs. Even minor frequency or intensity changes in a continuous tone elicit large N1 types of waves. Such a wave in response to a brief change from a continuous 600-Hz tone to 625 Hz is illustrated in Figure 2. Further, Arlinger, et al. (1982) recorded magnetic fields evoked by brief changes in the frequency of a continuous tone. Like the N1 type of response to tone onset, the response to a frequency glide was generated in the supratemporal plane. On the basis of this and further similar evidence, Naatanen (1988) suggested that the inherent attention-catching property of any changes in a continuous stimulus is based on the high sensitivity of the mechanism generating the supratemporal N1 component to these changes.

Correlates of Auditory Attention

EOG

Fz

cz

203

PZ

Figure 2. Across-subject ERPs to a transitory frequency modulation (onset at 'IS") of a 600-Hz continuous tone to 625 Hz for a duration of 50 ms. The mean interval between successive frequency modulations was 3.1 s. The trace length is 300 ms. (From Naatanen, Paavilainen, Alho, Reinikainen, & Sams, 1987. Reproduced with permission; see acknowledgements.)

ERP studies also suggest a mechanism for switching attention in response to a change in a repetitive homogeneous sequence of discrete stimuli. In these studies, the subject's attention is directed elsewhere while he/she is presented with a long sequence of auditory "standard" stimuli, one of which is occasionally replaced by a "deviant" stimulus. The two stimuli elicit quite similar N1 and P2 components which may, however, be slightly larger in response to the deviant stimuli, at least when the stimulus difference is not small. On the other hand, the deviant stimuli elicit a new component called the mismatch negativity (MMN) which is not elicited by the standard stimuli (Naatanen, Gaillard, & Mantysalo, 1978). The MMN can be derived from a difference wave obtained by subtracting the standard-stimulus ERP from that to the deviant stimulus. The negativity of the difference wave usually yields a rather good estimate of the MMN component because, as already mentioned, the N1 and P2 components elicited by the deviant stimuli are quite similar to those elicited by the standard stimuli when the magnitude of stimulus deviation is not large (Sams, Paavilainen, Alho, & Naatanen, 1985). This subtraction procedure, as well as a MMN in response to a change in tonal frequency, are illustrated in Figure 3 (Sams et al., 1985). Each stimulus block was composed of 80% standard stimuli of 1000 Hz and 20% deviant stimuli, differing in frequency from the standards, delivered in a random order and at constant ISIs of 1 s. The frequency

204

R Naatanen

of the deviant stimulus was, in separate blocks, either 1004, 1008, 1016, or Deu i ant FZ CZ 1032 Hz. The subject was instructed to ignore the auditory stimuli and to concentrate on reading a book. The N1 wave in response to the various deviant stimuli is similar to the one evoked by the standard stimulus, and does not vary in amplitude or latency with the degree DIFFERENCE of deviance. As shown by 1004 Hz the difference waveforms, a clear MMN is elicited by deviant stimuli with fre1008 HZ , quencies higher than 1008 Hz. (In a control experi1016 Hz Avr ment, a MMN was elicited when the subject instead of reading, performed a 1032 Hz difficult visual computer game.) In a separate condition, the subjects were found to be able to discriminate these stimuli from the standards. Even Figure 3. Top: Across-subject ERPs to 1000the 1008-Hz stimuli, which Hz standards (thin line) and to deviants (thick were found to be near the line) of different frequencies, as indicated on discrimination threshold, the left side. Bottom: The respective difference elicited a MMN although waves obtained by subtracting the standardstimulus ERP from the respective deviant-stimua very small one. The lus ERP. (From Sams et al., 1985). MMN peak is earlier for larger deviations in pitch. When the magnitude of deviation is further increased, the decreased MMN latency results in an increasing overlap between the MMN and N1 components (Naatanen & Gaillard, 1983).

-_ ---

a-=---

-L4

Correlates of Auditory Attention

205

The MMN seems to be elicited by any physical or temporal deviance in the auditory stimulus sequence. Even intensity and duration decrements elicit the MMN, and this occurs independently of attention (for a review, see Naatanen, in press). These and some further data indicate that the MMN cannot be explained by new afferent elements activated by deviants but not standards. Therefore the MMN appears to be generated when an input from a deviant stimulus encounters the neural representation or memory trace of the standard stimulus (see Naatanen, in press). The biological function of the process generating the MMN might be to initiate a sequence of further brain processes leading to an attention switch to a change in an unattended auditory input (Naatanen, 1985). Some of these further processes might be reflected by the P3a component of the ERP, a frontocentral positivity peaking at 250-300 ms (see Squires, Squires, & Hillyard, 1975). James (1890) and his followers assumed that an attention switch can also occur due to the semantic meaning of an unattended stimulus. This brings us into the middle of one of the central issues of attention research to date, that concerning the degree of automaticity of semantic processing mentioned. ERP data might be interpreted as suggesting that semantic processing does not have to be fully automatic (in the sense of being independent of attention). In the foregoing, we have examined ERP data suggesting attention-switching mechanisms which are frequently activated by certain physical events in the unattended input. Apparently, these mechanisms cause frequent momentary attention switches to this input (see Lyytinen, Blomberg, & Naatanen, 1989). According to MMN data, this input is fully processed with regard to its physical and temporal properties and stored for a few seconds in a memory system probably corresponding to cognitive psychologists' notion of sensory or precategorical memory. It is, presumably, this information store that is looked at by the attentional focus while momentarily caught by the "wrong" input, rather than the ongoing sensory processing per se. Such frequent attention switches to such fully processed representations of the physicaltemporal features of the auditory stimuli might well serve to introduce these representations into long-term memory and thus explain data interpreted in terms of attention-independent semantic processing (Naatanen, in press).

206

R. Naatanen

Selective Attention and its Physiology in Audition Research on ERP correlates of auditory selective attention was initially centered around the N1 wave. The first demonstration of a selective-attention effect on the N1 wave, which could not be questioned on methodological grounds, was the one reported by Hillyard, Hink, Schwent, & Picton, (1973) (for a review, see Naatanen, 1982). In Hillyard et al.'s dichotic-listening task with very short irregular ISIs, the left-ear tones were of a considerably higher pitch than the right-ear tones. Both sequences included occasional, randomly placed, slightly higher tones. The task was to count the deviants among the standards in a designated ear and to ignore all the input to the other ear. The vertex N1 deflection was larger in response to the attended than to the ignored stimuli. The authors regarded their "N1 effect" as an enhancement of the exogenous "N1 component" and suggested that it reflects Broadbent's (1971) stimulus-set mode of attention. Since this pioneering study, considerable research effort has been expended to clarify the conditions and limits of the N1 effect (for reviews, see Hillyard & Picton, 1979; Naatanen, 1982). Naatanen, et al. (1978), however, described a different kind of selective-attention effect on the ERP, which they called the processing negativity (PN). The effect was not produced by any exogenous ERP component but was rather a new component emerging during selective attention. The dichotic-listeningtask used differed from that of Hillyard et al. (1973) in having, among other things, a considerably longer and constant IS1 (800 ms). The peak amplitude of the N1 deflection was not affected, but the N1 peak was followed by a low-amplitude negative displacement of the ERP to the attended standards relative to the unattended standards. This displacement began at 150 ms, during the descending limb of the N1 deflection, and persisted for at least 500 ms. The authors proposed that the PN is an endogenous component generated by a cerebral mechanism different from that of the N1 component and, further, that even the N1 effect reported by Hillyard et al. (1973) might have been caused by a PN rather than by an intensification of the generator process of the N1 component. The considerably shorter ISIs used by Hillyard et a1.(1973) might have shortened the PN latency so that the PN overlapped the N1 component, causing an artificial increase in its measured amplitudes. Even to date, it is not entirely clear whether all ERP effects of auditory selective attention are due to the PN or whether an enhancement of some N1 component may also occur in some conditions (Woldorf

Correlates of Auditory Attention

207

et al., 1987; Hackley, Woldorff, & Hillyard, 1987; Naatanen, Teder, & Alho, in preparation). The PN has been verified by many studies (e.g., Okita, 1979; Parasuraman, 1980) and has also been observed by Hillyard and his colleagues (Hansen & Hillyard, 1980). Hansen and Hillyard's.data suggest that the onset latency of the attention effect depends on the difference between relevant and irrelevant stimuli. The subjects in this study were presented with a binaural sequence of equiprobable "high-pitch" and "lowpitch'' tones. The frequency of the high tones was, in separate blocks, either 350, 400, or 700 Hz, the low tones always being 300 Hz. Thus pitch separation between the two tones in a block was either 50, 100, or 400 Hz. In different blocks, either the low or high tones were designated as task-relevant, the subject's task being to discriminate occasional longerduration tones among the task-relevant tones. The right side of Figure 4 presents the attention effect called the "Nd" (negative difference) by the authors, obtained by subtracting the ERP to the unattended stimuli from that to the attended stimuli. With greater pitch separation, both onset and peak latencies of the Nd were shorter and the Nd duration longer. The late portion of the Nd had a topography anterior to that of the earlier portion, suggesting that there are in fact two partially overlapping components in the attention effect. Subsequent studies showed, among other things, that (a) the PN onset latency is shorter when the mean IS1 in studies using irregular ISIs is shorter (Parasuraman, 1980); (b) the PN is elicited even by complex stimuli, e.g., by speech sounds (Woods, Hillyard, & Hansen, 1984); (c) even (slowly) moving attended stimuli can elicit a PN (Okita, 1979); (d) at least some of the PN is, according to MEG recordings, generated in the supratemporal auditory cortex (Hari et al., in press; Arthur, Lewis, Medvick, & Flynn, 1989); (e) a few relevant stimuli must be delivered before the PN can be elicited (Hansen & Hillyard, 1988; see also Donald & Young, 1982). Naatanen (1982) proposed that the first PN component, probably generated in the auditory cortex, expresses selection of the to-be-attended stimuli among irrelevant stimuli. According to him, selective attention to one class of stimuli physically differing from the other stimuli is based on maintaining a neuronal representation of the physical feature(s) defining the relevant stimuli. This representation would be based on the presence of a corresponding stimulus representation in sensory memory, explaining why a few stimuli are needed before attention can be "tuned" to a particular stimulus. (For consistent behavioral results, see Treisman,

EJ 0

m

Low Tones

High Tones

y

50 I N

I

Attend minus Unattend

I

. .

................. . . ..

A

........

n

400 4c.

I 0

'......

xx)

-

Attend Una,ted ..............

Figure 4. Across-subject ERPs to standard tones at three different frequency separations in a binaural durationdiscrimination task. ERPs at left are to 300 Hz (low) tones and at center to 350, 400, or 700 Hz (high) tones, according to condition. ERPs to attended tones (solid lines) and unattended tones (dotted lines) are overlapped. Tracings at right are superimposed difference waves between ERPs to attended and unattended tones. (From Hansen & Hillyard, 1980).

kl

i -

L-r

Correlates of Auditory Attention

209

Squire, & Green, 1974). This voluntary neuronal representation, proposed to be located in the auditory cortex, was called the attentional trace. It was further suggested that the attentional trace is maintained by a frontallobe mechanism which might account for the later, frontal PN component. Some rCBF data of Roland (1981, 1982) support this proposal. During selective attention, each auditory input is, on this theory, compared with the attentional trace and, further, this comparison process lasts longer the more similar the stimulus is to that represented by the attentional trace. Thus the relevant stimulus generates the longest and largest PN but even irrelevant stimuli generate some PN. Hence selective attention in these kinds of situations would be based on a self-terminating matching or comparison process which selects, for further processing, stimuli meeting the criteria represented by the attentional trace and rejects all other stimuli. These predictions have recently been verified by Alho et al. (e.g., Alho, Sams, Paavilainen, & Naatanen, 1986; Alho, Tottola, Reinikainen, Sams, & Naatanen, 1987). Their results imply that the Nd obtained by subtracting the irrelevant-stimulus ERP from the relevant-stimulus ERP (see Figure 4) does not yield the whole PN, for the common initial PN elicited by both stimuli at the same onset latency is cancelled. This cancellation effect accounts for the later Nd onset with smaller separations (Figure 4), i.e., the later separation of the traces for the relevant and irrelevant stimuli. This later Nd onset is hence not due to a later PN onset to relevant stimuli. For details of the attentional-trace theory and results of further studies testing its predictions, see Naatanen (1982; and in press). In conclusion, ERP studies on auditory attention suggest that selective attention to a class of stimuli defined by some physical feature is realized via a matching type of process. This process expresses comparison of each auditory input with a voluntarily maintained representation of the physical features defining the relevant stimulus class. Further, in contrast to filtering or gating types of theories, the processing of physical features of irrelevant stimuli is not blocked or attenuated, but these stimuli are rejected from further processing following initial selection. Therefore these data tend to support early-selection rather than late-selection types of attention theory (see Naatanen, in press). The above-mentioned filtering or gating types of theories in turn would be supported by results suggesting that selective attention enhances exogenous ERP components. As already mentioned, such evidence should still be considered inconclusive with regard to the N1 component. Very

210

R. Niiutiinen

recent research (Woldorf et al., 1987; Hackley et al., 1987), however, suggests that under certain specific conditions, attended stimuli elicit a small positivity at a latency as short as 30-50 ms preceding the negative attention effect. The significance of this finding should be determined by future research.

Summary Physiological studies on human auditory attention were reviewed. These studies mainly involve event-related brain potentials (ERP) but some magnetoencephalographic (MEG) and regional cerebral blood-flow (rCBF) studies have also been performed. ERP studies on passive attention suggest cerebral mechanisms of attention switch to certain physical events in the unattended auditory input. These events include stimulus onsets and offsets as well as changes in continuous or discrete stimuli. MEG studies provide information on cerebral locations of these mechanisms. ERP studies on auditory selective attention reveal a specific ERP component, the processing negativity, which permits certain conclusions regarding the nature and mechanisms of attentional stimulus selection.

Acknowledgments The preparation of this article was supported by the Wissenschaftskolleg zu Berlin (Institute of Advanced Study Berlin) and the Academy of Finland. Figure 2 reproduced, with permission, from R. Naatanen, et al., (1987). "Interstimulus interval and the mismatch negativity. In C. Barber and T. Blum (Eds.), Evoked potentials 111: The third international evoked potentiah symposium. Stoneham, M A Butterworth Publishers.

References Alho, K., Sams, M., Paavilainen, P., & Naatanen, R. (1986). Small pitch separation and the selective-attention effect on the ERP. Psychophysiology, 23, 189-197. Alho, K., Tottola, K., Reinikainen, K., Sams, M., & Naatanen, R. (1987). Brain mechanisms of selective listening reflected by event-related potentials. Electroencephalography and clinical Neurophysiology, 68, 458-470.

Correlates of Auditory Attention

211

Arlinger, S., Elberling, C., Bak, C., Kofoed, B., Lebech, J., & Saermark, K. (1982). Cortical magnetic fields evoked by frequency glides of a continuous tone. Electroencephalography and clinical Neurophysiology, 54, 642-653. Arthur, D. L., Lewis, P. S., Medvick, P. A., & Flynn, E. R. (1989). A neuromagnetic study of selective auditory attention. Manuscript submitted for publication. Broadbent, D. E. (1971). Decision and Stress. New York: Academic Press. Celesia, G. G. (1976). Organization of auditory cortical areas in man. Brain, 99, 403-414. Davis, H., & Zerlin, S. (1966). Acoustic relations of the human vertex potential. Journal of Acoustical Society of America, 39, 109-116. Donald, M. W., & Young, M. J. (1982). The time course of selective neural tuning in auditory attention. Experimental Brain Research, 46, 357-367. Donchin, E:, Ritter, W., & McCallum, W. C. (1978). Cognitive psychophysiology: The endogenous components of the ERP. In E. Callaway, P. Tueting, & S. H. Koslow (Eds.), Event-related brain potentials in man (pp. 349-441). New York: Academic Press. Gaillard, A. W. K., & Lawson, E. A. (1984). Evoked potentials to consonantvowel syllables in a memory-scanning task. In R. Karrer, J. Cohen, & P. Tueting (Eds.), Brain and information: Event-relatedpotentials. Annals of the New York Academy of Sciences, 425, 204-209. Graham, F. K. (1979). Distinguishing among orienting, defence and startle reflexes. In H. D. Kimmel, E. H. van Olst, & J. F. Qrlebeke (Eds.), The orienting reflex in humans (pp. 137-167). Hillsdale, N.J.: Erlbaum. Hackley, S. A., Woldorff, M., & Hillyard, S. A. (1987). Combined use of microreflexes and event-related brain potentials as measures of auditory selective attention. Psychophysiology, 24, 632-647. Hansen, J. C., & Hillyard, S. A. (1980). Endogenous brain potentials associated with selective auditory attention. Electroencephalography and Clinical Neurophysiology, 49, 277-290. Hansen, J. C., & Hillyard, S. A. (1988). Temporal dynamics of human auditory selective attention. Psychophysiology, 25, 316-329. Hari, R. (In press). The neuromagnetic method in the study of human auditory cortex. In F. Grandori, G. L. Romani, & M. Hoke (Eds.), Advances in

Audiology. Hari, R., Hamalainen, M., Kaukoranta, E., Makela, J., Joutsiniemi, S.-L., & Tiihonen, J. (In press). Selective listening modifies activity of the human auditory cortex. Expenmental Brain Research.

212

R Naatanen

Hari, R., Kaila, K., Katila, T. Tuomisto, T., & Varpula, T. (1982). Interstimulus interval dependence of the auditory vertex response and its magnetic counterpart: Implications for their neural generation. Electroencephalography and clinical Neurophysiology, 54, 561-569. Hillyard, S. A., & Picton, T. W. (1979). Event-related brain potentials and selective information processing in man. In J. E. Desmedt (Ed.),

Cognitive components in cerebral event-related potentials and selective attention. Progress in Clinical Neurophysiology, 6 (pp. 1-52). Basel: Karger. Hillyard, S. A., Hink, R. F., Schwent, V. L., & Picton, T. W. (1973). Electrical signs of selective attention in the human brain. Science, 182, 177-180. James, W. (1890). The principles ofpsychology. New York: Holt. Kahneman, D., & Treisman, A. (1984). Changing views of attention and automaticity. In R. Parasuraman, & D. R. Davies (Eds.), Varieties of attention (pp. 29-61). London: Academic Press, . Kaufman, L., & Williamson, S. (1988). Recent developments in neuromagnetism: Implications for imaging. In G. Pfurtscheller, & F. H. Lopes da Silva (Eds.), Functional brain imaging (pp. 11-29). Bern: Hans Huber Publishers. Lyytinen, H., Blomberg, A-P., & Naatanen, R. (1989). Autonomic concomitants of event-related potentials in the auditory oddball paradigm. Submitted for publication. Makela, J. P., Hari, R., & Leinonen, L. (1988). Magnetic responses of the human auditory cortex to noisehquare wave transitions. Electroencephalography and clinical Neurophysiology, 69, 423-430. Naatanen, R. (1982). Processing negativity: An evoked-potential reflection of selective attention. Psychological Bulletin, 92, 605-640. Naatanen, R. (1985). Selective attention and stimulus processing: reflections in event-related potentials, magnetoencephalogram and regional cerebral blood flow. In M. I. Posner, & 0. S. Marin (Eds.), Attention and Performance X I (pp. 355-373). Hillsdale, N.J.: Erlbaum. Naatanen, R. (1986). The orienting response theory: An integration of informational and energetical aspects of brain function. In R. G. J. Hockey, A. W. K. Gaillard, & M. Coles (Eds.);Adaptation to stress and task demands: Energetical aspects of human infomtion processing (pp. 91 111). Dordrecht: Martinus Nijhoff. Naatanen, R. (1987). Event-related brain potentials in research of cognitive processes -- a classification of components. In E. van der Meer, & J. Hoffmann (Eds.), Knowledge Aided Information Processing (pp. 241273). Amsterdam: Elsevier. Naatanen, R. (1988). Implications of ERP data for theories of attention. Biological Psychology, 26, 117-163.

Correlates of Auditory Attention

213

Naatanen, R. (In press). Role of attention in auditory information processing revealed by event-related brain potentials. Behavioral and Brain

Sciences. Naatanen, R., & Gaillard, A. W. K. (1983). The N2 deflection of ERP and the orienting reflex. In A. W. K. Gaillard, & W. Ritter (Eds.), EEG correlates of information processing: Theoretical issues (pp. 119-141). Amsterdam: Elsevier. Naatanen, R., & Picton, T. W. (1987). The N1 wave of the human electric and magnetic response to sound: A review and an analysis of the component structure. Pvchophysiologv, 24, 375-425. Naatanen, R., Gaillard, A. W. K., & Mantysalo, S . (1978). Early selective attention effect on evoked potential reinterpreted. Actu Pychologica, 42, 313-329. Naatanen, R., Paavilainen, P., Alho, K., Reinikainen, K., & Sams, M. (1987). Inter-stimulus interval and the mismatch negativity. In C. Barber, & T. Blum (Eds.), Evoked potentials 111 (pp. 392-397). London: Butterworths. Naatanen, R., Teder, W., & Alho, K. (In preparation). Processing negativity and the "Nl effect": One or two selective-attention effects on the

auditory event-related brain potential. Newstead, S . E., & Dennis, I. (1979). Lexical and grammatical processing of unshadowed messages: A reexamination of the MacKay effect. Quarter& Journal of Experimental Psychology, 31, 477-488. Okita, T. (1979). Event-related potentials and selective attention to auditory stimuli varying in pitch and localization. Biological Psychology, 9, 271-284. Parasuraman, R. (1980). Effects of information processing demands on slow negative shift latencies and NlOO amplitude in selective and divided attention. Biological Psychology, 11, 217-233. Parasuraman, R., Richter, F., & Beatty, J. (1982). Detection and recognition: Concurrent processes in perception. Perception and Pychophysics, 31, 1-12. Picton, T. W., Hillyard, S . A., Krausz, H. I., & Galambos, R. (1974). Human auditory evoked potentials. I. Evaluation of components. Electroencephalography and clinical Neurophysiology, 36, 179-190. Picton, T. W., Stapells, D. R., & Campbell, K. B. (1981). Auditory evoked potentials from the human cochlea and brainstem. The Journal of Otolaryngology, 10, 1-41. Ritter, W., Vaughan, H. G., & Costa, L. D. (1968). Orienting and habituation to auditory stimuli. A study of short term changes in average evoked responses. Electroencephalography and clinical Neurophysiology, 25, 550-556.

214

R. Naatanen

Roland, P. E. (1981). Somatotopical tuning of postcentral gyrus during focal attention in man. A regional cerebral blood flow study. Journal of Neurophysiology, 46, 744-754. Roland, P. E. (1982). Cortical regulation of selective attention in man. A regional cerebral blood flow study. Journal of Neurophysiology, 48, 1059-1077. Sams, M., Paavilainen, P., Alho, K., & Naatanen, R. (1985). Auditory frequency discrimination and event-related potentials. Electroencephalographyand clinical Neurophysiology, 62, 437-448. Squires, N. K., Squires, K. C., & Hillyard, S. A. (1975). Two varieties of longlatency positive waves evoked by unpredictable auditory stimuli in man. Electroencephalography and clinical Neurophysiology, 38, 387-401. Starr, A., & Don, M. (1988). Brain potentials evoked by acoustic stimuli. In T.W. Picton (Ed.), Human event-related potentials. EEG handbook (revised series, Vol. 3) (pp. 97-157). Amsterdam: Elsevier. Treisman, A. M., Squire, R., & Green, J. (1974). Semantic processing in dichotic listening? A replication. Memory and Cognition, 2, 641-646. Vaughan, H. G., & Ritter, W. (1970). The sources of auditory evoked responses recorded from the human scalp. Electroencephalography and clinical Neurophysiology, 28, 360-367. Vaughan, H. G., & Arezzo, J. C. (1988). The neural basis of event-related potentials. In T. W. Picton (Ed.), Human event-relatedpotentials. EEG handbook (revised series, Vol. 3) (pp. 45-96). Amsterdam: Elsevier. Wolpaw, J. R., & Penry, J. K. (1975). A temporal component of the auditory evoked response. Electroencephalographyand clinical Neurophysiology, 39, 609-620. Wood, C. C., McCarthy, G., Squires, N. K., Vaughan, H. G., Woods, D. L., & McCallum, W. C. (1984). Anatomical and physiological substrates of event-related potentials. Two case studies. In R. Karrer, J. Cohen, & P. Tueting (Eds.), Brain and information: Event-related potentials. Annals of the New York Academy of Sciences, 425, (pp. 681-721). Volume in the N.Y. Academy Series of Publications. Woods, D. L., & Elmasian, R. (1986). The habituation of event-related potentials to speech sounds and tones. Electroencephalography and clinical Neurophysiology, 65, 447-459. Woods, D. L., Hillyard, S. A., & Hansen, J. C. (1984). Event-related brain potentials reveal similar mechanisms during selective listening and shadowing. Journal of Experimental Pychology: Human Perception and Pe$omurnce, 10, 761-777.

VOLITIONAL ACTION, W.A. Hershbevger (Editor) Elsevier Science Publishers B. V~(North-Holland), 1989

0

215

CHAPTER 9 THE PHYSIOLOGICAL STRESS OF THWARTED INTENTIONS Raymond P. Pavloski This chapter examines a conception of physiological stress derived from control theory. Physiological activity is modulated by neural processes that are tightly linked to the generation and control of behavior (e.g., Brener, 1987). Consequently, changes in physiological activity, including those changes commonly regarded as being stress related, necessarily involve the control mechanisms by which behavior itself is generated. In other words, physiological stress is a theoretical concept which derives its scientific meaning from its larger theoretical context, that is, from its position within a general theory of behavior. Only from within the context of a general theory of behavior can we precisely specify what is meant by the expression physiological stress, in the same way that something precise is meant by the terms stress and strain in the context of Newtonian mechanics (Gartenhaus, 1977, Chapters 1-12).

Defining Stress Requires a Theory of Behavior The general theory of behavior from which the present conception of physiological stress is derived is the theory that human beings are control systems employing negative feedback (for a detailed control-system analysis of human behavior, see Powers, 1973a, 1973b). A negative-feedback control system produces a behavior reliably by varying its output so that the monitored value of a controlled variable (a representation of what is--an input to the system) matches the system’s reference level (a representation of what is intended-the value of the reference signal for the system). Both artificial control systems and those natural systems that have been investigated maintain a near-zero difference between these two representations over time (see review in Pavloski, in press). A thwarted intention exists when the two representations do not closely match. Physiological stress is defined here as the physiological consequence of thwarted intentions; that is, of non-zero error in the operation of negativefeedback control systems. Defined in this way, physiological stress appears

216

Raymond P. Pavloski

to be related to the psychophysiologist's concept of physiological reactivity-enhancement of a physiological activity that would in certain circumstances have functional behavioral significance. Two examples are a pounding heart and sweaty palms. The pounding heart refers to increased cardiac performance that would meet the metabolic demands of increased striate muscle activity, but is metabolically inappropriate in an individual not experiencing any increased metabolic load (Brener, 1987; Obrist, 1981; Sherwood, Allen, Obrist, & Langer, 1986). Optimal activity of the eccrine sweat glands has been proposed to aid in manipulation by increasing friction, to increase tactile discrimination, and to make it more difficult to tear the skin, but is inappropriate when excessive and in the absence of hand-environment interactions (Edelberg, 1972). I shall concentrate on cardiovascular reactivity, which has been the focus of my research (e.g., Pavloski, 1988).

The Control-System Analysis of Person-Environment Transactions: Intentional Behavior Motor Equivalence is a Helpfil Observation The problem of motor equivalence posed by Lashley (1930) provides a nice point of introduction to a control-system mode of analysis. Lashley viewed motor equivalence as a problem. He wondered what type of mechanism the nervous system embodies that allows it, on different occasions of the same behavior, to provide suitably different outputs in order to achieve the same behavioral result. That is, how can a consistent behavior be produced when that behavior is only partially determined by the outputs of the nervous system, and is also partially determined by outside influences that (a) are different on different occasions of the behavior's occurrence, (b) are not necessarily influenced by the outputs of the nervous system, and (c) may not even be represented in the nervous system? In other words, what type of mechanism must the nervous system embody so that, without knowing beforehand the particular time course and constellation of outside influences, the nervous system produces the necessary, properly timed, outputs that combine with those outside influences to produce a given behavior and to produce it reliably? The phenomenon of motor equivalence poses a significant theoretical problem, one that has been recognized by cognitive psychologists (Bruner & Bruner, 1968, pp. 251-255), by behaviorists (Brunswick, 1952), and by psychologists studying motor behavior (Turvey, 1977, pp. 2 15 -216). The phenomenon of motor equivalence is clearly evident in our experiences of what might be called "high-level" behaviors. Consider what occurs

Stress of Thwarted Intentions

217

when we say that a woman leads or facilitates a group discussion. The behavior in this case is the discussion. Clearly, the behavior is the result of the actions of all of the participants, and not just those of the leader. Yet, a "good" leader or facilitator reliably sees to it that a good discussion occurs. Furthermore, she does so in very different ways on different occasions, and consistently good discussions result. The same phenomenon occurs in what might be called "low-order" behaviors. Consider what occurs in creating and maintaining a particular position of a limb--an arm, say. The nervous system outputs volleys of motor neuron impulses that arrive at neuromuscular junctions. The volley of motor neuron impulses is the only thing that the nervous system can be sure will occur (see Milner, 1970, pp. 59-67). A given volley of impulses will result in a release of neurotransmitter that will change from occasion to occasion; a given amount of neurotransmitter will result in different amounts of change in muscle tension from occasion to occasion that depend on the initial length and state of fatigue of the muscle fibers; and the change in position of the limb will vary from occasion to occasion with a given change in tension, depending on the actions of other muscles and on acceleratory forces that depend on dynamic changes in posture (McMahon, 1984). Despite the existence of influences that are uncontrolled by the nervous system, some of which are not even represented in the nervous system, it is clearly possible and in fact trivially easy to produce a certain limb position reliably. Finally, consider the behavior of keeping a motor vehicle centered on a road lane. Even if we assume that it is no problem to produce a given motion of the steering wheel, the effects of this motion combine with the influences of roadbed tilt, of acceleratory forces that change with the motion of the vehicle, of wind forces, and of forces produced by imperfections in the road to determine the actual position of the vehicle. It is commonplace to observe a driver continuously moving the steering wheel from side to side in an apparently random fashion that produces just those effects necessary, in combination with the other influences, to keep the car centered in the appropriate lane. Those who would "explain" these behaviors by appealing to operant conditioning, to maturation, to skill-learning, or to schema-driven outputs that are modified on the basis of existing inputs should become aware of the physics of behavior. Such an awareness provides a sobering perspective on the nature of such "explanations." An awareness of the mechanics of even the simplest movements (e.g., Hobbie, 1978, pp. 252-255; Turvey, 1977) reveals that such statements merely beg the question of the mechanism that produces the behavior. Physics describes the environment of the nervous

218

Raymond P. Pavloski

system. It shows us how the physical properties of the components of an organism and its surrounds are described as functions of time, and it explains how physical quantities affect one another. Explaining behavior means understanding what physical quantities must affect and be affected by the nervous system, and the forms of the relationships involved, so that behavior is produced. Motor equivalence actually makes this understanding easier, since it indicates that the mechanism involved must reliably produce certain effects in this physical world, even though at any moment in time those effects are only partially the result of the outputs of the nervous system. The observation of motor equivalence thereby puts stringent constraints on the type of theoretical mechanism or model deserving serious consideration. Control theory suggests that a single mechanism underlies all of the above examples, as well as countless other instances of motor equivalence. This mechanism is called a control system. The mechanism comprises one or more negative-feedback loops, and operates as follows.

The Continuous Realization of Intention: Near-Zero Error as the Hallmark of Control-System Operation Figure 1 shows the essential elements of a negative-feedback control system and its immediate environment. Two classes of time-dependent quantities are shown in this diagram. The physical quantities of the environment are called variables, while those within the system are called signals. Both are, in general, continuous functions of time. Beginning at the input end of the system, physical energy is transmitted from the controlled variable to the control system according to physical laws, and is transduced by an input function into a perceptual signal. The timedependent difference between the perceptual signal and a reference signal, generated outside the feedback loop, is called error. An output function transduces the error signal into an oytput variable. We are now back in the system’s environment. The feedback function represents those physical properties of the environment that determine how the output variable affects the controlled variable, again as a function of time. The controlled variable is, in general, also affected by influences that are independent of the system. These influences together provide a disturbance, an independent variable that affects the controlled variable according to physical properties of the environment; this is depicted in Figure 1 as the disturbance function, The system shown cannot affect these influences, and they are not even represented within the simple control system depicted here (i.e., there is no mental, neural, or other representation of them within the system).

219

Stress of Thwarted Intentions

Reference Signal

Function

Function Muscle

Physical of Muscle Tension

Controlled Condition

Influences on Controlled Condltlon

Figure 1. General model of a negative-feedback control system and its irnmediate environment. Reprinted with permission of the publisher (The Society for Physiological Research, Copyright 1989) from "A control system approach to cardiovascular reactivity: Behavioral models that behave," by R. Pavloski, in press, Psychophysiology; adapted with permission from William T. Powers, Behavior: The Control of Perception (New York Aldine-DeGruyter). Copyright 0 1973 by William T. Powers. @

It is difficult to grasp intuitively how a control system operates, because our intuition readily neglects physical time. Physics tells us that it takes time for any physical quantity to affect a second quantity, and this means that physical time must be properly included in any adequate description of the

220

Raymond P. Pavloski

system’s operation. A differential equation is required to describe the timedependent cause-effect relationship between any two physical quantities (variables and signals) whose values are continuous functions of time. An intuitive grasp is also likely to be based on a sequential description of the system’s operation. But a sequential description cannot be correct, since both members of any given pair of quantities in the loop are continuously exerting influences on each other. In pictorial terms, the arrows in Figure 1 depict physical relationships that are always in effect--for example, the effect of the value of the controlled variable on the output variable is not suspended temporarily while the value of the output exerts an effect on the controlled variable. The output variable continues influencing the controlled variable at the same time as the controlled variable is influencing the output. This is not mysterious as some believe (Bandura, 1983); the value of the controlled variable at time t is a function of the value of the output variable, and the reverse is also true. This simply shows that a simultaneous pair of equations is required to describe the interactions between the system and its environment. We can choose two quantities in the loop (the output and controlled variables, say), and write 1 equation that describes the output as a function of the controlled variable, and a second that describes the controlled variable as a function of the output. These equations can then be solved for either the controlled variable or the output variable as a function of time. Certain choices for the functions within the system will keep the values of the time-dependent perceptual and reference signals approximately equal over time. A system with such functions produces a continuously varying output variable whose continuously varying effects on the controlled variable almost exactly cancel any outside influences on the controlled variable that would cause the perceptual and reference signals to differ in value. Negativefeedback control systems embody such functions and thereby maintain the error signal near zero. If we call the reference level (value of the reference signal) the intention of the control system, then it follows that by cancelling the effects of influences on the controlled variable that would produce deviations between the values of the perceptual and reference signals, the system continually realizes its intention. The intention will in general vary over time, and its variation can result in the production of behavior. For example, changes in the value of a certain reference signal over time can have effects on a controlled variable whose value indexes the changing length of a muscle so that an object is picked up. Changes in the value of another reference

Stress of Thwarted Intentions

221

signal can have effects on another controlled variable whose value indexes progress in a group discussion.

Hierarchies of Control Systems-A view of Human Performance I have called the maintenance of error near zero a hallmark of controlsystem operation. It adds to the view of intentional behavior the notion that those intentions are satisfied in a quantitative sense over time. A hallmark of human behavior is that many intentions are being satisfied simultaneously, some of them by systems operating on one and the same time scale, and others by systems operating at different time scales. Consider a subject who is complying with an experimenter’s request to maintain a cursor in the center of a video screen by manipulating the position of a stick connected to a potentiometer while a computer simultaneously influences the position of the cursor with a disturbance that varies randomly over time. A great many intentions can be discerned here. The subject is intentionally maintaining the position of the cursor at a reference level by changing the position of the stick in just the manner required to cancel the effects of the unseen and unpredictable disturbance. Intentions operating on a faster time scale can be seen by asking how the cursor-position control system keeps the cursor in the center of the screen. The subject does this by moving a stick. Might not the position of the stick be a controlled variable, with its reference signal provided by the output of the cursor-position control system? And how does the stick control system achieve a given stick position--are muscle tensions controlled variables with reference signals provided by the stick control system? Clearly, the muscle-tension control systems must operate on a time scale faster than the stick control system, which operates on a scale faster than the cursor-position control system. If this were not so, then non-zero error in stick position, for example, would be counteracted not only by the stick system but also by the cursor control system; the reference to the stick control system would be modified even though its operation was faultless, and the hierarchy would be a dynamically unstable system, creating its own errors. More slowly operating systems can be found by asking why the subject is controlling the position of the cursor. Where does the reference signal for cursor-position control originate? The subject is controlling the position of the cursor because his professor asked him to. The cursor-position control system receives a reference signal from a higher order system, or perhaps a reference signal that is a weighted combination of the outputs from many higher order systems. These higher order systems control their input signals

222

Raymond P. Pavloski

by setting values for reference signals such as the one to the cursor position system. This process of asking, how and why, can be used to move up and down a hierarchy of control systems, as Powers (1973b) has done. It results in the view of a human being as continually satisfying intentions at many levels by controlling variables on different time scales related to different levels of intention.

Physiological Stress as the Result of Thwarted Intentions The Control-System-Error Theory of Cardiovascular Reactivity A proposal for a connection between the operation of human control systems and physiological activity follows easily once the basic principles of control-system operation are understood. Consider the problem of explaining how the activity of the heart is related to the production of behavior. We have evidence that cardiac performance is (appropriately) integrated with the metabolic demands of behavior under a variety of experimental conditions, and is (inappropriately) greater than metabolic demands warrant under others (Brener, 1987; Obrist, 1981; Sherwood et al., 1986). How might the production of behavior be related to both appropriately and inappropriately enhanced cardiac activity? The control-system-error theory of cardiovascular reactivity states that error in the operation of behavioral control systems is the modulator of cardiac performance in both types of situation. In the case of metabolicallyappropriate increases, we might suppose that error departs significantly from the zero value that it would have if control were perfect, and that it does so because the physical load placed on the muscles that form the output function of the control system is near the capacity of those muscles. Since we are assuming that we have convincing evidence that the behavior is produced by the operation of a control system, and since skeletal muscles form the output functions of such systems, we expect muscular effort to increase as the system tries to reduce the error. We know that cardiac performance increases in such situations (e.g., Brener, 1987), but are in need of a mechanism to link behavior to the increased cardiac performance. From the perspective provided by control theory, we are led to hypothesize that it is the error in the operation of the control system that signals the increased cardiac performance that we observe. Now consider another situation in which error departs significantly from zero, and in which the error signal is not converted into muscular efforts of proportional magnitude. This might occur for a variety of reasons. For

Stress of Thwarted Intentions

223

instance, it will occur if the rate of change of environmental influences on the controlled variable exceeds the speed of response of the control system (see Figure 2). In such a situation error will be changing faster than the system can respond. Other reasons are explored below.

Initial Tests of the Control System Error Theory My students and I have carried out several initial tests of the control system error theory. The results of completed experiments are summarized here. Rotated I/isual Image Task. We have carried out 1 experiment using this procedure (Pavloski, 1988). Subjects attempt to trace a line drawing on a sheet of paper while viewing a video image of the drawing surface (showing the sheet of paper with the drawing and the subject’s hand holding a pen). Control- system error can be increased from normal values near zero by rotating this visual image through an angle about the central axis of the lens of the video camera. A group of 27 subjects traced line drawings under 7 different angles of rotation, with the expectation that at least some of these would lead to large error values. Based on previous research using a similar procedure (Smith & Smith, 1962), it was expected that there would be large individual differences in control-system-error values that would permit an initial between-subjects test of the control-system-error hypothesis. Since we had available a group of subjects participating in a study of the relationship of cardiorespiratory fitness to heart rate (HR) reactivity (Arbitell & Pavloski, 1987), we were able to determine the relationship of both fitness and error to HR reactivity. Wide ranges of fitness and HR reactivity were obtained. Maximum oxygen consumption (VO, Max) (treadmill test) ranged from 35.5 ml/min/kg (poor) to 62.3 ml/min/kg (very fit), and HR reactivity ranged from 2 to 58 beats per minute. The rotation of the image was associated with increased error [F(6,138) = 16.99, p < .OOOS] and with increases in HR from baseline values [F(6,138) =7.19, p < .0005]. Multiple regression analysis with error, VO, Max and their interaction entered in that order revealed a significant between-subjects correlation between error and HR reactivity [r=.54, F(1,22)=9.20,pc.007]. VO, Max did not account for a significant proportion of the variance in HR reactivity, and was unrelated to error @>.05). Cursor-Position Control Tusk. Two subsequent experiments (Pavloski, 1988) employed a computer-controlled task in which a subject is asked to keep a cursor at a particular position on a video display. The position of the cursor is determined by the time-dependent values of two variables: the position of a handle that can be moved in 1 plane by the subject, and a

224

Raymond P. Pavloski

Figure 2. (to the right) Plots of slow- (a), medium- (b), and fast(c and d) rate-of-change disturbance functions shown together with the representative performances of a subject instructed to keep the cursor at the center of the screen. The subject moves the handle in such a way as to counteract the effect of the disturbance to cursor position. Both disturbance function and handle position are expressed in terms of their independent effects on cursor position. Error is the deviation of the cursor from the center of the screen. If the disturbance function and handle position were symmetric, error would remain at zero. When the disturbance’s rate of change is fast (plot c), error is so large that it must be shown separately (plot d) for clarity of presentation.

disturbance function produced by smoothing the output of a random number generator over time. Performance in this task has been simulated with great accuracy by control- system models (Marken, 1986; Powers, 1978). It was expected that increasing the rate of change of disturbance would increase control-system error and, by hypothesis, H R reactivity. In two experiments employing a total of 60 subjects, each subject was exposed to 5 final test trials at each of 3 rates of change of disturbance (slow, medium, fast). Each of 3 groups of subjects received 0, 1, or 2 15-minute training sessions prior to the test session. The slope and intercept of the regression line relating handle position to disturbance value for the test-session trials showed that subjects controlled cursor position as requested during the slow- and medium-rate-of-change trials, but may not have done so for the fast trials. Training had no significant effects on performance or reactivity. The rate of change of disturbance had a highly significant effect (for both experiments combined) on error, [F(2,114)=650.7, p<.0001] and on H R [F(2,114)=20.16, p<.0005]. HR increased as predicted from the slow to the medium and fast, but not from medium to fast trials (but there is no evidence that subjects controlled position as requested during the fast trials). Handle-potentiometer movement increased significantly both from slow to medium and from medium to fast trials, but the amount of movement (3.99, 7.17, and 9.97 full-scale potentiometer deflections per 1-minute trial, respectively) is too small to account for the HR data. For both experiments (60 subjects), the mean error during each interbeat interval (IBI) predicts the duration of the subsequent IBI

0

?

c

T

0

m

.

0

o

?

'

?

m

0

0

Stress of Thwarted Intentions

c

0

225

226

Figure 3. Representative graphs of mean IBI duration and standard deviations plotted against the category of the error on the previous beat (category 1: 0-100 screen units deviation from center; category 2: 101-200 screen units deviation; and so on). The number of beats on which each mean IBI is based is shown next to each data point. The left column shows 5 subjects’ graphs judged as fitting the predicted error-IBI relationship. The right column shows 5 graphs judged as not fitting the predicted error-IBI relationship.

Raymond P. Pavloski

730

806

t

i

1I’

8861

601

675

676

659

688

1

2

3

4

5

\ 43

3

1

831

1

a9

14

4

4

659

T E

Stress of Thwarted Intentions

227

[F(3,177)=13.38, pc.0005],with high error followed by shorter. IBIS. The 19/60 subjects not showing this relationship had significantly higher HRs on slow disturbance trials than subjects showing the relationship [t(58)=-2.31, pC.051, suggesting a ceiling effect. Figure 3 shows representative error-IBI plots, both for those subjects who show the relationship and those who do not.

An Experimental Approach to the Problem of Individual Diflerences in

Reactivity Research on cardiovascular reactivity has consistently shown the importance of individual differences in reactivity (e.g., Krantz & Manuck, 1984). For example, although certain tasks (called active coping tasks) result in greater reactivity than others (passive coping tasks), individuals who show reactivity greater than the group’s median value on the active task actually show greater reactivity on the passive task than do the remaining individuals on the active task (Obrist et al., 1983). The control system error theory holds that total control system error modulates cardiac performance. We must be careful not to equate this error with the error found in a particular control system involved in a task. For example, two subjects who show very similar error in controlling the position of a cursor may nevertheless have different total error in a hierarchy of control systems if they are doing the task for different reasons. A subject who is participating merely to fulfill a course requirement may show little reactivity, while another subject who is treating the task as a test of competence and as having an effect on university standing may show much greater reactivity. This suggests the hypothesis that reactivity increases with the number of control systems involved in a task. It is possible to conduct an experiment to test this hypothesis. It is based on the principle that control of a visual-spatial category (e.g., face) requires control of spatial relationships (distances), and that control of distance requires control of the positions of the elements that are in the spatial relationship considered. Thus, exactly the same behavior required to control positions over time can involve 1, 2, or 3 levels of control systems in a hierarchy. Subjects controlling distance who have the same position error as those controlling only position should have greater total error and reactivity than the latter, and subjects controlling category who have the same position error should have greater total error and reactivity than both of the former.

228

Raymond P. Pavloski

Emotion and Imagination in Living Control Systems Emotion Mandler (1984) concluded from an extensive review of research on emotion that there are two aspects to emotion, the first varying in intensity and the second varying in quality. The first comprises enhanced physiological activity of the viscera, triggered by the interruption of schema-driven actions and discrepancies in schema-based perception; this provides the intensity aspect of emotion. The quality of the emotion, seen by Mandler as the second aspect, is determined by meaning analysis--a hypothetical cognitive process that is said to involve our awareness of where we are, what we are doing, and why we are doing it. For example, an interruption of one's dinner schema provoked by an unexpected telephone call will lead directly to visceral arousal, providing emotional intensity. The quality of the emotional experience might be related to the identity of the caller (e.e., salesman or long-unheard-from friend), and is determined by the cognitive process (meaning analysis) that results in the contents of consciousness during the conversation. This is not a labeling or attributional explanation; Mandler (1984) is speculating about a mechanism responsible for emotional experience. The control-system mode of analysis provides a precise functional basis or mechanism for Mandler's cognitive speculations. For instance, what Mandler describes as interruptions of schema-driven actions and discrepancies of schema-based perceptions can, arguably, be defined operationally as controlsystem error, thus allowing precise quantification of this variable. We may similarly operationally define such putative psychological stressors as "effort" (Light, 1981), "hostility" (Dembroski, MacDougall, Shields, Pelitto, & Lushene, 1978), and the like (Harrell, 1980). The control system mode of analysis also suggests that the qualitative aspect of an emotion should be classified in terms of the error-inducing environmental disturbance, expressed either in terms of the disturbance itself or in terms of the compensatory output required to offset it. That is, it is precisely consistent with control theory to classify a state of enhanced physiological activity as being fear or anger in terms of the direction in which one is running, that is, away or toward the proverbial bear. Fear can turn to anger literally as quickly as one can change one's mind (reference signal), because the disturbance at any given moment is defined in terms of the control system's momentary intention. A man running in fear for his life may instantly turn in an angv rage upon a bear which menaces his child. The emotion changes, but the physiological arousal is not diminished because the

Stress of Thwarted Intentions

229

bear serves as a disturbance for either or both controlled variables: the man's control of his own welfare, and/or his control of the welfare of his child.

Imagination Powers (1973b) has suggested that the nervous system may be able to try out cognitively its various control loops by a process of imagination that short circuits both the muscles and the environment. The implied purpose of such dry runs would be to estimate the error signals likely to be encountered in an actual overt action. Irrespective of whether such putative error signals are considered cognitive or neural, to the degree that they are error signals, they may be expected to modulate cardiac performance and other indices of stress. Indeed, there is evidence that imagination does, in fact, have such effects. Much of the data on physiological reactivity demonstrates the existence of reactivity in the absence of environmental manipulation (Krantz & Manuck, 1984). In his doctoral dissertation, May (1987) found large increases in heart rate and systolic blood pressure in hypertensive adults who engaged in imagined episodes of interaction with the environment that they had previously indicated were "hassles." It is possible that experimental techniques developed by experimental cognitive psychologists to test imaginal actions (e.g., mental rotations, as described by Stillings et al., 1987, pp. 40-48) can also be used to explore the applicability of the control-system-errortheory to reactivity. For example, the research on visual representations has shown that subjects probably answer questions about familiar objects around the house (e.g., How many windows are in your house?) by modelling, in imagination, an environmental interaction that would provide the information requested (e.g., taking an imagined walk around the house). If subjects were also to imagine disturbances (e.g., an impassable hallway, wet with new wax), perhaps they would evince error induced stress in the form of increased cardiac performance, and the like.

Summary Control theory provides a precise, operationally defined conception of physiological stress that affords quantitative measurements. Control theory not only explains the generation of overt behavior, including the problem of motor equivalence; it also accounts for the causal links between overt and covert behavior, including the link we call stress. According to the controlsystem-error theory of reactivity, enhanced physiological activity is the result of frustrated intentions in the form of total system error. Finally, control

230

Raymond P. Pavloski

theory accounts for our cognitive interpretations of these visceral effects as emotions classified in terms of environmental circumstance. The advantages of the control-system mode of analysis can be seen clearly by comparing the control-system-error theory of reactivity with a typical 'kompeting" hypothesis, such as the notion that reactivity increases with the amount of subjective frustration involved in a task (see the discussion in Pavloski, in press for a thorough comparison of the control-system-error theory and alternative positions on reactivity). The advantage of the controlsystem-error theory is its testability. Hypotheses that relate reactivity to fitness and other such measurable variables are administrative. They can be used to provide advice to people and to calculate the probability of a given magnitude of reactivity on an empirical basis, even in the absence of any understanding of reactivity per se. In contrast, the control-system mode of analysis offers the possibility of actually understanding reactivity, that is, understanding how reactivity is related to the mechanisms of behavior and thought.

References Arbitell, M., & Pavloski, R. (1987). Cardiorespiratory fitness does not predict cardiovascular reactivity. Paper presented at the annual meetings of the American Psychological Association, New York, N.Y. Bandura, A. (1983). Temporal dynamics and decomposition of reciprocal determinism: A reply to Phillips and Orton. Psychological Review, 90, 166-170. Brener, J. (1987). Behavioural energetics: Some effects of uncertainty on the mobilization and distribution of energy. Psychophysiology, 24, 499-512. Bruner, J. S., & Bruner, B. M. (1968). On voluntary action and its hierarchical structure. International Journal of Psychology, 3(4), 239-255. Brunswick, E. (1952). The conceptualframework ofpsychology. Chicago: University of Chicago Press. Dembroski, T. M., MacDougall, J. M., Shields, J. L., Pelitto, J., & Lushene, R. (1978). Components of the Type A coronary-prone behavior pattern and cardiovascular responses to psychomotor performance challenge. Journal of Behavioral Medicine, I, 159-175. Edelberg, R. (1972). Electrical activity of the skin: Its measurement and uses in psychophysiology. In N. S. Greenfield & R. A. Sternbach (Eds.), Handbook of psychophysiology. New York: Holt, Rinehart and Winston. Gartenhaus, S. (1977). Physics: Basic principles. (Combined edition). New York: Holt, Rinehart and Winston. Harrell, J. P. (1980). Psychological factors and hypertension: A status report. Psychological Bulletin, 87, 482-501.

Stress of Thwarted Intentions

231

Hobbie, R. K. (1978). Intermediate physics for medicine and biology. New York: John Wiley & Sons. Krantz, D.S., & Manuck, S. B. (1984). Acute psychophysiologic reactivity and risk of cardiovascular disease: A review and methodologic critique. Psychological Bulletin, 96, 435-464. Lashley, K. S. (1930). Basic neural mechanisms in behavior. Psychological Review, 37 1-24. Light, K. C. (1981). Cardiovascular responses to effortful active coping: Implications for the role of stress in hypertension development. Psychophysiology, 18, 216-225. Mandler, G. (1984). Mind and body: Psychology of emotion and stress. New York: W. W. Norton & Co. Marken, R. (1986). Perceptual organization of behavior: A hierarchical control model of coordinated action. Journal of Experimental Psychology: Human Perception and PerJomnce, 12, 267-276. May, R. (1987). Cognitive appraisal and blood pressure variation among essential hypertensives: A preliminary investigation. Unpublished doctoral dissertation, Indiana University of Pennsylvania. McMahon, T. A. (1984). Muscles, reflexes, and locomotion. Princeton, N.J.: Princeton University Press. Milner, P. M. (1970). Physiological psychology. New York Holt, Rinehart and Winston. Obrist, P. A. (1981). Cardiovascular psychophysioloay:A perspective. New York: Plenum. Obrist, P. A., Langer, A. W., Grignolo, A., Light, K. C., Hastrup, J. L., McCubbin, J. A., Koepke, J. P., & Pollack, M. H. (1983). Behavioral cardiac interactions in hypertension. In D. S. Krantz, A. Baurn, & J. G. Singer (Eds.), Handbook of psychology and health: Cardiovascular disorders and behaviors (Vol. 3, pp. 199-226). New Jersey: Lawrence Erlbaum Associates, InC. Pavloski, R. P. (1988). Error in behavioral control system operation is a determinant of heart rate reactivity. Presented at the Annual Meetings of the Society of Behavioral Medicine, April 27-30, Boston, MA. Pavloski, R. P. (In press). A control system approach to cardiovascular reactivity: Behavioral models that behave. PsychophysioZogy. Powers, W. T. (1973a). Feedback: Beyond Behaviorism. Science, 179, 351-356. Powers, W. T. (1973b). Behavior: The control of perception. Chicago: Aldine. Powers, W. T. (1978). Quantitative analysis of purposive systems: Some spadework at the foundations of scientific psychology. Psychological Review, 85, 417-435.

232

Raymond P. Pavloski

Sherwood, A., Allen, M. T., Obrist, P. A., & Langer, A. W. (1986). Evaluation of beta-adrenergic influences on cardiovascular and metabolic adjustments to physical and psychological stress. PqchophysioZogy, 17, 202-207. Smith, K. U. & Smith, W. M. (1962). Perception and motion: An analysis of space-stiuchred behavior. Philadelphia: W. B. Saunders Co. Stillings, N. A., Feinstein, M. H., Garfield, J. L., Rissland, E. L., Rosenbaum, D. A., Weisler, S. E., & Baker-Ward, L. (1987). Cognitive science: An introduction. Cambridge, Massachusetts: The MIT Press. Turvey, M. T. (1977). Preliminaries to a theory of action with reference to vision. In R. Shaw & J. Bransford (Eds.), Perceiving, acting, and knowing: Toward an ecoZogicaZpqchoZogy. Hillsdale, I?. J.: Lawrence Erlbaum, pp. 211- 266.

SYSTEMS-MODELING PERSPECTIVE

This Page Intentionally Left Blank

VOLITIONAL ACTION* W.A. Hersh berger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

235

CHAPTER 10 A CONTROL-THEORY ANALYSIS OF INTERFERENCE DURING SOCIAL TRACKING

W. Thomas Bourbon When we engage in purposeful behavior, the actions of one person sometimes interfere with those of another. This kind of interference occurs any time the behavior of one person disturbs a variable controlled by another, who must then compensate for the disturbances produced by the first. Such interactions occur frequently, as when we walk along a hallway, or drive along a highway, and one of us (perhaps unknowingly) interferes with the progress of the other. In such cases, we usually compensate successfully, even when we do not know the origin of the disturbance: all we need do is act to cancel any deviations in the variable(s) we control. In this chapter, I use control theory (CT) to model interference. First I describe a simple pursuit tracking task in which each of two people controls a different visually-perceived relationship. One of them unintentionally, but unavoidably, interferes with the performance of the other. Next, I use CT to analyze and predict their performance. Control theorists (e. g., Bourbon, 1982; Powers,l973a, 1973b; Powers, Clark & McFarland, 1960a, 1960b) assume that, to achieve control, people act on the environment so as to create and stabilize specific perceptions. In the CT model, an environment equation represents variables affected by the person, and a person equation represents the hypothetical organization of the individual’s perceptions, goals and actions. To model social interaction, I use a separate personequation to represent each individual, while interactions between them are modeled as connections between environmental variables, not as links between features of the individuals. Each person is modeled as working to accomplish a perceptual goal independent of that for the other person. In this chapter I analyze interactions between people, but the principles discussed here probably characterize the control behavior of most living things. I focus on unilateral interference and compensation, but obviously there are other forms of interaction. In particular,

W Thomas Bourbon

236

individuals might mutually interfere with one another, or they might cooperate to control variables that one alone cannot control. The model in this chapter simulates both cooperation and mutual interference as effectively as it does unilateral interference (Bourbon, 1988; Chong, 1988), but because of limited space I cannot discuss those topics here.

Social Tracking When One Person Interferes With Another Many experiments based on CT examine individual performance on tracking tasks (e.g., Bourbon, 1987; Bourbon & Powers, 1988; Marken, 1980, 1982, 1986, 1988; Murphy, 1982; Powers, 1978), but in this chapter I analyze data from simultaneous pursuit tracking by two people. First, I describe the experimental conditions and the data from the pair; then I demonstrate how CT explains and predicts the data. Figure 1 presents a diagram of the experimental setup for the task.

Condition I Experimentalprocedures. In Condition 1, each of two people (Left and Right), seated beside each other at a table, moves a handle (HL or HR). The position of each handle affects the vertical position of a mark (a cursor) on the display screen of an IBM AT personal computer, equipped with a game port to sense the handles. The computer scales the influence of the handles so that when one moves through its full range the associated cursor moves through 200 units of vertical resolution on the screen. One cursor (CL) is on the left side of the screen; the other (CR), on the right. Their positions are determined entirely by the positions of the handles. The person on the left tries to keep the left cursor even with another mark (the left target, TL), which moves randomly in a vertical path on the left side of the screen under the influence of a smooth series of random numbers generated by the computer. The person on the right tries to keep the right cursor even with a target (TR) that follows a path on the right side of the screen that is the inverse of the path of TL. No one knows in advance the paths the targets will follow. A person could easily keep the cursor in any desired relationship with the target. The "even" position was selected to simplify a written presentation of the results. The right handle affects only the right cursor: the left handle, both cursors, but with an effect on the right cursor that is only half as great as on the left. Thus, if the left handle moves through its full range and

Social Tracking

237

Screen

Figure 1. Interactions between handles and random-number sets as they affect the computer screen in social tracking. (H = Handle; C = Cursor; T = Target. The designations "L" and "R' indicate "Left" and "Right,"respectively. The random lines are the vertical positions of the targets across time.) the right handle remains still, the left cursor moves through 200 units of resolution on the screen and the right cursor moves 100 units. An experimental run lasts 40 s, during which the computer samples and saves the positions of the handles and cursors into an array once every 1/30 s; the targets are created and saved before the run. Thus, there are 1200 data values for each variable. Figures 2 through 5 show the vertical positions of the targets, handles and cursors, as functions of time, relative to the vertical center of the screen. No plots of handle positions appear on the screen during a run, but after the run, handle

238

W; Thomas Bourbon

positions are scaled and plotted in units of displacement on the screen, rather than mechanical displacement. Figures 2 through 5 were traced from original data with "paint" software that has lower resolution than the data. Consequently, they show only about every third value for each variable, but no important feature of the data is lost. Results and discussion. Stated simply, each person kept the designated cursor close to the appropriate target. Figure 2 shows the data for Left, whose cursor correlated highly with the left target (r = .991, n = 1200 data pairs): the mean difference between them was -0.2 units of resolution (S.D. = 2.2). Because the left handle completely determined the position of the left cursor, their plotted paths were congruent. (The results labeled "model" in the figures are described later.)

+ Figure 2. Condition 1: Relative positions of target, handle and cursor over time for the person on the left. The upper figure is the experimental data. The lower figure is the modeled reconstruction of the data. (The y origin for the target and cursor is the center of the screen; the y origin for the handle is top dead center, with + and corresponding to pushing and pulling, respectively. Only the left handle affects the left cursor, so their paths are identical, and close to the target.)

I

Data: L e f t

+I I

Model: L e f t

0 Time

Is)

40

Social Tracking

239

Figure 3 shows the data for Right, whose cursor correlated highly with the right target (r = .988): the mean difference between them was -0.1 units of vertical resolution (S.D. = 2.6). The paths of the right hundle and cursor were similar (r = .999), but not congruent, because the right cursor was affected by both handles. The person on the right moved the handle to keep the right cursor even with its target, while also compensating for the influences of the person on the left, consequently the right handle was highly negatively correlated with the left (r = -.996).

Figure 3. Condition 1: Relative positions of target, handle and cursor over time for the person on the right. The upper figure is the experimental data. The lower figure is the modeled reconstruction of the data. (The y origin for the target and cursor is the center of the screen; the y origin for the handle is top dead center, with + and corresponding to pushing and pulling, respectively. Both the left and right handles affect the right cursor, so the paths of the right handle and cursor are not identical. The cursor is closest to the target.)

IData: Right O

I

40

+I

- IModel:

Right

40

D Time ( s )

240

W . Thomas Bourbon

Condition 2 Experimental procedures. Condition 2 was identical to Condition 1 in every detail but one: the two targets were determined by two new and independent tables of random numbers (r = -.379), each containing 1200 values. The tables, part of a set of three generated when Condition 1 began, were constrained so that none of the three could correlate any higher that .4, absolute value, with either of the other two. Each person again moved the handle to keep a cursor even with the appropriate target.

+I Figure 4. Condition 2: Relative positions of target, handle and cursor over time for the person on the left. The upper figure is the experimental data. The lower figure is the modeled reconstruction of the data. Only the left handle affects the left cursor, so their paths are identical, and close to the target. (The y origin for the target and cursor is the center of the screen; the y origin for the handle is top dead center, with + and - corresponding to pushing and pulling, respectively.)

I

Data: L e f t

IMadel: L e f t

I

4tl

0 Time (s)

Social Tracking

241

Results and discussion. Each person successfully tracked the appropriate target. Figure 4 shows the data for Left, who caused the left cursor to closely track its target (r = .992); the mean distance between them was -0.5 units of resolution (S.D. = 2.5). The paths of the left cursor and handle were congruent. Figure 5 shows the data for Right, who caused the right cursor to track its target (r = .976). The mean difference between cursor and target was -0.4 units of vertical resolution (S.D. = 3.5). Once again, the

Figure 5. Condition 2: Relative positions of target, handle and cursor over time for the person on the right. The upper figure is the experimental data. The lower figure is the modeled reconstruction of the data. Both the left and right handles affect the right cursor, so the paths of the right handle and cursor are not identical. The cursor is closest to the target. (The y origin for the target and cursor is the center of the screen; the y origin for the handle is top dead center, with + and - corresponding to pushing and pulling, respectively.)

Data: Right

a

IModel: Right

I

0

40 Time ( s )

242

W; Thomas Bourbon

0.5

~~~~~~~~~~~~

~~~~~

Figure 6. Interacting control-systems used to model social tracking with unilateral interference. (i = input function [sensors]; o = output function [effectors]; H = Handle; C = Cursor; T = Target; P = immediate perception of C and T; P' = ideal perception [reference level] of C and T. Designations "L" and "R" indicate "Left" and "Right,"respectively. Coefficients "1.0"and "0.5"indicate relative influences of handles on cursors.)

Social Tracking

243

paths of the right handle and cursor were highly correlated (r = .903), but not congruent, because the right cursor was affected by both handles. The correlation between the left and right handles was once again negative (r = -.691), but was smaller than in Condition 1, because in Condition 2 the people tracked targets that were not the inverse of one another. There is no question about what happened in the two conditions: people caused the cursors to track the targets, closely. In both conditions, the one on the right successfully compensated for disturbances caused by the one on the left. To test the idea that the CT model of interacting, but independent, control systems might explain their success, I used that model to reconstruct and predict the performance of the two people.

Modeling the Data One way to test a model is to require that it accurately reconstruct quantitative behavioral data. Another, more stringent, test requires that the same model produce quantitatively detailed and accurate predictions of data that will be gathered later. If the values created by the model closely resemble those in the behavioral data, then the organization hypothesized in the model might be at least a plausible explanation for the behavior. Both of those tests were applied to the CT model. The mathematical modeling described in this section occurred between the data-gathering sessions for the two experimental conditions. After the participants completed Condition 1, the CT model reconstructed their data. Then the same model predicted their data on Condition 2, which they performed 10 minutes after the predictions.

Reconstructing Condition 1 The modek The equations for the CT model are recurrence equations (Watt, 1966). Recurrence equations are common in dynamic system models (Davisson & Uhran, 1979), where the results of one "round'' of computations (ie., one set of solutions of the simultaneous equations) enter back into the equations as initial values for the next round. Thus, the specific values of variables in the equations for person(s) and environment at time t + 1 are functions of their values at time t: cause and effect still operate, but in a continuous recursive path between individual(s) and environment. Each round of computations

244

W: Thomas Bourbon

corresponds to a measured period of time in the original events being modeled; variables must be specified relative to clock-time in the "real world," rather than in terms of cycle-time in the computer. In the present example, a model of the results of Condition 1 must reconstruct the value of every measured variable for each of the 1200 intervals in the experiment. Figure 6 is a diagram of the two interacting CT models. The equations for the model follow. For the person on the left,

dHL = kL

(PLi

- P*L)

dt

(2)

where dHL is the change in HL during dt, which is one interval, (dt = 1/30 s); TL,., CLi and HLi are the values of TL, CL and HL in one interval, while TLitI, CLi+I and HLi+l are their values in the next; PLi is the present relationship between the target and cursor (TL - CL) (as a simplifying assumption, this value is used as an estimate of the present perceived difference between cursor and target); P*L is the desired (reference) difference between cursor and target (as a simplifying assumption, this is estimated as the mean difference between 1200 pairs of values for the two variables--in the original data it was -0.2 units of resolution); and kL is the integration factor, representing the velocity with which the person moved HL to eliminate any discrepancy between PL and P*L. (The person is assumed to eliminate only a portion of the perceived error during any given interval of time.) The value of kL, the integration factor for Left, was estimated from the original data by inserting the values of TL and CL, saved from the first interval in the original behavioral data, into equation 1; solving for PLi; inserting PLi, P*L and an arbitrary low value of kL into equation 2 and solving for dHL; adding dHL to HLi to obtain HLi+l in equation 3; then setting HLi+] equal to CLi+l in equation 4; then inserting CLi,I and the next value for TL into equation 1, which begins the next cycle of computations. That procedure was repeated 1199 times to reconstruct the original data. The Pearson correlation coefficient between the reconstructed and original values of CL was calculated and saved. Next,

245

Social Tracking

a slightly larger value of kL was inserted into equation 2 and the round of calculations was repeated. This entire procedure was repeated with increasingly larger values of kL until the correlation between the reconstructed and original values of CL reached a maximum. The value of kL that produced the best re-creation of the path of CL was used as the estimate of the person’s integration factor. For the behavioral data in Figure 2, the estimate of kL was 5.70, meaning the person moved HL the equivalent of 5.70 units of screen resolution per second per screen unit of difference between PL and P*L. For the person on the right, the equations were identical to those for the person on the left, with one exception: CR was affected by both HR and HL. Thus, PRj = TRi dHR = kR

CRi,,

-

CR,

(5)

(PRi - P*R)

= HRi,,

+ (.5

dt

(6)

HLi+l)

where all of the terms have the same meanings as in the equations for the person on the left, except that they refer to variables in the person on the right, or to variables on the right side of the computer display. Thus, this person has a reference level (P*R) for the perceived difference, (TR - CR). While solving the equations to estimate kR, the modeled position of CR for any interval was calculated as the sum of the calculated position of HR for that interval, plus one-half of the actual value of HL for that interval that was saved from the original tracking data. The calculated value of kR was 4.05,slightly less than that for Left, indicating that Right moved the handle a little less rapidly to eliminate perceived discrepancies between the positions of cursor and target. In the final step of modeling, the calculated values of kL, kR, P*L and P*R all were substituted in the appropriate equations for each person and all of the equations were solved recurrently, for 1199 intervals after t = 0. The resulting arrays of values for each variable were plotted and analyzed as the modeled reconstructions of the original data. Results and discussion. The results of reconstructing the data for Condition 1 are shown in Figures 2, for Left, and 3, for Right. For Left, the modeled paths of CL, TL and HL were very nearly congruent.

246

U? Thomas Bourbon

Obviously, the model portrayed the person as controlling a bit more precisely than actually occurred, but a comparison of the actual and modeled results in Figure 2 reveals that the reconstruction closely resembled the original data. For the actual and modeled positions of the left handle, r = .995, which was also the correlation between the actual and modeled paths of the cursors. Examination of Figure 3 reveals that the model also accurately simulated the data for Right: for the actual and modeled positions of the handle, r = .996; and of the cursor, r = .991. The accuracy of these reconstructions resembles that obtained when modeling data for tracking by individuals (e.g., Marken, 1982, 1986). Bourbon (1987) reported similarly high correlations between actual and reconstructed handle positions for individuals in 50 replications, 5 each by 10 people (mean r = .997, S.D. = .OOl).

Predicting Condition 2 The models. The models that predicted the results of Condition 2 were identical in every way to those in Condition 1. For the predictions, I modeled each person as trying to maintain the same cursor-target relationship as in Condition 1 and as moving the handle at the same velocity as in the first condition to eliminate perceived error. The values of kL, kR, P*L and P*R derived during reconstructions of Condition 1 were substituted into the equations, along with the new and as-yet-unseen values of TL and TR, and the equations were solved recurrently for new values of HL, HR, CL and CR: 1199 iterations of the equations generated the predicted values of the variables. The predictions were calculated 10 minutes before the subsequent tracking task and were not seen by the persons until after the tracking. Results and discussion. The predictions are shown in Figures 4 and 5, for Left and Right, respectively. The figures also contain data from the tracking task. The predictions were very accurate. For Left, the correlations were high between actual and predicted values of the handle (r = .993) and the cursor (r = .993). The same was true for Right: for the handle, r = .993; for the cursor, r = .991. People often refer to reconstructed data as "predictions." In a sense, that usage is appropriate, but I prefer to use the word more literally to refer to predictions-before-the-fact, because the CT model often allows us to make detailed quantitative predictions before we collect empirical data. That was clearly the case for the data reported here. In another instance, Bourbon (1988) reported similar accuracy for predictions made one year before tracking by an individual.

Social Tracking

247

General Discussion In the examples reported here, CT provided accurate quantitative predictions of both the actions of individuals and the environmental consequences of those actions: CT is as much a model for, how variables in the environment reach and remain at particular values as it is for behavior. The specific predictions reported here are not important. No one can anticipate every path a target will follow during tracking, or every interference a person will encounter during a lifetime, and that is the point. Only an omniscient being could anticipate such things, but control and tracking of important variables are routine achievements of even "simple" forms of life. What is important is that the organization hypothesized in control theory allows us to predict that, whatever the source and nature of the disturbances acting on variables controlled by an organism, the organism will act to cancel their effects. Given that organization, the task of making predictions as precise as those reported here is simple; without it, the task often seems impossible. Until recently, quantitative analyses using control theory focused on performance by individuals (e.g., Bourbon, 1987; Bourbon & Powers, 1988; Marken, 1980, 1982, 1986, 1988; Murphy, 1982; Powers, 1978), but there was no similarly detailed modeling of social interactions. There is an extensive literature using models other than CT to analyze interaction, but it does not discuss interactions as examples of control behavior. And most social research does not address real-time interactions among variables that are continuous in time. As part of an extensive program of research independent of that on CT, K. U. Smith and his colleagues did study continuous real-time tracking by dyads (see the reviews by Smith & Smith, 1987, 1988). They provide abundant evidence, some of it elegant, for the importance of negative feedback as a feature of behavior, but they offer no formal model of control, whether by individuals or dyads. The analyses offered by Smith and his associates cannot precisely predict the outcomes of their research because, in contrast with the CT approach, they use no quantitative model. Instead, they offer descriptive accounts of their often elegant empirical studies. Nonetheless, they can claim priority for the term "social tracking," which they used to identify interactions based on "social feedback" (e.g., Smith, 1972). The present chapter is the first published quantitative application of CT to the analysis of social interaction, but there were earlier published attempts at conceptual analyses derived loosely from CT. For example,

248

U? Thomas Bourbon

White (1986a, 1986b) described families and armies as control systems. He said individuals occupy specific niches in an organization, where each fulfills one function in a control system. Thus, one person might be a "sensor," another a "comparator," and another an "effector," and the hypothesized "social system" is the source of control. Lord and Hanges (1986) applied a similar description to corporations. These authors offer a verbal "model" for control of individuals, not control by individuals. In his comments on White (1986a), Powers (1986) suggested that such construals of people and control are mistaken and that interactions between people can be modeled as occurring between independent control systems, not between "components" of larger "real" systems that exist independently of the individuals. The functions and variables in the person-equations of the CT model reside in the individuals, not in free space between them, and not in a superordinate system. There are no effective goals, social roles, norms, values and the like, outside of individuals; only from within do they determine an individual's voluntary behavior by playing the role of "reference signals," against which controlled perceptions are compared. Several writers have successfully applied the control-theory model in conceptual analyses of social behavior: Powers (1980), in a paper with limited circulation, has used CT to postulate essential stages in the historical development of social interactions; Plooij (1984, 1987), and Ritj-Plooij and Plooij (1987) have used CT to interpret the development of control, via social interactions, in chimpanzee and human infants; McPhail and Wohlstein (1986) have used CT to interpret interactions between people in various types of social locomotion, ranging from spontaneous walking and running to ritualized processions; Soldani and Ford (1983) have used CT as a basis for personnel management in industry; and Ford (1987) has used CT as a basis for the practice of family counseling. In each case, they have demonstrated that the CT model for individuals affords a plausible explanation for social phenomena; that is, individuals act to eliminate discrepancies between their own individual goals and perceptions, and the interactions occur between environmental variables affected by the individuals. The notion that individual volitional behavior, in the form of selfcontrolled perceptual inputs, is the basis of social life is one which deserves careful experimental evaluation. The notion appears sound, as illustrated by the quantitatively precise modeling described in this chapter. However, the evaluation and the modeling have just begun.

Social Tracking

249

References Bourbon, W. T. (1982, October). Models of control. Paper presented at the annual meeting of the American Society for Cybernetics, Columbus, Ohio. Bourbon, W. T. (1987, October). Control theory as a source of accurate and reliable predictions of tracking performance. Paper presented at the third annual meeting of the Control Theory Group, Haimowoods, Wisconsin. Bourbon, W. T. (1988, September). Modeling control by dyads. Paper presented at the fourth annual meeting of the Control Theory Group, Haimowoods, Wisconsin. Bourbon, W. T., & Powers, W. T. (1988, September). The scientific analysis of control behavior. Paper presented at the fourth annual meeting of the Control Theory Group, Haimowoods, Wisconsin. Chong, E. (1988). Cooperation interpreted in t e r n of control Jrstem theory. Unpublished master’s thesis, Stephen F. Austin State University, Nacogdoches, TX. Davisson, W. I., & Uhran, J. J. (1979). A primer for NDTRAN: A continuous Jrstem intepreter. Notre Dame, Indiana: University of Notre Dame. Ford, E. (1987). Love guaranteed: A better martiage in 8 weeks. San Francisco: Harper & Row. Lord, R. G. & Hanges, P. J. (1987). A control system model of organizational motivation: theoretical development and applied implications. Behavioral Science, 32, 161-178. Marken, R. S. (1980). The cause of control movements in a tracking task. Perceptual and Motor Skills, 51, 755-758. Marken, R. S. (1982). Intentional and accidental behavior: A control theory analysis. Psychological Reports, 50, 647-650. Marken, R. S. (1986). Perceptual organization of behavior: A hierarchical control model of coordinated action. Journal of Experimental Psychology: Human Perception and Pe$omzance, 12, 267-276. Marken, R. S. (1988, September). Degrees of freedom in behavior. Paper presented at the fourth annual meeting of the Control Theory Group, Haimowoods, Wisconsin. McPhail, C., & Wohlstein, R. T. (1986). Collective locomotion as collective behavior. American Sociological Review, 51, 447-463. Murphy, P. E. (1982). A n analysis of a tracking task from the perspective of control theory. Unpublished master’s thesis, Stephen F. Austin State University, Nacogdoches, TX.

250

FK Thomas Bourbon

Plooij, F. X. (1984). The behavioral development of pee-living chimpanzee babies and infants. Norwood, NJ: Ablex. Plooij, F. X. (1987). Infant-ape behavioral development, the control of perception, types of learning and symbolism. In J. Montangero, A. Tryphon & S. Dionnet (Eds.), Symbolisme et connuissance I Symbolism and knowledge, Cahier No. 8, Foundation Archives Jean Piaget (pp.35-63). Geneva, Switzerland. Powers, W. T. (1973a). Feedback Beyond behaviorism. Science, 179, 351-356. Powers, W. T. (1973b). Behavior: The control of perception. Chicago: Aldine. Powers, W. T. (1978). Quantitative analysis of purposive systems: Some spadework at the foundations of scientific psychology. Psychological Review, 85, 417-438. Powers, W. T. (1980). Control-theory psychology and social organizations: Background for a theory of the corruption of social indicators when used for social decision making. Unpublished manuscript, The Maxwell School, Syracuse University, Syracuse, NY. Powers, W. T. (1986). Interaction: The control of perception. Commentary on a paper by J. M. White. (In CSR Working Paper No. 86-4). Alberta, Canada: University of Alberta, Center for Systems Research, pp. 1-9. Powers, W. T., Clark, R. K., & McFarland, R. I. (1960a). A general feedback theory of human behavior: Part I. Perceptual and Motor Skills, 11, 71-78. Powers, W. T., Clark, R. K., & McFarland, R. I. (1960b). A general feedback theory of human behavior: Part 11. Perceptual and Motor Skills, 11, 309-323. Ritj-Plooij, H. H. C. van de, & Plooij, F. X. (1987). Growing independence, conflict and learning in mother-infant relations in free-ranging chimpanzees. Behaviour, 101, 1-86. Smith, K. U. (1972). Social tracking in the development of educational visual skills. American Journal of Optometry and Archives of American Academy of Optometry, 149, 50-59. Smith, T. J., & Smith, K. U. (1987). Feedback-control mechanisms of human behavior. In G. Salvendy (Ed.), Handbook of human factors (pp. 251293). New York: Wiley-Interscience. Smith, T. J., & Smith, K. U. (1988). The cybernetic basis of human behavior and performance. Continuing the Conversation: A Newsletter of Ideas in Cybernetics, 15, 1-28. Soldani, J. C., & Ford, E. E. (1983). Money isn’t enough: Managingpeople eflectively through control system theory. (Available from E. E. Ford, 10209 N. 56th St., Scottsdale, AZ.)

Social Tracking

25 1

Watt, K. E. F. (1966). The nature of systems analysis. In K. E. F. Watt (Ed.), Systems analysis in ecology (pp. 1-14). New York: Academic Press. White, J. M. (1986a). The family system of inteTersona1perception: Interaction as the control of perception. (CSR Working Paper No. 86-1). Alberta, Canada: University of Alberta, Center for Systems Research. White, J. M. (1986b). Interaction: The control of perception. Reply to comments by William I: Powers. (In CSR Working Paper No. 86-4). Alberta, Canada: University of Alberta, Center for Systems Research, pp. 13-15.

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hershberger (Editor) Elsevier Science Publishers B. V. (North-Holland). 1989

0

253

CHAPTER 11 VITE AND FLETE NEURAL MODULES FOR TWECTORY FORMATION AND POSTURAL CONTROL Daniel Bullock’ and Stephen Grossberg? 1. Factorization Of Pattern And Energy In Trajectory Formation And Tension Control The modern study of neural networks has shown that even the simplest actions may involve the coordinated activity of millions of biological elements. For example, a simple skeletal action such as rotation around a single joint involves coordinated activity of two muscles, one of which shortens while the other lengthens. Each muscle in turn is composed of a population of contractile fibers and sensory organs, which, respectively, are linked by efferent and afferent nerves with neuronal populations in the spinal cord. These spinal neuron populations are in turn linked, by a bewildering array of descending and ascending pathways, with a large number of discrete supraspinal centers. Volitional activity may at first seem to belie this complexity, for when we exercise voluntary control, it seems that we do something quite simple. How can voluntary activity appear so simple if every action is inherently so complex? The answer lies in noticing the nature of the control we voluntarily exercise, and also the large range of action parameters over which we lack direct control. We do seem to exercise fairly direct control over where and how fast we move (Sections 2-6), over how forcefully we try to hold a posture (Sections 7-15), even over the vigilance with which we perform tasks (Carpenter & Grossberg, 1987,

Supported in part by the National Science Foundation (NSF IR I-87-16960). Supported in part by the National Science Foundation (NSF IRI-87-16960) and the Air Force Office of Scientific Research (AFOSR F49620-86-C-0037& AFOSR F49620-87-C-0018).

254

Daniel Bullock and Stephen Grossberg

1988). On the other hand, we do not have direct control over many of the processes that automatically compensate for changes in limb position and inertia during rapid trajectory formation, or that adapt to changes in the parameters of the muscle plant due to exercise, aging, or accidents. One of the devices whereby voluntary control is simplified is the use of non-specific control signals. A nonspecific signal is a scalar signal that is generated at a single command source and broadcast, through a parallel fan-out of pathways, to many target cells. It is then up to the target cells to react appropriately to the widely broadcast signal. If each cell reacts in a state-dependent manner, a nonspecific signal can control an entire array of events without requiring conscious knowledge of the controlled array. In order for such a simplification of control to work, the target cell array must be appropriately designed. For example, in a real neural network, all neurons have finite activities, or potentials. Broadcasting the same signal to an entire array of cells could raise the baseline level of activity across the array and push the activities of many cells to their maximal potentials. The nonspecific signal could hereby homogenize, or compress, the spatial pattern of cell activities. Because information in a neural network is carried by such spatial patterns, a nonspecific signal could easily become information-destroying. Thus, any widely broadcast volitional signal could destroy information in the target network unless the target network is designed to automatically compensate for the potentially saturating effects of the nonspecific signal. The target network must guarantee that the volitional signal has its intended modulatory effect without destroying its pattern registration and processing characteristics. The above mentioned issue has been called the pattern-energy factorization problem (Grossberg, 1970, 1973, 1978, 1982) to emphasize that many networks are designed to factor pattern differences from overall activity levels. This problem arises in perceptual and cognitive contexts, no less than during voluntary motor control. In this paper, we will illustrate how pattern-energy factorization is realized within two model networks that may play a role in solving two fundamental problems of motor control: How movements are performed at different intended speeds, and how postures are held at different intended levels of rigidity. The first network, called the Vector Integration To Endpoint, or VITE model (Bullock & Grossberg, 1986, 1988a), indicates how a scalar signal, called the GO signal, affords voluntary control of movement rate without disrupting the spatial pattern characteristics, such as distance and

WTE and FLETE: Neural Modules

255

direction, of the movement trajectory. We will survey operating characteristics of the VITE network, and describe some of the physiological data that support these properties, notably properties of the GO signal. The second network model, introduced in Bullock and Grossberg (1988d), is called the FLETE model, as a mnemonic for Factorization of LEngth and TEnsion. It models a part of the spino-muscular system that cooperates with the VITE network to generate the forces needed to ensure that the limb follows the trajectory commanded by the VITE circuit. The general outlines of the trajectory-matching capabilities of the spino-muscular system have been discussed elsewhere (e.g., Gielen & Houk, 1987). We focus here on how to design the circuit to enable holding of an arbitrary posture at multiple levels of rigidity via a scalar signal sent in parallel to the motor channels responsible for posture maintenance. As with the VITE network, the FLETE network indicates how a scalar signal resulting from a voluntary choice can have a desired effect on a high-dimensional target system without knowledge of the characteristics of the target system. Besides discussing the FLETE network's ability to factorize length from tension despite the tendency of a muscle's tension to covary with its length, we will present physiological evidence for the network's components in the spino-muscular system. The FLETE model's mathematical treatment of the spino-muscular system sheds new light on why certain properties, such as the size principle of motoneuron recruitment (Henneman, 1957, 1985), are so prevalent among vertebrates. The FLETE model also suggests a new functional interpretation for the Renshaw-mediated recurrent inhibition seen in higher vertebrates. Our analysis of the size principle indicates that recurrent inhibition via the Renshaw-Ia pathway serves a more fundamental function than "controlling the gain of the stretch reflex," the most common textbook statement of its primary role. In particular, we argue that this pathway compensates for a threat to postural stability that occurs when the size principle is combined with the co-contractive signals needed to control postural rigidity. Rather than contradicting other views, our new account helps complete the picture of Renshaw function, and broadens the theoretical context for understanding how "supra-spinal convergence on Renshaw cells allows recurrent inhibition to serve as a variable gain regulator" (Hultborn, Lindstrom, & Wigstrom, 1979) for the final common path. In particular, we make novel predictions regarding supraspinal Renshaw

256

Daniel Bullock and Stephen Grossberg

modulation during postural versus movement states by showing through computer simulations that the desirable decoupling of alpha-motoneuron and Renshaw activities observed in some movement contexts (PierrotDeseilligny & Morin, 1980) would be disastrous in many postural contexts.

I Figure 1. Main variables of the VITE circuit: T = target position command, V = difference vector, G = GO signal, P = present position command. The circuit does not include the opponent interactions that exist between the VG and P stages of agonist and antagonist muscle commands. For these interactions, see Figure 2.

2

P

2. Emergent Invariants Of The Vite Circuit In the VITE model, motor planning occurs in the form of a Target Position Command, or TPC, an array that specifies the lengths to which all trajectory-controlling muscles are intended to move, and an independently controlled GO command, which specifies the movement’s overall speed. Automatic processes convert this information into an arm trajectory with invariant properties, notably properties of synchrony among muscle synergists. These automatic processes include computation of a Present Position Command, or PPC, and a Difference Vector, or DV. The DV is the difference between the TPC array and the PPC array at any time. The PPC is gradually updated by integrating the DV through time. A time-varying GO signal multiplies, or gates, the DV before it is integrated by the PPC. The PPC generates an outflow movement

257

HTE and FLETE: Neural Modules

Figure 2. Opponent interactions among VITE circuit channels controlling agonists and their antagonists enable coordinated, automatic updating of their present-position commands (PPCs). Outputs from the PPC stage serve as the basis for reciprocal control of opponent muscles' contractile states.

GO

u AGONIST

ANTAGONIST

PPC STAGE

command to motoneurons controlling its target muscle groups and sends a corollary discharge (inhibitory efference copy) back to the DV stage (Figure 1). Opponent interactions are also needed to regulate the PPC's to agonist and antagonist muscle groups at each joint (Figure 2). To generate a movement, a TPC different from the current PPC is instated. This generates a non-zero DV which is multiplied by the GO signal to generate an input to the PPC. The PPC integrates this signal through time until the PPC equals the TPC. The model hereby explicates how a limb is commanded to reach the same goal (TPC) from arbitrary initial states (initial PPCs) at variable rates (determined by the size of the GO signal). In its simplest form, excluding terms expressing opponent interactions, the VITE circuit obeys the equations: Difference Vector

d dt

vi= a(+

+Ti -Pi)

and Present Position Command

d

a Pi = G[Q]+,

258

Daniel Bullock and Stephen Grossberg

Figure 3. (A,B): With equal GO signals, movements of different size have equal durations and perfectly superimposable velocity profiles after velocity axis rescaling. Shown are GO signals and velocity profiles for 20 and 60 unit movements lasting 500 ms. (C,D,E): Velocity profiles associated with small, medium, and large GO magnitudes result in slow, medium, and fast performance of a 20 unit movement. Each SR value gives the trajectory's symmetry ratio; that is, the time taken to move half the distance divided by the total movement duration. These ratios indicate progressive symmetrization at higher speeds, within the range of speeds shown. (F): The velocity profiles shown in (C),(D), and (El are not perfectly superimposable after time and velocity normalization.

NORMALIZED TIME

tn

WTE and FLETE: Neural Modules

259

1 SYNERGIST BEGINS CONTRACTION Simulation f SYNERGIST ENDS CONTRACTION results showing contraction offset times for three synergistic muscles with different onset times, as a function of the GO signal scalar (the voluntarily chosen multiplier of the time-varying GO signal). In each block, the DV component corresponding to muscle one begins to be read z out 0 ms after the start of GO signal buildup, W muscle two 150 ms after the start of GO buildup, and muscle three 300 ms I I1 I11 IP after the start of GO buildup. The GO signal scalar was 10, 20, 40, and 80 in blocks I-IV, respectively. Results indicate automatic VITE circuit compensation for staggering of contraction onset times. Figure 4.

----

where a is a positive constant and [?I+ = max(?,O). Equations (1) and (2) describe interactions of a generic component of a target position command ( T I , T2,..., T,), a difference vector (VI, V2,..., Vn),a present position command (PI, PI,..., P,), and a time-varying velocity command, or GO signal G(t) L 0. The difference vector computes a mismatch between target position and present position, and is used to update present position at a variable rate determined by G ( t ) until the present position matches the target position. The GO signal is thus a nonspecific command, and the circuit as a whole factors pattern, in the form of a vector ( T I , T2,..., T,), and energy, in the form of a scalar G(t). Because the updating rate for each PPC component Pi is a function of the corresponding DV component q, such a scheme permits multiple muscles to contract synchronously even though the total amount of contraction, scaled by Tj(0)-Pi(O), may be different for each effector (Figure 3).

260

Daniel Bullock and Stephen Grossberg

Because the GO signal multiplies the DV signal (Equation 2) on its way to the PPC stage, the VITE circuit is capable of motorpriming. In particular, even if a new TPC is instated causing a new DV to be computed, there will be no PPC updating until G(t)becomes greater than zero. A new target may hereby be primed before it is released by a volitional act that generates a positive GO signal. Variable speed control and motor priming are thus intimately linked in the VITE model. In addition, the interaction between the DV and GO signal provides automatic compensation for staggered onset times of synergetic muscles. As shown in Figure 4,even if motor commands to different muscles in a

Figure 5. Learning of an intermodal associativetransformation between target position maps is gated by a DV process which matches TPC with PPC to prevent incorrect associations from forming between eye-head TPC's and hand-arm TPC's. Learning only occurs when

GATE

the DV is small.

synergy are initiated at very different times, the times at which all the PPCs reach their target TPCs are much more synchronous. This property may help to generate linear hand-movement paths (Hollerbach, Moore, & Atkeson, 1986). It is also needed to execute motor plans involving a rapid series of movements, because asynchrony in the completion of one movement in the series could destabilize the correct execution of the next movement in the series.

261

U T E and FLETE: Neural Modules

The VITE circuit generates trajectories without regard to the forces required to realize those trajectories. It is not sufficient in itself to accomplish all the tasks required of a variable-speed, variable-load limb movement system. In particular, a key difference between the VITE model and many other model proposals is that the PPC movement command is based on outflow, or feedforward, signals, not inflow, or feedback, signals from the muscles. The process of guaranteeing that the PPC command actually moves the limb to a corresponding position in space is accomplished, in part, by a separate muscle linearization network (MLN) that does use inflow signals in the form of outflow-inflow mismatches to generate error signals that adaptively alter the gain of the total outflow movement command (Grossberg & Kuperstein, 1986, 1989). The cerebellum has been implicated as the locus of adaptive gain change in this model. For arm movements, the VITE circuit and an MLN circuit operate in parallel with the spino-muscular FLETE model system to realize flexible and adaptive trajectories free from the rigid performance inherent in systems that preplan an entire trajectory before beginning to generate the forces needed to implement it.

Figure 6. A passive update of position (PUP) circuit. An adaptive pathway PPC + DVp calibrates PPC-outflow signals in the same scale as inflow signals during intervals of posture. During passive movements, output from GO equals zero. Hence the passive difference vector DVP updates the PPC until it equals the new position caused by any passive movements that may occur due to the application of external forces.

PASSIVE MOVEMENT SIGNAL

SIGNAL

262

Daniel Bullock and Stephen Grossberg

3. Actively Gated Learning Of Target Position And Present Position In addition to their role in trajectory formation, TPC, PPC, and DV computations are predicted to actively modulate, or gate, the learning of associative maps between TPC’s of different modalities, such as between the eye-head system and the hand-arm system (Figure 5) in a manner consistent with Piaget’s (1963) analysis of how circular reactions promote map learning. The gating process prevents learning from occurring except when the PPC is close to the TPC; that is, except when the DV is small. Such gating helps to prevent spurious correlations from being learned, say between a fixed target position of the hand and all present positions which the eye assumes while moving to look at the hand. By using such an intermodality associative map, looking at an object can activate a TPC of the eye-head system that is associatively mapped into an appropriate TPC of the hand-arm system, as in the recent model of Kuperstein (1988). A VITE circuit translates this latter TPC into a synchronous movement command for moving the hand to the object. Active gating is also needed to regulate updating and learning of present position commands during passive arm movements. In particular, an auxiliary circuit, called the Passive Update of Position, or PUP, Model uses inflow signals to update the PPC during passive arm movements due to external forces (Figure 6), but not during active arm movements or during actively maintained posture. Because the scales of outflow position command signals Pi and inflow position sensing signals Ii cannot be assumed to be the same, the PUP circuit incorporates a synaptic modification mechanism for adaptively recalibrating the gain ziof the PPC outflow signals that are matched with inflow signals until they are computed in the same numerical scale, thereby ensuring that the PUP circuit accurately updates the PPC. When the VITE Model and PUP Model are combined, the following equations result: Difference Vector

d- 5 = a(-Vi +Ti -Pi) dt

U T E and FLETE: Neural Modules

263

Present Position Command

Oucflow-Inflow Match

Adaptive Gain Control

d &

2.

1

= dGp(-tzi

+[Mi]+).

Active and Passive GO Signals

G

+ Gp > 0 and GGP = 0

(6)

Equation (3) supplements equation (2) with a position update signal Gp[Mi]+. This signal is turned on only when the passive GO signal gating function, or "pauser" signal, G becomes positive as G becomes zero in (6), and when the outflow-inffow match function Mi>O. Equation (4) shows that Mi>O only when the outflow signal, z f j , is less than the inflow signal, yZi . Function zi in (5) is a long term memory trace, or associative weight, which adaptively recalibrates the gain of outflow signals Pi until they are in the same scale as inflow signals yZi in (4). Then the match function Miis able to update the PPC in equation (3) until it accurately registers the new position caused by external forces. In summary, offset of the GO signal within the VITE circuit enables a pauser signal within the PUP circuit to drive its learning and reset functions. Such pauser-modulated learning during mismatches has been suggested to occur in several adaptive sensory-motor control circuits (Bullock & Grossberg, 1988~;Grossberg and Kuperstein, 1986, 1989). Onset of the GO signal is also suggested to inhibit inflow channels controlling "long-loop" postural reflexes, which would disrupt trajectory formation if allowed to function during fast movements (Evarts & Fromm, 1978).

264

Daniel Bullock and Stephen Grossberg

4. Behavioral Operating Characteristics And Vector Cell Properties In Precentral Motor Cortex Because the VITE model proposes that trajectories are generated as the arm tracks the evolving state of a neural circuit's output stage (the PPC), the model can be tested in two ways: by comparing trajectories of the neural circuit's output stage (e.g., Figures 3,4) with behavioral data concerning actual arm trajectories, and by checking for the existence of the neural components postulated in the model. Detailed quantitative comparisons of model predictions with behavioral data can be found in Bullock and Grossberg (1988a, 1988b). Among the properties treated therein are: Woodworth's distance-error law (Woodworth, 1899); the speed-accuracy trade-off known as Fitts' law (Fitts, 1954); peak acceleration as a function of movement amplitude and duration, and isotonic arm movement properties before and after arm-deafferentation in animals deprived of visual feedback (Bizzi, Accornero, Chapple, & Hogan, 1984); synchronous and compensatory "central error correction" properties of isometric contractions (Gordon & Ghez, 1987); velocity amplification during target switching (Figure 7 and Georgopoulos, Kalaska, & Massey, 1981); velocity profile invariance across different movement distances (Figure 3A,B and Freund & Budingen, 1978) and change in profile asymmetry across different movement durations (Figure 3C-F and Beggs & Howarth, 1972; Moore and Marteniuk, 1986; Zelaznik, Schmidt, and Gielen, 1986); and changes in the ratio of maximum to average velocity during discrete vs serial movements (Ostry, Cooke, & Munhall, 1987). Neurophysiological data support the existence of the major stages in the VITE model. In particular, the VITE model includes a DV stage, the analogue of which does not exist within mass-spring models of trajectory formation (e.g., Cooke, 1980). Cell populations have been identified that possess all the properties required of an in vivo analogue of DV stage neurons. For example, Georgopoulos and associates (Georgopoulos, Kalaska, Caminiti, & Massey, 1984; Schwartz, Kettner, and Georgopoulos, 1988) have located a class of cells in the shoulder-elbow zone of the precentral motor cortex (Area 4). Called vector cells, they have the following properties in common with VITE model DV cells: (1) activity levels correlate with arm movement direction but not arm movement endpoint; (2) activity levels may be primed prior to movement, as required by the postulate that actual movement depends on GO signal activation; (3) the time course of vector cells is highly correlated with the time course of the model DV; (4) vector cell coding of movement

WTE and FLETE: Neural Modules

265

direction does not reverse during the second half of the movement, indicating pure kinematic coding with no braking-force component, consistent with the VITE-FLETE-MLN parsing strategy; and (5) vector cells project to interneurons rather than directly to motoneurons, as required by the V I E model postulate of an outflow PPC stage that must be supplemented by FLETE and MLN signals to generate the total movement command. Thus the VITE model provides a mechanistic understanding of how the neural population vectors measured by Georgopoulos and associates may be computed by a distributed neural circuit (for further details, see Bullock & Grossberg, 1988a; Georgopoulos, this volume).

5. Physiological Evidence For Globus Pallidus As A Component Of The GO Signal Pathway In addition to evidence of Georgopoulos and his colleagues that cell populations in precentral motor cortex behave like an in vivo analogue of model DV stage neurons, physiological support for the VITE model comes from recent experiments involving lesions, electrical stimulation and microelectrode recording studies of the basal ganglia. Data from a set of experiments by Horak and Anderson (1984a, 1984b; Anderson & Horak, 1985) are consistent with the interpretation that the internal segment of the globus pallidus is a component of an in vivo analogue of the VITE model's GO signal pathway. An in vivo candidate for a GO signal pathway must pass four tests. First, stimulation at some site in the proposed pathway must have an effect on the rate of muscle contractions. Second, it must have this effect without affecting the amplitude of the contractions. Thus stimulation should have no effect on movement accuracy, except possibly for effects caused by imperfect motor realization of the PPC commands. Third, this rate-modulating effect should be non-specific: it should affect all muscles that are typically synergists for the movement in question. Fourth, because movement depends on the conjunction of a positive DV and a positive GO signal (Equation 3), no movement should occur in the absence of either signal. The studies conducted by Horak and Anderson (1984a, 1984b) pertain to all of these issues. Horak and Anderson (1984a) showed that "when neurons in the globus pallidus [of Macaque monkeys] were destroyed by injections of kainic acid (KA) during task execution, contralateral arm movement times (MT) were increased significantly, with

266

Daniel Bullock and Stephen Grossberg

little or no change in reaction times" (p.290). This satisfies the rate criterion. Moreover, the rate of motor recruitment was depressed "in all the contralateral muscles studied at the wrist, elbow, shoulder, and back, but there were no changes in the sequential activation of the muscles" (p.20). This satisfies the non-specificity criterion. Finally, the authors also noted that "animals displayed no obvious difficulty in aiming accurately... they did not miss the 1.5-cm target more often following KA injections, and there was no noticeable dysmetria around the target" (p.300). This satisfies the accuracy criterion. Horak and Anderson (1984b) used an electrical stimulation paradigm instead of a lesion paradigm. They found that "stimulation in the ventrolateral internal segment of the globus pallidus (GPJ or in the ansa lenticularis reduced movement time, whereas stimulation at many sites in the external pallidal segment (GP,), dorsal GP,, and putamen increased movement times for the contralateral arm" (p.305). Once again, these effects were non-specific: "no somatotopic effects of stimulation were evident. If stimulation at a site produced slowing, it produced a depression of activity in all the muscles studied. Even stimulus currents as low as 25 p 4 affected proximal as well as distal muscles, flexor as well as extensor muscles, and early-as well as late-occurring activity" (p.309). The conjunction criterion for a GO-signal pathway was also met. In the VITE model, activation of the GO-signal pathway produces movement only if instatement of a TPC different from the current PPC leads to the computation of a non-zero DV, regardless of the value of G ( t ) . In agreement with this property, Horak and Anderson (1984b) observed that "stimulation at sites that speeded movements did not induce involuntary muscle activation in resting animals nor did it change background EMG activity prior to self-generated activity during task performance'' (p.313). In Bullock and Grossberg (1988a) we noted that "very rapid freezing can be achieved by completely inhibiting the GO signal at any point in the trajectory.'' This property of the model has been partially shown to be a property of the GP system by the demonstration that stimulation in inhibitory zones adjacent to GPi significantly slowed movement, as noted above. Interestingly, Horak and Anderson also reported that "stimulation with 50 or 100 p A at...sites ventral and medial to typical GPi neuronal activity completely and immediately halted the monkey's performance in the task" (p.315). Though the sites producing halting in the Horak and Anderson studies apparently do not inhibit the GP,, they may inhibit targets of the GP, output pathway. Prior studies using much larger currents in zones known to inhibit GP, have produced halting (Van

WTE and FLETE: Neural Modules

267

Buren, Li, & Ojemann, 1966). Taken together, their experiments led Horak and Anderson (1984b) to conclude that '?he basal ganglia...determine the speed of the movement" (p.321). Consistent rate-control data for speech movements have been reported by Mateer (1978). In a study of timing relations between natural pallidal neuron discharges and the earliest detectable EMG activity, Anderson and Horak (1985) observed that though about 30% of pallidal neurons began firing 50-150 ms before mechanically detectable movement, "only 13 of 108 neurons showed changes in activity before the earliest EMG activity recorded during the same trials, and for only two of them did the initial changes in firing rate precede the initial changes in EMG activity by more than 25 ms" (p. 444). From this they conclude that "it is unlikely that changes in pallidal firing would be important in determining the initiation of the arm movement... but they could be important in controlling the buildup or scaling of EMG activity and thus the duration of the movement'' (p. 444). Similar timing relations in monkeys have been reported by Mitchell, Richardson, Baker, and D e h n g (1987). These timing relations have several alternative interpretations that require further discussion, especially in the light of cat data consistent with an initiating role for pallidal output signals (Neafsey, Hull, & Buchwald, 1978). Before beginning our discussion, we note that even if we accepted Anderson and Horak's caveat regarding initiation, the GP would still be implicated in GO-signal buildup, but in such a way that the GO signal can begin its gating action before the pallidal part of the circuit becomes engaged. Such a picture is consistent with one aspect of the proposal of Penney and Young (1983), that the globus pallidus is part of a positive feedback loop, which could be used to help generate a GO signal that grows in time with the shape shown in Figure 3. However, both theoretical and empirical considerations suggest that Anderson and Horak may have underestimated the role of the GPi in movement initiation. In any planned movement context, there are likely to be a set of central events, all of which may be jointly involved in "determining the initiation of the arm movement." In particular, an arm movement will be more successful if the muscles controlling body segments that serve as the postural base for the arm are activated before the phasic arm movement is itself initiated. Gahery and Massion (1985) have reported central and muscular postural adjustments with lead times in excess of 25 msec before the onset times for central and muscular arm-movement producing activations.

268

Daniel Bullock and Stephen Grossberg

From this perspective, the data of Anderson and Horak do not rule out the GPj as a GO signal generator. Rather, they further buttress the argument that a gradually increasing GO signal exists in the GPi . In particular, each animal individually showed some pallidal activity at least 25 msec prior to the earliest EMG activity. Inspection of Anderson and Horak's (1985) Figure 8 reveals that this "short" 25 msec lead time held only for the thoracic paraspinal muscle, whose activity was probably generated by the separate circuit responsible for preparing the postural base for the forthcoming arm movement (Gahery & Massion, 1985). In contrast, pallidal activity led EMG activity in all arm-projection muscles (biceps, deltoid, radialis) by at least 50 ms. Such a lead time is compatible with an initiating role because the GPi may be tri-synaptically linked to motoneurons via two separate pathways. In addition, Anderson and Horak used a simple RT task, which allows complete DV priming. Such a task would be expected to eliminate any effect of the GO signal on RT as well as reduce to a minimum the lag between GO activation and initial muscle activation. Initial muscle activity is affected by the product of the GO signal and the large initial DV. Because the GO signal is assumed to start small and to grow gradually, only a small proportion of pallidal neurons should become active prior to initial muscle activity. Thus the Anderson and Horak (1985) observations of gradual recruitment of active pallidal neurons are consistent with the model postulate of a gradually growing GO signal (Figure 3). The hypothesis of a gradually growing GO signal was needed to quantitatively explain parametric data about arm movement velocity profiles (Bullock & Grossberg, 1988a). The internal segment of the globus pallidus is one of two main output nuclei for the basal ganglia (BG). An assessment of its role as a GO signal generator thus needs to consider inputs to the basal ganglia. In particular, it is necessary to ask whether the basal ganglia receive the afferents one would expect to govern the final decision to execute a primed motor command. This issue has recently been addressed by Passingham (1987), who concluded "that it is the basal ganglia that finally direct the action to be taken" (p.90). Regarding BG inputs, he noted that for a correct evaluation of the context for action, "the motor system must be influenced by information from all of the cortical regions....In fact there is a massive projection from all these areas, but it runs not across the cortex but downward to the basal ganglia" (p.85). Moreover "the ventral striatum [one of the BG input zones] receives a heavy projection from the amygdala...[which] plays a crucial role in the learning of

U T E and FLETE: Neural Modules

269

motivational and emotional associations" (p.85). Thus the basal ganglia receive inputs whereby cognitive and motivational information may be integrated to arrive at decisions to act. The other main output nucleus of the basal ganglia--the substantia nigra (SN) pars reticulata--is known to gate read-out of movement commands controlling saccadic eye movements. It does this by disinhibiting deeper layers of the superior colliculus (Sparks & Jay, 1986; Wurtz and Hikosaka, 1986). Grossberg and Kuperstein (1986, 1989) have modelled how this gating action enables planned and attentionally modulated eye movement commands to effectively compete with more rapidly computed visually reactive eye movement commands to decide which type of information will determine where the eye looks in any given situation. In baboons, SN lesions produced a marked increase in the duration of a forelimb pointing movement without causing a change in movement accuracy, and the slowing involved the whole trajectory (Viallet, Trouche, Beaubaton, Nieoullon, & Legallet, 1983), consistent with Equation (2).

6. Target Switching During Movement Sequences By supporting VITE model predictions regarding separate DV and GO signal processes, the data of Georgopoulos et al. and Horak and Anderson also support the more general hypothesis that motor systems, like sensory systems, implement factorization of pattern and energy (Section 1). In the VITE component, this factorization means that a movement's speed ("energy") can be scaled up or down over a wide range without disrupting the movement's direction or spatial endpoint ("pattern"). By using a GO signal that grows gradually during the movement time (Figure 3), all synergists complete their contractions at approximately the same time even if movement onset times of different synergists are staggered by a large amount (Figure 4). These properties of the model, together with the strong evidence for separate DV and GO signal pathways in vivo, provide a basis for understanding how primates can achieve space-time equifinality--all synergists reaching their length targets at equal times--yet retain separate control of rate and position. Ratecontrol models relying on static stiffness adjustments (e.g., Cooke, 1980) lack this critical temporal-equifinality property. A closely related property of the VITE model gains importance during the many occasions when the TPC is updated one or more times during a movement or movement sequence. This may occur, for example,

270

Daniel Bullock and Stephen Grossberg

0.W

0.10

0.20

0 19

0.59

0.49

TIME

TIME

TIME

(A)

(a

Figure 7. A much higher peak velocity is predicted by the model whenever a target, T, is activated after the GO signal has already had time to grow. (A): The control condition, in which T and the GO-signal growth process are activated synchronously. (B): Same T as in (A), but here T was activated after the GO signal G(t) had been growing for 300 ms. if the position of the object to be reached unexpectedly changes. Alternatively, a subject reaching for an object that is initially in the visual periphery may make a better estimate of object location after performing a saccadic eye movement to foveate the object. Saccades take less time than an arm movement that may be unfolding in parallel. In either case, the TPC and DV are rapidly updated, and this late-arriving information affects the arm's trajectory more quickly because the GO signal is already fully developed. Thus the factorization of TPC and GO signal, along with the hypothesis of a gradually growing GO signal, implies that a higher peak velocity will be achieved as a result of a mid-trajectory switch in TPC (Figure 7). Such an amplification of velocity facilitates reaching the target after the incorrect initial TPC is updated. This speed-up occurs "on-the-fly" as the effects of the perturbation flow through the system via

WTE and FLETE: Neural Modules

271

dynamic real-time computations. Georgopoulos et al. (1981) have reported such an increase of peak velocity during target-switching experiments in monkeys. An experiment by Goodale, Pelisson, and Prablanc (1986), analogous to the Georgopoulos et al. (1981) study with monkeys, showed that humans also possess the ability to compensate for in-course target switches. Their experiment was also consistent with an explanation in terms of TPC updating and flow-through, because they eliminated the possibility that corrections could be based on visual comparisons of the relative positions of hand and target. In particular, compensations to a change in target position occurred in the arm’s trajectory even when the arm and hand were invisible to the subject. Fisk and Goodale (1988) have offered an interpretation of lateoccurring in-course error corrections, also proposed by Cooke and Diggles (1984), that is consistent with VITE model mechanisms. They concluded that many terminal error corrections are not based on either proprioceptive feedback from the limb or on visual comparisons of the relative positions of hand and target. Rather, such corrections are based on a comparison made between an internal representation of the target’s locus and an internal representation of the hand’s estimated location based on movement commands. These results support the VITE model as well as the classical hypothesis that even infants typically perform reaches without needing to compare the position of their seen hand with the seen target (Piaget, 1963).

7. From Kinematics To Dynamics: Generating Forces To Ensure That The Arm Tracks The Evolving PPC The VITE circuit places stringent requirements on other components of the sensory-motor system because it requires continuous or nearcontinuous adjustment of the balance of forces acting on the limb to ensure that the limb tracks the evolving PPC without significant lags or overshoots. Some of these components are modelled herein to explain how they autonomously generate the force-time patterns required to track VITE-generated trajectories. When both types of circuits are understood, a quantitative mechanistic understanding of two of the most fundamental problems in sensory-guided motor control would then be approached: how to generate continuously modifiable kinematic plans, and how to generate the continuously modifiable force-time patterns needed to realize them.

272

Daniel Bullock and Stephen Grossberg

8. Factorization Of Length And Tension (FLETE) Model A critical step toward a coherent theory of force-time pattern generation is a principled explanation of the basic spinal-motor-receptor architecture that exists across species. Modelling and computer simulations of this system (Figure 8) suggest that all the major components of the basic architecture play a role in guaranteeing that force (muscle tension) and position (muscle length) are independently controllable by descending control signals of a few simple types. To emphasize this model's central claim of Factorization of LEngth and TEnsion, the model is called the FLETE model. The next sections describe some results that were reported in preliminary form in Bullock and Grossberg (1988d). Past research in this direction has focused on muscle operating characteristics and the alpha-gamma subsystem. An excellent recent fusion of these two traditions can be found in Gielen and Houk (1987), which shows that muscles, receptors, and alpha-gamma circuits can together act as a length servo at low speeds and a velocity servo at high speeds. However, several aspects of the basic system have remained obscure, such as the role of the recurrent inhibition mediated by Renshaw cells, and the contribution of the size principle of motor neuron recruitment. Many recent treatments of Renshaw inhibition relegate it to the role of controlling the gain of the stretch reflex (e.g., McMahon, 1984). FLETE simulations indicate that viewing the Renshaw + MN and Renshaw + Ia -* MN circuits (Figure 8) as merely an "epicycle" to the alpha-gamma circuit is insufficient, and suggest a revised functional basis for the size principle of motor neuron recruitment. A summary of both results follows.

9. Wide Force Range At Each Fixed Muscle Length Requires Size Principle By Newtonian mechanics, the position of a limb is controlled by the balance of forces acting on it. In the body, muscles are the force generators by which this balance is controlled. Muscle is a springy tissue that can actively contract. In a spring, the amount of force depends on the amount of stretch beyond the resting length, the threshold-length for force development. Because muscle can actively contract, muscle has a variable threshold-length for force development. The basic spring-like

YITE and FLETE: Neural Modules

273

P

Figure

Components of the FLETE model of the peripheral skeleto-motor system. Neuron populations comprising two channels together control the contractile states of opponent muscles (AG for agonist, ANT for antagonist) acting on a joint. Descending signal P to both channels allows co-contraction and joint stiffening. Adjusting the balance between descending signals A , and A, allows reciprocal contractions and joint repositioning. For clarity, subpopulations of neurons and some signal pathways are not depicted. Key: Iai = la interneuron population in channel i, i = 1,2; yj = gamma motoneurons; MN, = alpha motoneurons; Ri = Renshaw cells; + = excitatory input; - = inhibitory input.

214

Daniel Bullock and Stephen Grossberg

property can be approximated by the equation

where F is force, L is muscle length, I' is a threshold length, g is a monotone increasing function, and notation [O]+

=

o if w>O 0 if 050.

Because active contraction can lower the threshold, (8) is replaced by

F = g([L -(r -C)]+)

(9)

or equivalently

where C measures the amount of contraction. Note that active contraction does not result in muscle shortening if another force--such as that developed by an opponent muscle--counteracts the contractive force. For the subsequent discussion it is very important to remember that muscles often actively contract without changing their length, as measured from origin to insertion. Suppose that two opponent muscles exert forces FI and F2 from opposite sides of a rotating limb segment (Figure 9) and that the outputs from antagonistic PPC stages in a VITE circuit (Figure 2) are sent to a stage capable of directly adjusting C, and C2 in the following equations for F1 and F2: (11) FI = g([LI -r, +c11+)

where the ri are constant thresholds. If initially F, = F2, then reducing C, and increasing C2 by AC creates a force imbalance such that F2

'F1

275

VITE and FLETE: Neural Modules

Figure 9. Equilibrium joint angle depends on the balance of forces developed by opponent muscles. Each muscle’s force depends on its length L , its resting length ri,and its active contractile state, Ci.

Rotation occurs until

[LI +AL

-rI +CI -AC]+

=

[L2 -M

-r2 +C2

+AC]+

(14)

and a force balance is once again restored. In vivo, the alpha-motoneurons transiently activate contractile fibers from a finite population of fibers. As in equations (11) and (12), let Ci be the degree of contraction, and let Mi be the output signal of the motoneuron pool in channel i, i = 1,2. Then a simple law for change in contraction in a finite population is:

This says that M iincreases Ci by activating unactivated fibers, which number (Bi-Ci),from a population of size Bi,and that contraction spontaneously decays at rate 6.

276

Daniel Bullock and Stephen Grossberg

However, it is known that contracted fibers yield, or decontract, when the force acting to stretch them is sufficiently large. Thus (15) is replaced by

where

o< pi < l a

In equation (16), when the force exceeds the threshold level, TF,it acts to reduce contraction. Inequality (17) acknowledges that any active contraction caused by neural input Mi is slow relative to the fast decontractive effect of suprathreshold forces. The simple model (16) provides a new perspective for understanding the functional role of a widely observed, but imperfectly understood, physiological law. At equilibrium, $Ci = 0 in (16), so that the equilibrium value of Ci is

Given (18), how is it possible to generate and sustain forces much larger than rF at a fired muscle length? Because

F~ = &Li

-ri +ci]+)

(19)

in (11) and (12), where ri is a constant, greater force at a fixed length Li can be generated only by increasing Ci . However, if pi is constant and less than 1, then (18) shows that the negative force feedback will cancel the effects of increasing Mi, and Ciwill not be able to increase, at least if a near linear or faster-than-linear g ( o ) is assumed. To overcome this deficiency within the constraints imposed by equation (16), let the contraction rate parameter pi and the number of sites Bi increase with Mi. Such a relation is well documented empirically and is often called "Henneman's size principle" (Henneman, 1957; 1985): As total excitatory input to the alpha motoneuron population grows, it recruits additional, progressively larger motoneurons which have faster

MTE and FLETE: Neural Modules

277

conducting axons, whose collaterals reach many more motor fibers and whose potentials evoke more rapid muscle contractions. Prior treatments of the function of the size principle have focused on either the contraction-time effects or the force-magnitude effects seemingly implied by activation of cells that project to larger numbers of motor fibers. The present analysis suggests that both aspects of the size principle help realize a large force range at any fixed muscle length. In particular, merely making Bi increase with Mi is not enough. The contraction rate parameter pi also needs to increase with Mi . Thus contractile rate is as critical a component of the system for yielding compensation as the more frequently cited reflex circuits (Houk & Rymer, 1981). Assuming that g(o) is slightly faster than linear, e.g.,

then the system

-ri +ci]+>a

Fi = ( [ L ~

generates a family of curves like those shown in Figure 10 where equilibrium force is plotted as a function of muscle length for four different values of Mi with Pi and Bi constant. Data from experiments in which equilibrium muscle tension (in areflexive muscle) was measured as a function of length and frequency of electrical stimulation have the same form with one difference: the higher stimulation rates lead to slightly larger slopes and higher asymptotic tensions (Rack & Westbury, 1969). From the above analysis, both the slope change and the raised peak tension can be attributed to the size principle: higher stimulation rates evoke faster-contracting fibers, which leads to a higher equilibrium Ci and Fi, and a higher multiplier of Li .

10. Size Principle With Co-Contraction Poses A Threat To Invariant Position Coding The above analysis suggests that the size principle plays a major role in generating a wide range of forces at each fixed muscle length. This

278

Daniel Bullock and Stephen Grossberg

f

SHIFT AS C i INCREASES

Figure 10. In first approximation, the effect of increased muscle stimulation is a shift in the threshold length for force development.

property is important both during high co-contraction conditions when the arm resists changes in position due to variable external forces (e.g., Humphrey & Reed, 1983), and in carrying out planned arm movements wherein the arm undergoes a continuous change of position. In the latter situation, the ability to generate a wide range of forces at each position of a movement trajectory is needed to enable the arm to accurately track the PPC commands that are continuously read-out at variable rates by central circuits. During trajectory formation, a wide range is also needed to compensate for variable forces due either to external perturbations of the arm or to variable arm inertias caused by variable velocities or trajectories of variable shape (e.g., Lestienne, 1979). However, we now show that the size principle, while extending the force range, can also pose a threat to invariant position coding. In particular, we describe an example that shows how the size principle could prevent any simple PPC code from being realized by the arm, such as one based upon adjusting the relative sizes of PPC commands to agonistantagonist muscle groups. We then present the additional neural circuit needed to negate this threat.

K'TE and FLETE: Neural Modules

279

As illustration of the invariance problem, suppose that a limb segment is initially at equilibrium, such that FI = F2

.

where CI(AI)denotes the equilibrium value of Cl when MI = f(A,) in (16). Now suppose that we attempt to hold the limb at the same position, but more rigidly, by increasing the level of muscle contraction on both sides of the joint. The simplest way to do this is by adding a constant, P, to each motoneuron input. Thus M I = f(A,+P) and M2. = f(A2+P). However, in a system that obeys the size principle, equation (23) implies

for arbitrary values of P and the same initial values of L, only if A , = A2 (see explanation in Figure 11). In other words, sending the same cocontractive input P to both motoneuron pools in an attempt to further stabilize current limb position could instead cause a limb rotation. This is a prime example of a failure of factorization of length and tension: an attempt to change only tension inadvertently changes length. Because the motor cortex appears to follow the simplest strategy (adding a constant P ; see section 14), some mechanism must exist to prevent unwanted limb rotations. In the light of this problem, many researchers have proposed that Ciand Li should interact multiplicatively to produce force. Though this would reduce the problem, the proposal amounts to a claim that the primary effect of changing Mi is a change in the stiffness (AF/AL) of areflexive (deafferented) muscle. However, experimental data show that stiffness changes relatively little as Mi changes; the primary effect of changing Miis a change in the threshold length for force development, as suggested in Equation (9) and Figure 10 (Feldman, 1986; Rack & Westbury, 1969).

280

Daniel Bullock and Stephen Grossberg

'Figure 11. When opponent motoneuron populations obey the size princiMOTONEURON POOL (A) ACTIVATION LEVEL ple, a co-contractive sigBIG CELL nal P sent in parallel to ZONE both populations can RECRUITMENT THRESHOLD disrupt the joint position SMALL CELL code. (A) Signals A, and ZONE A1 A, supraliminally activate only small cells in opposing channels and their relative sizes determine BIG CELL the balance of muscular ZONE forces and thus the equiRECRUITMENT THRESHOLD librium joint position. (B) SMALL CELL With A, > A,, co-conZONE tractive signal P causes *1 P the total input A, + P to exceed the big cell threshold while input A, + P remains below the big cell threshold. Thus part of the signal P is subjected to greater amplification in channel 1 than in channel 2. Unless compensated, this would create a new balance of forces and cause an unwanted joint rotation.

11. Compensatory Properties Of Renshaw-Ia Pathway An alternative solution may be sought in known neural circuits. Is there any neural site that is sensitive to the amplification factor (Figure 11) introduced by progressive recruitment within an alpha-motoneuron population? The alpha-gamma system cannot provide the type of compensation desired because gamma motoneurons are not sensitive to the amplification. However, inspection of Figure 8 reveals the Renshaw cells as sole neural targets of a-MNaxon collaterals, and that a-MN cells receive feedback signals from Renshaw cells. Thus Renshaw cells are well situated to measure and modify the final output of the motor channel. Moreover, it is known that collaterals of larger, later recruited motoneurons make many more synaptic contacts with Renshaw cells than smaller, earlier recruited motoneurons (Cullheim and Kellerth, 1978). These properties are consistent with the hypothesis that Renshaw cells

VlTE and FLETE: Neural Modules

281

play the required compensatory role. We now present computer simulations of the system in Figure 8 which show how model Renshaw cells can compensate for position code distortions that would otherwise be generated when co-contraction is combined with the size principle of motoneuron recruitment. This property depends upon how,the feedback pathways in the circuit are organized into opponent, or antagonistic, muscle channels.

Table 1 Major FLETE Model Variables

Descending reciprocal input to alpha motoneuron and Ia interneuron population i, i = 1,2 Descending co-contractive input to both alpha motoneuron populations Force developed by muscle i, i = 1,2 Contractile state of muscle i, i = 1,2 Origin-to-insertion length of muscle i, i = 1,2 Alpha-MN population activity, i = 1,2 Renshaw population activity, i = 1,2 IaIN population activity, i = 1,2 Composite signal from spindle organs, i = 1,2

The compensation occurs as follows. Opponent Renshaw populations R, and R, measure the output of their respective alpha-motoneuron populations, a-MN,, and a-MN,, and compare those outputs via mutually inhibitory signals (Figure 8). A consensus emerges regarding which MN channel to inhibit via Renshaw feedback, and which to disinhibit via feedback from the Ia interneuron (IaIN) pathway. Suppose that a cocontractive input, P, to a-MN, and a-MN2 occurs when input A, exceeds Suppose that the activity of a-MN, is consequently multiplied by a larger factor than that of a-MN, due to the size principle (Figure 11). Then R, also becomes much more active due to the size-correlated synaptic weighting on a-MN, axon collaterals to R,. Because the opposing R, has not experienced as large an input increment, R, will transiently become more active than R , by an amount that scales with the

A,.

282

Daniel Bullock and Stephen Grossberg

diflerence between the a-MN output increments due to the change in P. Thus this system calculates an error due to unequal amplifications of cocontractive inputs. This error signal then directly inhibits a-MN1 and, by inhibiting laIN1, indirectly activates a-MN,. Both actions work to zero the error without negating either the shared increment in a-MNi activation required to increase joint stiffness, or the joint angle setting determined by the difference in descending inputs, exclusive of P, to opponent a-MN and IaIN populations. Readers not interested in the details of the simulations should skip ahead to the Results subheading. Others may consult Table 1 for definitions relevant to equations (25)-(37). As in Figure 9, we assumed a rotary joint affected by two opponent muscles, each of which is inserted in the moving segment one unit from the axis of rotation. The distance from muscle origin to the axis of rotation was 20 units, and the midpoint of the limb's 180" excursion was stipulated to be at joint angle 0 = 0". Origin-to-insertion muscle lengths, L , were thus functions of 0:

In the simulations reported here, we were interested only in large-scale effects on equilibrium joint angle. Thus we ignored moment-arm and force-velocity effects and chose the simple force law

Fi = k[Li -ri +CJ+ where k = .5, ri = 20.9 and i = 1,2. Limb dynamics were governed by equation

zd2G0 = -m1 (F1 -F2 - n dx0 ) where rn represents mass and n is a damping coefficient. Cbntradle state Ciwas governed by

U T E and FLETE: Neural Modules with 6 = 1, and ,?I

= 0. Variables pi and

pi = .05

+ .02(Ai

283

Biwere defined by:

+P)

(29)

Both variables grow as a function of total descending input Ai+P to the MN pools in channel i, but pi grows with a smaller slope. Equations (16), (29), and (30) use parameters pi and Bi to approximate a-MN recruitment effects that occur in viva Which a-MNs exceed threshold may depend not only on the total descending drive, Ai+P, as in (29) and (30), but also on inhibitory inputs from IaINs and Renshaw cells as well as on inputs arising from sensory organs in the muscle. In this lumped model, sensory feedback was omitted to isolate the potential compensatory effect of the Renshaw pathway. The lumped model does, however, include the critical assumptions that Renshaw populations associated with opponent muscles are mutually inhibitory, and that each Renshaw populations’s activity be sensitive to the amplification factor introduced by recruitment of larger motoneurons. Such sensitivity requires that growth in a Renshaw population’s activity not saturate prior to saturation of growth in a-MN population activity, and that the input to the Renshaw population scale with the net amplification due to recruitment. As noted above, there is evidence that this scaling is effected in vivo by increasing the synaptic weighting factor associated with Renshaw-directed axon collaterals from larger a-MNs. In our lumped model, this effect was absorbed into a single variable coefficient, zi, which was made a function of recruitment extent, as approximated by A,+P. The lumped equations for opponent Renshaw populations were thus

$ R, = (lB1 -Rl)z1Ml

-R1(1 +R2)

with 1 = 5 and

zi = .2

+ .8 (Ai+P)

(33)

284

Daniel Bullock and Stephen Grossberg

for i = 1,2. Parameter Bi in (31) and (32) approximates the property that the Renshaw population has a continuum of recruitment thresholds similar to the a-MN population, and that the number of suprathreshold Renshaw sites increases as more a-MNs are recruited (with increasing Ai+P). We modeled the opponent alpha-motoneuron populations via

where 9 = .2, x = 0, and 52 = 0 or 1. The inhibitory Zi inputs represent signals from the IaINs (see Figure 8) and the excitatory El inputs represent signals from the muscle spindles. We assumed that IaINs were not subject to any recruitment effects. Thus their dynamics were modeled without a direct dependence on BP and without a co-activating input P:

3

I = @(lo -Z2)(A2 +xE2) -Z2(1 2

+m, +I,)

(37)

Composite spindle feedback signals E, and E2 (in Equations 3437) were gated off in our simulations by setting x = 0. This corresponds to destroying the stretch reflex via deafferentation, and it allowed us to test the ability of the Renshaw-Ia-MN feedback circuit to achieve position code invariance. The Renshaw feedback signals were gated on or off, respectively, by setting parameter C2 (in Equations 34-37) equal to 1 or 0.

Results Table 2 shows representative numerical results. Variables A l , A,, and P (see Equation 24) represent constant inputs and variables L,, F,, and 0 represent equilibrium values of dependent variables. Because L, and L, are complements and F, = F2 at equilibrium, L2 and F2 are omitted from the table. Because small length changes can imply large joint rotations, the most informative column in Table 2 is that showing

285

U T E and FLETE: Neural Modules

0,the equilibrium joint angle. When Renshaw feedback was absent (Q = 0), changing P while A , and A, remained fixed led to large undesirable rotations (10" in block 1, 22" in block 3). However when Renshaw feedback was present (Q = l),rotations due to changing P with fixed A, and A, were extremely small ( < l o in block 2 and <4". in block 4). Generally, when the system was not forced to operate in the saturation range, excursions were
Table 2

FLETE Model Simulations: Wide Force Range at Each Length and Independent Control of Force and Length When 52 = 1

0 0 0

.12 .12 .12

.1 .1 .1

.o

1

.1 .1 .1

.o

1

.14 .14 .14

0 0 0

.14 .14 .14

1 1 1

.22 .22 .22

1

19.82 19.66 19.70

0.02 0.32 7.8

.1 .8

19.88 19.87 19.89

0.002 0.13 3.1

.1 .1 .1

.O .1 .8

19.58 19.26 19.37

0.04 0.36 7.9

26.0 48.5 39.9

.1 .1 .1

.o

19.55 19.54 19.61

0.03 0.19 3.4

28.2 28.9

.1 .8

.1 .8

11.6 21.5 18.8 8.27 9.16 7.89

24.5

286

Daniel Bullock and Stephen Grossberg

Because all stretch feedback was turned off in these simulations, this property of Renshaw feedback may not be described as "controlling the gain of the stretch reflex." Indeed, from the current perspective, the tonic stretch reflex mediated by type Ia and I1 feedback fibers from length-sensitive spindle organs might be viewed as a secondary system designed to compensate for residual errors left uncompensated by the Renshaw-Ia system. This impression is reinforced by data (Fromm, Haase & Wolf, 1977) indicating that type I1 fibers from spindles activate a-MN populations and inhibit Renshaw populations in their own outflow channel. This means that if Renshaw feedback is either too strong or too weak to fully correct cocontraction-related positioning errors, Renshaw feedback gain will be automatically adjusted in the correct direction by type I1 feedback from muscle spindles. This pathway would also help compensate for moment-arm effects not included in our simulations (see Hasan & Enoka, 1985).

12. Evidence For Assumed Distribution Of Renshaw Connectivity Two critical hypotheses of our model are (a) that Renshaw cells participate in the size principle and (b) that the computational unit is the pair of opponent muscle channels. A variety of evidence supports the hypothesis that Renshaw cells participate in the size principle. In the simulations that generated Table 2,. this assumption was implemented by scaling up Renshaw population parameters in parallel with motor-unit rescaling as Ai +P grows. Recent experiments surveyed by Pompeiano (1984) concur that ''recurrent inhibition is produced mainly by large phasic neurons that are recruited late" (p. 526). In particular, Pompeiano and Wand (1976; Wand and Pompeiano, 1979) produced functional evidence for such a size-dependency, and Cullheim and Kellerth (1978) produced convergent anatomical evidence by showing that larger, phasic motoneurons make many more synaptic contacts with Renshaw cells than smaller, tonic motoneurons. The second hypothesis has been well supported since Sherrington's (1906) observations of reciprocal inhibition, but is oddly ignored in many treatments. Our treatment extends the reciprocal inhibition principle, which is a "biggest competitor wins" principle at the IaIN stage (Figure 8), by including Renshaw populations which compete before supplying inhibitory feedback to the model's IaINs and alpha-motoneurons (Miller & Scott, 1977; Pompeiano, 1984). Because the channel with the larger

U T E and FLETE: Neural Modules

287

Renshaw activity receives more inhibition, reciprocal inhibition at the Renshaw stage follows a "biggest competitor loses" principle. This property extends the classical role of the Renshaws in stabilizing the peripheral skeleto-motor system; Renshaw inhibition works against extreme joint angle excursions and complements the intrinsic damping characteristics of muscles. More generally (Figure 8), the FLETE model assumes that Renshaw cells have an inhibitory effect at three sites. The recurrent inhibition to the alpha-motoneuron population that excites them has long been well known (Renshaw, 1941; Eccles, Fatt, & Koketsu, 1954; see Pompeiano, 1984, for recent review). Inhibition of the IaIN population in the same outflow channel was demonstrated by Hultborn, Jankowska, and Lindstrom (1971) and confirmed by others (see Pompeiano, 1984, pp. 512513). Renshaw inhibition of the Renshaw population of the opposing muscle channel, suspected since Renshaw (1946), has been convincingly demonstrated by Ryall (1970). Though there is also evidence that Renshaws have an inhibitory effect on gamma motoneurons (Pompeiano, 1984, pp. 509-511), the effect is known to be attenuated relative to that on alpha-MNs. Though nearly all alpha-MNs are inhibited by Renshaws, only about half of gamma-MNs are so inhibited, and in lesser degree. The model also assumes that Renshaw cells are directly affected by an excitatory input from the alpha-motoneurons (see Renshaw, 1941), and an inhibitory input from the opposing-channel's Renshaw cells (noted above). The inhibitory stretch feedback noted earlier from spindle organs via group I1 fibers, which carry a length-correlated signal, will be incorporated in future simulations. In this connection, we note that a descending inhibitory input from the red nucleus to Renshaw populations is also well documented (Henatsch, Meyer-Lohman, Windhorst, & Schmidt, 1986). This inhibitory red nucleus output is coupled with another rubral output that excites alpha-motoneurons in the same outflow channel. Thus this descending rubral signal is analogous to the inflowing type I1 spindle signal. If this parallel rubral output is a reciprocal command (always increasing in one channel while decreasing in the opposing channel), it can be seen to be part of a feedforward adaptive gain control system (Grossberg & Kuperstein, 1986), which gradually learns to supply predictively the compensations the peripheral circuit can only supply reactively (see also Bullock, Carpenter, & Grossberg, 1989; Kawato, Furukawa, and Suzuki, 1987).

Daniel Bullock and Stephen Grossberg

13. Prior Proposals Regarding Renshaw Function Proposals regarding Renshaw function have evolved rapidly in recent years. Shepherd (1979) acknowledged that their function remained mysterious despite the long-standing hypothesis that they might serve as a source of surround inhibition (and thus perhaps to contrast enhance the motor-output signal). In the same year Hultborn et al. (1979) proposed that the Renshaws were well situated to control the gain of the alpha-motoneuron pool's response to excitatory inputs. This proposal was often restated in terms of controlling the gain of the stretch reflex (e.g., McMahon, 1984), a picture since reinforced by discovery of the descending (rubrospinal) pathways that both inhibit Renshaw cell activity (thus disinhibiting alpha-motoneurons) and excite alpha-motoneurons, resulting in a higher-gain stretch reflex (Henatsch et al., 1986) among other effects. The common scenario imagined for such Renshaw modulation was during muscle contraction intended to produce movement. Thus this consensus proposal is not in conflict with the present proposal, in which muscle contraction intended to prevent movement is seen to require that the Renshaw pathway not be inhibited by descending signals. Rather, our proposal can be seen as the logical complement to the consensus view, and as further evidence for the need to have mechanisms (Section 4) that automatically define active posture, active movement, and passive movement as distinct computational states. A model by Miller and Scott (1977) shares our emphasis on competition between opponent Renshaw populations. However, the authors assumed that such competition implicated the Renshaw-la pathway in locomotor pattern generation, a different function than the one here proposed. Subsequent research (Pratt & Jordan, 1987) appears to have ruled out the possibility that the Renshaw-la pathway is part of a spinal locomotor generator. Finally, though some aspects of our model are similar to Feldman's (1986) well known "A" model of skeleto-motor control, neither of our descending control signals, Ai and P, correspond to Feldman's stretchreflex parameter 1. Moreover, we believe that continued use of lumped parameters like A, and a kindred overemphasis on the concept of stretch reflex, may hinder attempts to understand how the neuromuscular system is parsed into functional subsystems. A case in point is the discovery, upon unlumping reciprocal and co-contractive inputs, that the Renshaw-

U T E and FLETE: Neural Modules

289

Ia pathway may play a role far more interesting than being an epicycle of the stretch-reflex.

14. Physiological Evidence For Separate Cortical Control Of Non-Selective Co-Contractive Input To Motoneurons FLETE model simulations of Renshaw function were based on the assumption that the co-contractive signal, P,is sent in parallel and without differential weighting to small and large MNs alike in both outflow channels (Figure 11). This hypothesis is supported by data of Humphrey and Reed (1983), who subjected monkeys to high-frequency, alternatingdirection, torque perturbations at the wrist joint after they trained the monkeys to actively maintain their wrist angle within a small angular tolerance zone. To prevent the imposed torques from rotating their wrists to angles outside the desired range, monkeys instated high levels of tonic co-contraction in wrist flexors and extensors. Measurements of motor unit activity showed that high levels of co-contraction were achieved nonselectively and in accord with the size principle. In particular, Humphrey and Reed (1983) concluded that "Asthe speed of joint perturbation rises, the modulated [reciprocal] input to the MN pools is increased and a tonic coactivation signal is added....Thus, an explanation of our observed MN firing patterns requires no assumption of selectivity of descending inputs to motor units of different type, nor of any recruitment order different from that established in previous studies....Both [reciprocal and coactivating] control signals appear to converge on both fast and slowtwitch MNs" (p.366). Humphrey and Reed (1983) were also able to identify a central source of co-activation signals. In Section 4, we cited evidence from Georgopoulos that precentral motor cortex (Area 4) served as a site of VITE-like DV computations and thus as a source of reciprocal commands received by spinal motor centers. Whereas Humphrey and Reed (1983) observed similar reciprocally-engaged precentral cells, they also discovered a new class of tonically active neurons they called S" cells (S = steady, A = shift). These neurons, also found in precentral Area 4, predominated in a zone slightly anterior to the DV-like cells, and "when the animal voluntarily co-contracted his wrist muscles, as in stabilization of the wrist or tightening of a grip on the handle, these cells discharged in a brisk and tonic manner" (Humphrey & Reed, 1983, p.363). Moreover, microstimulation (12 to 20 ,LA) in the anterior, S" cell, zone evoked a co-

290

Daniel Bullock and Stephen Grossberg

activation of flexor and extensor muscles at the wrist and in some cases at other arm joints. Thus the primary motor cortex seems to be a source of both the specific (reciprocal) and the non-specific (co-contractive) signals assumed to ultimately converge on the spinal motoneurons in the FLETE model.

15. Tracking A Rapidly Changing PPC, And The Genesis Of Tri-Phasic EMG Bursts Computer simulations have also been carried out of how the peripheral skeleto-motor system, as modelled by the F L E E equations, responds to a fast ramp input, similar in shape to the VITE circuit output in response to a large GO-signal. These simulations were directed at the question of whether a quickly changing monotonic ramp input would lead to a multiphasic burst pattern at the alpha motoneuron stage of the model. In vivo, a so-called "tri-phasic" burst pattern is typically observed during fast movements (Lestienne, 1979). If such a triphasic pattern cannot emerge from an interaction of a monotonic descending signal with feedback within the peripheral skeleto-motor system, then any simple form of the VITE model would be untenable. Feldman (1986) presented a qualitative analysis suggesting that the peripheral system could compute the monotonic-to-multiphasic transform given appropriate parameters. The FLETE simulations indicate that a triphasic burst pattern of the correct form (large AG1 burst, then large ANT burst, then much smaller AG2 burst) does emerge when there is strong velocity-dependent feedback via the Ia pathway to alpha motoneurons and IaINs. This issue is currently controversial (see Berkinblit, Feldman, & Fukson, 1986) because bursts may also be observed in the absence of Ia feedback signals. The controversy may stem from the fact that after their genesis by reactive, feedback systems within the peripheral FLETE circuits, the burst pattern may be "copied" into and amplified by a supplementary feedforward command channel by adaptive predictive learning of movement gains that is mediated by the cerebellum (Grossberg & Kuperstein, 1986). Thus after learning, the burst pattern can occur even after Ia afferent feedback is eliminated. In addition, simulations of the Renshaw-IaIN pathway suggest that bursting can emerge under certain conditions ("Renshaw size-principle") without participation by Ia feedback from spindle organs. Preliminary studies suggest that this effect could account for sub-bursts that have been observed in experiments utilizing EMG

U T E and FLETE: Neural Modules

29 1

recording techniques with high temporal resolution (Brown & Cooke, 1986), as well as bursts appearing in isometric studies (Gordon & Ghez, 1987). Even isometric studies present interpretive difficulties however, because intrafusal muscle activation via gamma motoneurons can evoke spindle discharge and Ia feedback in the absence of joint rotations. Detailed treatments of these simulations as well as a fuller report on refinements of the Table 2 simulations will appear elsewhere (Bullock & Grossberg, 1989a,b).

16. Conclusions In the VITE circuit, a single GO signal sent in parallel to a large number of primed muscle-control channels can initiate goal-oriented synchronous movement, and control its rate without disrupting its form. In the FLETE circuit, a single co-contraction signal sent in parallel to a large number of muscle channels can control joint rigidity without disrupting postural stability. Both circuits clarify an old mystery in the theory of volitional action: if every act is so complex, why do volitional acts, or acts of "will," seem to be so simple? Once such invariance-preserving components evolve, they can be expected to be incorporated into many subsequent evolutionary specializations (Bullock, 1987; Grossberg & Kuperstein, 1986, 1989; Lieberman, 1984; Powers, 1973; Simon, 1969). Here and elsewhere, we have argued that the VITE architecture or close variants may have been replicated across many systems which control phasic goal-oriented movements, including both arm and speech movements, and the circuit of Figure 8, which is mathematized in the FLETE model, is known to exist throughout the higher vertebrates. Similarly, the cerebellum serves as an adaptive gain control stage in a wide range of motor systems (Grossberg & Kuperstein, 1986, 1989; Ito, 1984; Kawato, Furukawa, & Suzuki, 1987). Despite initial appearances of overwhelming complexity, perhaps we may reasonably hope that the discovery of a modest number of robust and broadly applicable circuits will allow us to explain a large portion of the basic motor competence of higher vertebrate species.

Acknowledgements: We thank Carol Yanakakis for her expert help in preparing the manuscript.

292

Daniel Bullock and Stephen Grossberg

References Anderson, M. E., & Horak, F. B. (1985). Influence of the globus pallidus on arm movements in monkeys. 111. Timing of movement-related information. Journal of Neurophysiology, 54, 433-448. Beggs, W. D. A., & Howarth, C. I. (1972). The movement of the hand towards a target. Quarterly Journal of Experimental Psychology, 24, 448-453. Berkinblit, M. B., Feldman, A. G., & Fukson, 0. I. (1986). Adaptability of innate motor patterns and motor control mechanisms. Behavioral and Brain Sciences, 9, 585-638. Bizzi, E., Accornero, N., Chapple, W., & Hogan, N. (1984). Posture control and trajectory formation during arm movement. Journal of Neuroscience, 4 ( I I ) , 2738-2744. Bullock, D. (1987). Socializing the theory of intellectual development. In M. Chapman & R. A. Dixon (Eds.), Meaning and the growth of understanding: Wittgenstein's significance for developmentalpsychology. New York: Springer-Verlag, 187-218. Bullock, D., Carpenter, G. A., & Grossberg, S. (1989). Self-organizing neural network architectures for adaptive pattern recognition and robotics. In V. Milutinovic & P. Antognetti (Eds.) Neural networks: Concepts, applications and implementations. Englewood Cliffs, NJ: Prentice Hall. Bullock, D., & Grossberg, S. (1986). Neural dynamics of planned arm movements: Synergies, invariants, and trajectov formation. Paper presented at the symposium on Neural Models of Sensory-Motor Control at the annual meeting of the Society for Mathematical Psychology, Cambridge, MA, August 20. Bullock, D., & Grossberg, S. (1988a). Neural dynamics of planned arm movements: Emergent invariants and speed-accuracy properties during trajectory formation. Psychological Review, 95( l), 49-90. Bullock, D., & Grossberg, S. (1988b). The VITE model: A neural command circuit for generating arm and articulator trajectories. In J. A. S. Kelso, A. J. Mandel, & M. F. Shlesinger (Eds.), Dynamic patterns in complex systems (pp. 305-326). Singapore: World Scientific. Bullock, D., & Grossberg, S. (1988~).Self-organizing neural architectures for eye movements, arm movements, and eye-arm coordination. In H. Haken (Ed.), Neural and synergetic computers (pp. 197-228). Berlin: SpringerVerlag. Bullock, D., & Grossberg, S. (1988d). Neuromuscular realization of planned trajectories. Neural Networks, I , Supplement 1, 329. Bullock, D., & Grossberg, S. (1989a). An opponent neuromuscular design for

factorization of length and tension, I: Cocontraction, the size principle, and recurrent inhibition. Manuscript submitted for publication.

WTE and FLETE: Neural Modules

293

Bullock, D., & Grossberg, S. (1989b). An opponent neuromuscular design for

factorization of length and tension, IZ: Force-time pattern generation during movement. Manuscript submitted for publication. Brown, S. H., & Cooke, J. D. (1986). Initial agonist burst is modified by perturbations preceding movement. Bruin Research, 377, 311-322. Carpenter, G. A., & Grossberg, S. (1987). Art 2: self-organization of stable category recognition codes for analog input patterns. Applied Optics, 26(23), 4919-4930. Carpenter, G. A., & Grossberg, S. (1988). The ART of adaptive pattern recognition by a self-organizing neural network. Computer, 21, 77-88. Cooke, J. D. (1980). The organization of simple, skilled movements. In G. E. Stelmach & J. Requin (Eds.), Tutorials in motor behavior. Amsterdam: North-Holland, 199-212. Cooke, J. D., & Diggles, V. A. (1984). Rapid error correction during human arm movements: Evidence for central monitoring. Journal of Motor Behavior, 16, 348-363. Cullheim, S., & Kellerth, J. 0. (1978). A morphological study of the s o n s and recurrent axon collaterals of cat a-motoneurones supplying different functional types of muscle unit. Journal of Physiology (London), 281, 301-313. Eccles, J. C., Fatt, P., & Koketsu, K. (1954). Cholinergic and inhibitory synapses in a pathway from motor-axon collaterals to motoneurones. Journal of Physiology (London), 126, 524-562. Evarts, E. V., & Fromm, C. (1978). The pyramidal tract neuron as summing point in a closed-loop control system in the monkey. In J. E. Desmedt (Ed.), Cerebral motor control in man: Long-loop mechanisms (pp. 56-69). Basel, Switzerland: Karger. Feldman, A. G. (1986). Once more on the equilibrium point hypothesis (A model) for motor control. Journal of Motor Behavior, 18, 17-54. Fisk, J. D., & Goodale, M. A. (1988). The effects of unilateral brain damage on visually guided reaching: hemispheric differences in the nature of the deficit. Experimental Brain Research, 72, 425-435. Fitts, P. M. (1954). The information capacity of the human motor system in controlling the amplitude of movement. Journal of Experimental P ~ c h o l o 47(6), ~ , 381-391. Freund, H. -J., & Biidingen, H. J. (1978). The relationship between speed and amplitude of the fastest voluntary contractions of human arm muscles. Experimental Brain Research, 31, 1-12. Fromm, C., Haase, J., & Wolf, E. (1977). Depression of the recurrent inhibition of extensor motoneurones by the action of group I1 afferents. Brain Research, 120, 459-468.

294

Daniel Bullock and Stephen Grossberg

Gahery, Y., & Massion, J. (1985). Co-ordination between posture and movement. In E. V. Evarts, S. P. Wise, & D. Bousfield (Eds.), The motor system in neurobiology (pp.121-125). Amsterdam: Elsevier. Georgopoulos, A. P., Kalaska, J. F., & Massey, J. T. (1981). Spatial trajectories and reaction times of aimed movements: Effects of practice, uncertainty, and change in target location. Journal of Neurophysiology, 46(4), 725-743. Georgopoulos, A. P., Kalaska, J. F., Caminiti, R., & Massey, J. T. (1984). The representation of movement direction in the motor cortex: Single cell and population studies. In G. M. Edelman, W. E. Gall, & W. M. Cowan (Eds.), Dynamic aspects of neocoitical function (pp. 501-524). New York: Wiley. Gielen, C. C. A. M., & Houk, J. C. (1987). A model of the motor servo: Incorporating nonlinear spindle receptor and muscle mechanical properties. Biological Cybernetics, 57, 217-231. Goodale, M.A., Pelisson, D., & Prablanc, C. (1986). Large adjustments in visually guided reaching do not depend on vision of the hand or perception of target displacement. Nature, 320, 748-750. Gordon, J., & Ghez, C. (1987). Trajectory control in targeted force impulses, 111: Compensatory adjustments for initial errors. Experimental Brain Research, 67, 253-269. Grossberg, S. (1970). Neural pattern discrimination. Journal of Theoretical Biology, 27, 291-337. Grossberg, S. (1973). Contour enhancement, short-term memory, and constancies in reverberating neural networks. Studies in Applied Mathematics, 52, 217-257. Grossberg, S. (1978). A theory of human memory: Self-organization and performance of sensory-motor codes, maps, and plans. In R. Rosen & F. Snell (Eds.), Progress in theoretical biology, Vol. 5 (pp. 233-374). New York: Academic Press. Grossberg, S. (1982). Studies of mind and brain: Neural principles of learning, perception, development, cognition, and motor control. Boston: Reidel Press. Grossberg, S., & Kuperstein, M. (1986). Neural dynamics of adaptive sensorymotor control: Ballistic eye movements. Amsterdam: Elsevier/NorthHolland. Grossberg, S., & Kuperstein, M. (1989). Neural dynamics of adaptive sensorymotor control: Expanded Edition. New York: Pergamon Press. Hasan, Z., & Enoka, R. M. (1985). Isometric torque-angle relationship and movement-related activity of human elbow flexors: implications for the equilibrium-point hypothesis. Experimental Brain Research, 59, 441-450.

V7TE and FLETE: Neural Modules

295

Henatsch, H. D., Meyer-Lohmann, J., Windhorst, U., & Schmidt, J. (1986). Differential effects of stimulation of the cat’s red nucleus on lumbar alpha motoneurons and their Renshaw cells. Experimental Brain Research, 62, 161- 174. Henneman, E. (1957). Relation between size of neurons and their susceptibility to discharge. Science, 26, 1345-1347. Henneman, E. (1985). The size-principle: A deterministic output emerges from a set of probabilistic connections. Journal of Experimental Biology, 115, 105-112. Hollerbach, J. M., Moore, S. P., & Atkeson, C. G. (1986). Workspace effect in arm movement kinematics derived by joint interpolation. In G. Gantchev, B. Dimitrov, & P. Gatev (Eds.), Motor control. Plenum Press. Horak, F. B., & Anderson, M. E. (1984a). Influence of globus pallidus on arm movements in monkeys, I. Effects of kainic acid-induced lesions. Journal of Neurophysiology, 52, 290-304. Horak, F. B., & Anderson, M. E. (1984b). Influence of globus pallidus on arm movements in monkeys, 11. Effects of stimulation.J o u m l of Neurophysiology, 52, 305-322. Houk, J. C., & Rymer, W. Z. (1981). Neural control of muscle length and tension. In Handbook of physiology: The nervous system 11 (pp. 257-322). Bethesda, MD: American Physiological Society. Hultborn, H., Jankowska, E., & Lindstrom, S. (1971). Relative contribution from different nerves to recurrent depression of Ia IPSPs in motoneurones. Journal of Physiology (London), 215, 637-664. Hultborn, H., Lindstrom, S., & Wigstrom, H. (1979). On the function of recurrent inhibition in the spinal cord. Experimental Brain Research, 37, 399-403. Humphrey, D. R., & Reed, D. J. (1983). Separate cortical systems for control of joint movement and joint stiffness: Reciprocal activation and coactivation of antagonist muscles. In J.E. Desmedt (Ed.), Motor control mechanisms in health and disease (pp. 347-372). New York: Raven Press. Ito, M. (1984). The cerebellum and neural control. New York: Raven Press. Kawato, M., Furukawa, K., & Suzuki, R. (1987). A hierarchical neural-network model for control and learning of voluntary movement. Biological Cybernetics, 57, 169-185. Kuperstein, M. (1988). Neural network model for adaptive hand-eye coordination for single postures. Science, 239, 1308-1311. Lestienne, F. (1979). Effects of inertial load and velocity on the braking process of voluntary limb movements. Experimental Brain Research, 35, 407-418.

296

Daniel Bullock and Stephen Grossberg

Lieberman, P. (1984). The biology and evolution of language. Cambridge, MA: Harvard University Press. Mateer, C. (1978). Asymmetric effects of thalamic stimulation on rate of speech. Neuropsychologia, 16, 497-499. McMahon, T. A. (1984). Muscles, reflexes, and locomotion. Princeton, NJ: Princeton University Press. Miller, S.,& Scott, P. D. (1977). The spinal locomotor generator. Experimental Brain Research, 30, 387-403. Mitchell, S. J., Richardson, R. T., Baker, F. H., & DeLong, M. R. (1987). The primate globus pallidus: neuronal activity related to direction of movement. Experimental Brain Research, 68, 491-505. Moore, S. P., & Marteniuk, R. G. (1986). Kinematic and electromyographic changes that occur as a function of learning a time-constrained aiming task. Journal of Motor Behavior, 18, 397-426. Neafsey, E. J., Hull, C. D., & Buchwald, N. A. (1978). Preparation for movement in the cat, 11. Unit activity in the basal ganglia and thalamus. Electroencephalography and Clinical Neurophysiology, 44, 714-723. Ostry, D. J., Cooke, J. D., & Munhall, K. G. (1987). Velocity curves of human arm and speech movements. Experimental Brain Research, 68, 37-46. Passingham, R. E. (1987). From where does the motor cortex get its instructions? In S. P. Wise (Ed.), Higher brain &nctions (pp. 67-97). New York: Wiley. Penney, J. B., & Young, A. B. (1983). Speculations on the functional anatomy of basal ganglia disorders. Annual Review of Neuroscience, 6 , 73-94. Piaget, J. (1963). The origins of intelligence in children. New York: Norton. Pierrot-Deseilligny, E., & Morin, C. (1980). Evidence for supraspinal influences on Renshaw inhibition during motor activity in man. In J. E. Desmedt (Ed.), Spinal and supraspinal mechanisms of voluntary motor control and locomotion. Basel: Karger. Pompeiano, 0. (1984). Recurrent inhibition. In R.A. Davidoff (Ed.), Handbook of the spinal cord, Vols. 2 and 3. Anatomy and physiology. New York: Marcel Dekker. Pompeiano, O., & Wand, P. (1976). The relative sensitivity of Renshaw cells to static and dynamic changes in muscle length. Progress in Brain Research, 44, 199-222. Powers, W. T. (1973). Behavior: The control of perception. Chicago: Aldine. Pratt, C. A., & Jordan, L. M. (1987). Ia inhibitory interneurons and Renshaw cells as contributors to the spinal mechanisms of fictive locomotion. Journal of Neurophysiology, 57, 56-71. Rack, P. H. M., & Westbury, D. R. (1969). The effect of length and stimulus rate on the tension in the isometric cat soleus muscle. Journal of Physiology, 204, 443-460.

U T E and FLETE: Neural Modules

297

Renshaw, B. (1946). Central effects of centripetal impulses in axons of spinal ventral roots. Journal of Neurophysiology, 9, 191-204. Renshaw, B. (1941). Influence of discharge of motoneurons upon excitation of neighboring motoneurons. Journal of Neurophysiology, 4, 167-183. Ryall, R. W. (1970). Renshaw cell mediated inhibition of Renshaw cells: Patterns of excitation and inhibition from impulses in motor axon collaterals. Journal of Neurophysiology, 33, 257-270. Schwartz, A. B., Kettner, R. E., & Georgopoulos, A. P. (1988). Primate motor cortex and free arm movements to visual targets in three-dimensional space. I. Relations between single cell discharge and direction of movement. Journal of Neuroscience, 8, 2913-2927. Shepherd, G. M. (1979). The synaptic organization of the brain. New York: Oxford. Sherrington, C. S. (1906). The integrative action of the nervous system. New Haven: Yale University Press. Simon, H. A. (1969). The sciences of the artificial. Cambridge, MA: MIT Press. Sparks, D. L., & Jay. M. F. (1986). The functional organization of the primate superior colliculus: A motor perspective. In H.-J. Freund, U. Butner, B. Cohen, & J. Noth (Eds.), The oculomotor and skeletal-motor systems (pp.235-241). Amsterdam: Elsevier. Van Buren, J. M., Li, C. L., & Ojemann, G . A. (1966). The fronto-striatal arrest response in man. Electroencephalographyand Clinical Neurophysiology, 21, 114-130. Viallet, F., Trouche, E., Beaubaton, D., Nieoullon, A., & Legallet, E. (1983). Motor impairment after unilateral electrolytic lesions of the Substantia Nigra in baboons: Behavioral data with quantitative and kinematic analysis of a pointing movement. Brain Research, 279, 193-206. Wand, P., & Pompeiano, 0. (1979). Contribution of different size motoneurons to Renshaw cell discharge during stretch vibration reflexes. Progress in Brain Research, 50, 45-60. Woodworth, R. S. (1899). The accuracy of voluntary movement. Psychological Review, 3, 1-114. Wurtz, R. H., & Hikosaka, 0. (1986). Role of the basal ganglia in the initiation of saccadic eye movements. In H.-J. Freund, U. Buttner, B. Cohen & J. Noth (Eds.), Progress in brain research, Vol. 64 (pp. 175-190). Amsterdam: Elsevier. Zelaznik, H. N., Schmidt, R. A., & Gielen, C. C. A. M. (1986). Kinematic properties of rapid aimed hand movements. Journal of Motor Behavior, 18, 353-372.

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hershberger (Editor) 0 Elsevier Science Publishers B. V; (North-Holland), I989

299

CHAPTER 12 BEHAVIOR IN THE FIRST DEGREE Richard S . Marken Control theory has been around for some time now, the basic equations having been worked out in the 1920s by H. S. Black (Waldhauer, 1982). Yet the application of control theory in the behavioral sciences is still considered novel and, to some extent, revolutionary (Powers, 1978). What is revolutionary about control theory is not so much its content as its subject matter. The subject matter of conventional psychological theories is behavior; the subject matter of control theory is control (Marken, 1988). The difference between behavior and control is obscured by the fact that both phenomena are typically referred to by the same name-"behavior." But these phenomena are not the same and the difference is significant. Behavior, as the term is typically used in scientific psychology, refers to any observable result of an organism's muscle actions-lever presses, rating responses and verbal reports are familiar examples. Control, on the other hand, refers only to intended results of an organism's actions. Control is behavior in the first degree. In order to properly apply the theory of control we must be able to tell the difference between control and other kinds of behavior. This amounts to distinguishing intentional from unintentional (accidental) behavior. Many psychologists have tacitly assumed that the difference is obvious. Freud, for example, wrote about the significance of unintentional behavior, taking for granted that he knew an accident when he saw one (Freud, 1951). The same is true in modern studies of human error (Norman, 1981; Reason & Mycielska, 1982) where it is assumed that certain behaviors are slips or mistakes and others are not--you can just tell by looking (or asking). In fact, there is nothing intrinsically intentional or accidental about the appearance of any behavior. Nor is there any reason to believe that people can, or will, always tell their true intentions. We do tend to see certain behaviors (like hitting a nail with a hammer) as intentional and

300

Richard S. Marken

others (like hitting a finger with the hammer) as accidental, perhaps because we can imagine plausible reasons for intending to produce one result and not the other. But results are just results. A person could intend to hit a finger and hit the nail by mistake. While our guesses about the intentionality of behavior are probably often correct, a science of behavior should be built on something more solid than guesses.

Intended and Unintended Behavior The problem of distinguishing intended from unintended behaviors can be illustrated in a situation where there are no preconceptions about the intentionality of the behavior involved. Such a situation was created in an unusual computer tracking task (Marken, 1982; 1983) which will be referred to as the ''five squares demonstration." The task is illustrated in Figure 1. A person sits at a computer console and moves objects on the computer screen with a "mouse" controller. The objects are five squares of different sizes located at different points on the screen. The squares can be moved in any two-dimensional pattern by making appropriate movements of the mouse. The mouse moves all five squares simultaneously. A subject sitting at the computer is asked to move one of the five squares around the screen in a circle, "figure eight,'' or just an arbitrary

Figure 1. Movement of squares caused by movement of the mouse in the five squares demonstration.

Behavior In the First Degree

301

Figure 2. Movement paths of the squares during a 20 s period of the five squares demonstration.

pattern. This request is easily carried out, resulting in patterned movement of all the squares as shown in Figure 2. The figure traces out the paths of all five squares during a 20 s period of movement. The paths are identified with the appropriate sized square superimposed on each trace. The subject was intentionally making a circular pattern with the largest square. Nevertheless, a11 squares move in the same pattern so that it is impossible to tell which square is being moved intentionally. The five squares demonstration shows the difficulty of distinguishing intentional from accidental behavior. The movement of every square is a behavior of the subject--it is something that the subject is "doing." However, from the subject's perspective, movement of only one of the squares is his doing--movements of the other squares are just accidental side-effects. But this is invisible to anyone watching the demonstration. All that one sees is the objective behavior of the squares moving around the screen. In this version of the five squares demonstration there is, indeed, no way to determine which behaviors (square movements) are intended and which are accidental. Monitoring eye movements might be considered a possibility, under the reasonable assumption that the subject visually tracks the intentionally moved square. But the subject can easily carry out this task, even switching intention from one square to another, with eyes fixated on a spot in the center of the screen. The subject's intention remains invisible.

302

Richard S. Marken

Figure 3. Movement paths of squares during a 20 s period of the five squares demonstration when a different disturbance contributes to the movement of each square. Only the large square makes a consistent pattern.

Disturbances In the "real world," as in the five squares demonstration, intentions would remain invisible if behavior were always produced by the actor alone. But most "real world" behavior is the result of effects produced by the actor and the environment combined. For example, the behavior called "driving," which consists of a car going in a particular direction, is the result of effects on the car that are produced by both the driver (steering wheel and accelerator movements) and the road (friction, load shifts). The five squares demonstration can be made more realistic by having the behavior of the squares be the result of both mouse movement (an effect produced by the actor) and artificially created environmental effects called disturbances. Computer generated random disturbances are added to mouse effects to produce the movements of the squares. A different disturbance is added to each square. These disturbances act like breezes that gently push the squares horizontally and vertically. If the subject does nothing the disturbances move the squares slowly around the screen in smooth, random patterns. The subject can still move any one of the squares in a desired pattern, but to do this mouse movements must be appropriately combined with the random effects of the disturbance. The paths of the five squares during a 20 s period with disturbances present is shown in Figure 3. The subject is again moving the large square but the circular pattern now stands out amongst the irregular paths of the other squares. The traces of the paths are again identified by the appropriate squares but now it is difficult to match squares (other than the largest one) to movement traces.

Behavior In the First Degree

303

Pattern, Consistency and Control With disturbances present, all squares roam about the screen in different patterns. Only one square moves in a nice circle. This happens to be the intentionally moved square. But how could we know this just by looking at the behavior of the squares? What is it about the behavior of the largest square that indicates intentionally? One possibility is that this square makes a circle rather than some other arbitrary pattern. A "meaningful" pattern (one that has meaning for the observer) might be seen as a less likely or more structured result of action than some other, "meaningless" pattern, suggesting that these results were created "on purpose." But any pattern of movement is possible in the five squares demonstration; the circle could be an accidental (though unlikely) side effect. Another possibility is that the consistency of the pattern is a sign of intent. The large square traces out a circle over and over. The other squares rarely repeat the same pattern. Consistency is a necessary condition for determining that a behavior is intentional, but it is not sufficient. When there were no disturbances, all squares moved in a consistent pattern, both the intentionally moved square and the accidentally moved squares. There can be consistency in both intentional and accidental behavior. We can tell that the large square is being moved intentionally because the consistency of the behavior of this square is achieved in the face of factors that should produce inconsistency. Consistency is a sign of intentionality when it is not expected--indeed, when it shouldn't occur. The random disturbances to the squares should produce inconsistency-and they do, in the case of four of the squares. The consistency of the behavior of the fifth (largest) square is not expected. The production of consistent results in the face of factors that should produce inconsistency is called control. The behavior of the large square is under control. The subject is acting to prevent this square from moving in a random pattern (like the other squares). The subject intends to see a particular pattern and adjusts his actions to satisfy this intention.

Intention and Control We are led to the conclusion that intentional behavior and control are the same phenomenon. It is difficult for some students of behavior to consider such an equation anything more than a metaphor. The term

304

Richard S. Marken

"intention" carries philosophical baggage that is not associated with the term "control." Some feel that equating intention and control somehow trivializes intentionality. This feeling might come from the intuition that intention represents much of what we have in mind when we distinguish "life" from "non-life"; living things have intentions, non-living things do not. As long as intention remains a mystery, so does the essence of life. In fact, control itself is quite a remarkable phenomenon, none the less so because it is no longer a mystery. Once we recognize intended behaviors as controlled results of action, it becomes possible to be more objective and precise in the identification of intentional behavior. Objectivity comes from the fact that we can identify controlled results of action without having to ask the subject "what did you intend to do?" Precision comes from understanding how control works. The theory of control suggests objective methods for determining exactly what aspect of an intended result is actually under control. Computer-based experiments, like the five squares demonstration, can be used to illustrate both the objectivity and precision that comes from recognizing intended behavior as controlled results of action. We will use the five squares demonstration to show an objective and quantitative means of tracking intention--in real time. We will then use another simple experiment to illustrate how to make a precise identification of exactly what result is intended.

Objective Identification of Intention Measuring disturbances is basic to the objective identification of intention. In everyday life the effects of disturbances on behavior are often invisible to bystanders--just as they are in the five squares demonstration. However, disturbances are readily apparent to agents who create them; therefore, the random disturbances in the five squares demonstration comprise information available to the computer. This information allows the computer to identify the square being moved intentionally. In the five squares demonstration there are (at least) five possible intended behaviors, corresponding to the positions of the five squares on the computer screen. The position, pi, of each of the i squares is determined by two factors; the position of the mouse, m, and the value

Behavior In the First Degree

305

of the disturbance, di, to each square. Thus, pi = m

+ di

The two-dimensional position of the ifh square (at any instant in time) depends on the position of the mouse (which is the same for all squares) and the value of a disturbance (a different disturbance for each of the i squares). All the variables in equation (1) are changing relatively slowly over time. The positions of the squares vary because two physically independent variables, mouse position, m, and disturbances, di, vary. If the mouse and disturbance variables have independent effects on the squares, we would expect the variability of pi over time to equal the sum of the variances of these two variables:

var(pi) = var(m)

+ var(di)

where var( ) is the variance of the quantity in parentheses. This is what is expected if pi (the position of any of the squares) is not controlled. Indeed, if the subject moves the mouse around randomly while looking away from the screen, the variance of the position of any square is almost exactly equal to the sum of the mouse variance and the variance of the disturbance to that square (as predicted by equation (2)). When a variable, like pi, is under control, then system outputs (m in this case) will act to prevent disturbances, di, from having their expected effect, in this case moving pi from the intended position. There will be a net correlation between mouse variations and variations of the disturbance to the position of the square being controlled. This will make the variance of the position of the controlled square, var(pi), far less than expected based on equation (2). A quantity called the stability factor (Powers, 1978) can be computed to determine whether the expected variance of a potential controlled variable is significantly smaller than expected. The stability factor is defined as follows: Si

= [var(m)+var(di)]/var(pi)

(3)

where Si is the stability factor of the ith square. The term in the brackets, [var(m) +var(di)], represents the expected variance of square i. If the square is not controlled the expected variance will be equal to

306

Richard S. Markn

the actual variance of the square, var(pi), and the stability factor for the square will be 1.0. For the five squares demonstration, the computer is programmed to sample the value of m, pi, and di about once every 100 ms, and to calculate the respective variances using a running sample of size n = 50 (i.e., as each new datum is added to the running sample, the oldest datum in the sample, entering the calculation 51 sampling cycles earlier, is removed). Since the actual measures of variance are based on a time sample of the total behavior, the value of the stability factor, as measured at any time, will differ somewhat from 1.0 during different sampling periods even if a square is, indeed, not being controlled. The distribution of S is approximately the same as F in the analysis of variance. Like F, S is most likely to be 1.0 under the null hypothesis--that there is no control. If a square is being controlled, its variance will be considerably less than expected; var(pi) will be much smaller than [var(m) var(di)] and the stability factor will be much greater than zero. Based on 50 time samples of the variables involved in the calculation of the stability factor, the value of S is typically about 15.0 or more for the controlled (intentionally moved) square and rarely greater than 2.0 for any uncontrolled square. In the five squares demonstration the value of S associated with each square (based on 50 samples of each variable) is calculated on a continuous basis. Thus, it is possible to continuously monitor the value of S associated with each square. I have set up the demonstration so that a square gets filled in black when its S value exceeds some criterion value (15.0, for example) and gets filled in white (the "default color") when its S value falls below the criterion. At the start of the demonstration all squares are unfilled (white). The subject is asked to select one square and make an arbitrary but consistent pattern with it on the screen. Once the square is filled the subject is to switch to another square and make another arbitrary, but consistent, pattern with it. This switch is mental--there is no physical change in the subject's action (mouse movements). It takes the computer about 20 s to detect the intentionally moved square (when the appropriate S value exceeds the criterion value) after a mental switch in intention. The demonstration works best if the subject is very skillful at carrying out the intention. This means that the subject must be able to counteract most of the effect of the disturbance to the movement pattern of the intended square--the subject must be in control. Subjects who can

+

Behavior In the First Degree

307

perform this demonstration skillfully (after a brief period of training) get the feeling that the computer is "reading their mind." The computer indicates which square the subject is moving intentionally (by filling it in)-without the subject revealing his intention verbally. The computer is almost always right. It can make mistakes because the disturbances to the squares, over the short interval of the computation of S, are not completely uncorrelated. This means that there is some chance that mouse movements which oppose the effects of a disturbance to the intentionally moved square will, coincidentally, oppose the disturbance to another square with a correlated disturbance. This is rare but it can lead to a brief period during which an unintended square gets a higher S than the intended square and is filled in. This demonstration is particularly dramatic because the computer is rather reliably identifying an intended pattern of movement that is not obviously visible to someone watching the squares move around the screen. The intended pattern of movement can be completely arbitrary-a ''squiggly line,'' for instance--so that it "blends" with the arbitrary movements of the other squares. Nevertheless, the stability factor for the intentionally moved square will eventually exceed the criterion, and the square will be filled. The stability factor provides an objective, albeit statistical, basis for identifying intentional behavior.

Precise Identification of Intention The stability factor can be used to identify intentional behavior when we have only a general idea about what a person might be controlling. In the five squares demonstration we know that the subject is intentionally moving one of the squares but we don't know what the pattern of movement might be. The subject is free to make any two dimensional pattern using fast or slow movements or some combination thereof. The stability factor can detect the intended behavior to the extent that a disturbance to an intended result (whatever that result might be) will tend to require systematic opposing actions by the subject. Disturbances to unintended results will tend not to be opposed systematically. This works in general, but it is not very precise--the subjects movements are only statistically related to disturbances and there is some chance of spurious correlations between the subject's actions and disturbances to unintended results. A more precise method of identifying intended results is called the "test for the controlled variable" (Powers, 1979). This test is based on the

308

Richard S. Marken

fact that, in a well designed control system (such as a skillfully behaving organism), disturbance of a controlled result will produce a precisely equal and opposite response from the system such that the disturbance has virtually no detectable effect on the controlled result.

Controlled Variables A controlled (or intended) result is always some variable property of the environment. Controlled results are called controlled variables. Anything in the environment that changes from time to time (varies) might be a controlled variable. Muscle tensions (muscles are in the environment with respect to the nervous system), joint angles, limb movements, events resulting from these movements, configurations of events, rates of change of events, and relationships between events are all potentially controlled variables. In order to be controlled, a variable must be influenced by actions of the organism. The position of the sun in the sky is variable but it is not controllable. The effect of the sun on the retinas, however, is controllable since it can be influenced by the organism, for example, by bringing a hand in front of the eyes. Also, in order to be controlled a variable must be perceptible; we can unknowingly influence how much infrared radiation gets into our eyes but we cannot control the amount of infrared because we cannot perceive it. The test for the controlled variable starts with an hypothesis regarding what variable a system (organism) is controlling. The source of hypotheses about controlled variables is as elusive as is the source of hypotheses in any field of science. However, naturalistic observation of "real" behavior and predictions derived from relevant theories (Powers, 1973) can serve as a guide. The next step is to apply disturbances to the hypothesized controlled variable. If the hypothesized variable is, indeed, controlled, the subject will act to compensate for the effect of the disturbance; the more accurate the hypothesis, the more precise the subject's response to the disturbance.

Reference Levels The intended level of a controlled variable is called its reference level. Reference levels are the "set points'' of a control system--they are the levels of controlled variables that the system "wants" to experience. Reference levels can be fixed or variable. They vary in order to control other variables that are affected by the value of the controlled variable. For example, the reference level for the tension in muscles can be varied

Behavior In the First Degree

309

in order to control the angles at joints. Similarly, the reference level for joint angle can be varied in order to control the position of objects in space. At some point in this chain we will arrive at a controlled variable with a fixed (for the time being) reference--such as the direction in which we are walking. Thus, muscle tensions vary in order to vary joint angles that change the position of the limbs which keep me 'walking in a south-easterly direction. The test for the controlled variable is most precise when we can be fairly sure that the reference level for the controlled variable is fixed during the time of testing. This precision is particularly important when it is clear that some variable is being controlled but we don't know exactly what it might be. The problem is illustrated in an experiment in which the test for the controlled variable is used to determine what variable is being controlled when a subject controls the size of a quadrilateral figure.

What is Size? The subject in this experiment sees two figures on the computer screen, as shown in Figure 4. The square in the upper left corner of the screen stays the same throughout the experiment. The quadrilateral figure in the center of the screen changes continuously (the appearance of the quadrilateral at three different points in time is shown in Figure 4). The height (y) of the quadrilateral changes slowly from a minimum of about .5 cm to a maximum of 10 cm. These height variations are created by a smoothly varying random waveform. The width (x) of the quadrilateral also varies, but this variation is caused by movements of the mouse controller. Moving the mouse to the left decreases the width (to a minimum of .5 cm) and moving it to the right increases the width (to a maximum of 10 cm). The subject is asked to keep the size of the quadrilateral equal to that of the fixed square in the corner of the screen. This can be done by moving the mouse, thus changing the width of the quadrilateral, in order to compensate for changes in height. The subject is asked to control a variable (called "size") relative to a fixed reference level, defined by the size of the fixed reference square. One interesting thing about this experiment is that the "size" of the reference square must define the reference level-not the shape. The quadrilateral is continuously changing shape throughout the experiment-it is rarely a square or even an approximation thereof (as shown in Figure 4). Thus, the subject must control a rather abstract variable-- one

310

Richard S. Marken

that is independent of the shape of the reference square and the quadrilateral figure.

The Test for the Controlled Variable In this experiment we are concerned primarily with determining what the controlled variable might be. What is the subject controlling when controlling the "sizef'of a quadrilateral figure? There are at least two possibilities; area, x * y, and perimeter, 2(x+y). There are other possibilities, like the diagonal length of the quadrangle, but two possibilities is enough for now. What we have done is take the first step in the test for the controlled variable, which is to form an hypothesis regarding what the controlled variable might be. The next step in the test is to create a disturbance to the hypothesized controlled variable. The disturbance consists of changes in the height (y) of the quadrilateral. If the variable that corresponds to "size" is controlled at a fixed reference level, the subject will produce responses (changes is the width, x, of the quadrilateral) that have an effect which is equal and opposite to the effect of the disturbance. For example, suppose that "size" corresponds to x plus y (perimeter). This would mean that the height plus width of the quadrilateral, p, would be kept at a fixed reference value, pr, such that

(x2+9fh,

pl=x+y

(4)

where y is the disturbance and x is the subject's response. In order to keep p at the reference level, pr, the subject varies responses to offset the effect of disturbances to the perimeter. If p is kept perfectly constant (which is to be expected when a variable is controlled at a fixed reference level) then variations in the subject's response over time will be precisely and linearly related to the disturbance, with a slope equal to -1, and a y intercept of p':

And since the reference square is 2 cm high and 2 cm wide, p' should be 4 cm. Figure 5 shows a plot of 60 bivariate samples of x and y (in cm) during one 2 minute run of the experiment. The straight diagonal line running through the data points is the line defined by equation 5

311

Behavior In the First Degree

. Figure 4. Reference square (upper left corner of the screen) and the quadrilateral figure at three different times during the size control experiment.

n Y

x

‘

Time 3

with p’ = 4. The rms deviation of the data points from this line is less than .2 cm--not a bad fit. A slightly better fit can be achieved (rms error of 0.157 cm) by finding the best-fitting straight line using regression techniques. The intercept of this line is not exactly 4 suggesting that the reference level (if perimeter is the controlled variable) is not exactly the same as the perimeter of the reference square. The curved line running through the data points in Figure 5 is the relationship between height and width if the subject is controlling the area, a, of the quadrilateral. This relationship is predicted if area is controlled at a fixed reference level, a’, equal to the area of the reference square. If the subject is controlling area, then a’=x*y

(6)

and the expected relation between response (x) and disturbance (y) is x

=

a’ly

(7)

where a’ is the area of the reference square, 4 cm2. The rms deviation of data points from the line defined by equation (7) is .076 cm, a considerable improvement over the result when we assumed that perimeter was the controlled variable. We conclude that the controlled variable in this experiment is most appropriately viewed as the area of the quadrilateral figure.

312

Richard S. Marken

Figure 5. Plot of width (x) versus height (y) of quadrilateral figure at equally spaced time periods during the size-control experiment. The experimental data is compared to the perimeter-control model (x = 4 y) and the areacontrol model (x = 4/y) of the controlled variable.

-

x = 4-y 0

2

4

6

Y (cm)

More Complex Controlled Variables Finding that a subject is controlling area rather than perimeter when controlling the "size" of a quadrilateral may not seem to be of earth shattering importance. But the size-control experiment illustrates an important principle--which is really the theme of this chapter--namely, that it is difficult to tell what a person is "doing" unless we know what he is intending to do. This is true in both the five squares demonstration and the size-control experiment where we cannot tell what the subject is doing by just looking at behavior. Watching for systematic opposition to the effects of disturbance is the key to determining what a person is doing. Disturbance resistance is a sign that some variable is under control. Systematic opposition to disturbance implies that there is an intention regarding what the results of action should be. The actor is not simply producing results (behavior); rather, he is doing whatever is necessary to make the net result of his actions, combined with the effects of prevailing disturbances, a particular, consistent behavior pattern. By looking at the results of an organism's

Behavior In the First Degree

313

actions as potentially controlled variables we can develop and test hypotheses about which of these are actually under control (intended). The studies described in this chapter deal with very simple results of action--the position of a square, the size of a quadrilateral figure. We deal with simple variables for the same reason that the pioneers of physics studied balls rolling down inclined planes. We study what is simple in the hopes that what we learn will be applicable in more complex situations. Indeed, controlling the position of a square on the computer screen is, in principle, no different than controlling the honesty of one's business dealings or the neatness of one's home. Honesty and neatness are variables--just like the position of a square. The level of these variables is a simultaneous result of effects produced by an actor and the environment. If these variables remain at some constant level it is because the actor is adjusting his actions to do this. The politician can't maintain a relatively high level of honesty by "going with the flow." The house won't keep itself neat. Honesty and neatness are words that refer to variables that seem to be controlled by human organisms. Nevertheless, it would be difficult to start a science of intentional behavior by studying such complex variables. While informal observation suggests that these variables are controlled, a precise definition of honesty or neatness is needed in order to test this hypothesis. Hopefully, we can work our way to an understanding of complex controlled variables by studying quantifiable variables that are likely to be their constituents.

Conclusions A science of intentional behavior--behavior in the first degree--is just beginning. Two of the tools of that science were described in this chapter; the stability factor, to detect the fact that control is occurring, and the test for the controlled variable, to precisely identify controlled variables. The main, near-term goal of a science of intentional behavior is to carefully identify controlled variables and look for relationships between them. One plausible organizational model of the relationship between controlled variables is the hierarchical structure proposed by Powers (1973). This model has the virtue of being consistent with what we know about the nervous system (the ultimate cause of intentional behavior). It is also consistent with the way we tend to describe behavior, with more

314

Richard S. Marken

complex behaviors (like driving) being the result of "lower level'' behaviors (like turning wheels and pushing pedals). Control theory provides the basic framework for understanding intentional behavior. The essentials of the theory (discussed elsewhere in this book) are simple, but this does not mean that control theory deals only with simple behavior. Control theory is "limited" only insofar as it is a theory of intentional behavior--behavior in the first degree.

References Freud, S. (1951). The psychopathology of everyday life. New York Mentor

Books. Marken, R. (1982). Intentional and accidental behavior: A control theory analysis. Psychological Reports, 50, 647-650. Marken, R. (1983). "Mind reading": A look at changing intentions. Psychological Reports, 53, 267-270. Marken, R. (1988). The nature of behavior: Control as fact and theory. Behavioral Science, 33, 196-206. Norman, D. A. (1981). Categorization of action slips. Psychological Review, 88, 1-15. Powers, W. T. (1973). Behavior: The control of perception. Chicago: Aldine. Powers, W. T. (1978). Quantitative analysis of purposive systems: Some spadework at the foundations of scientific psychology. Pychological Review, 85, 417-435. Powers, W. T.(1979). The nature of robots: Part 4. Looking for controlled variables. Byte, 4, 96-112. Reason, J., & Mycielska, K. (1982). Absent minded? Thepsychology of mental lapses and everyday error. Englewood Cliffs, NJ: Prentice-Hall. Waldhauer, F.D. (1982). Feedback. New York: Wiley.

VOLITIONAL ACTION9 W.A. Hershberger (Editor) Elsevier Science Publishers B. V. (North-Holland), 1989

0

315

CHAPTER 13

QUANTITATIVE MEASUREMENT OF VOLITION: A PILOT STUDY William T. Powers In cybernetic control theory, overt intentional behavior is operationally defined as a controlled input or perceptual variable being maintained in a publicly-observable reference condition. In a control-system model the observable reference condition corresponds to a reference signal inside the behaving organism. The reference signal is the physical embodiment of the intention that is directing the volitional action, The volitional actions of others are not always obvious. Their discovery requires finding a variable that the person's actions are maintaining in some identifiable state despite disturbances that act directly on the variable. From the behavior of the controlled variable it is possible to infer the behavior of the internal reference signal and thus get a picture of the directing intention (Marken 1982, 1983). This inference is modeldependent, but as we will see in this study, it can be made with more internal consistency than might seem reasonable.

Experimental Procedure This analysis will be done in the context of a "compensatory tracking" task modified to include an interval of spontaneous behavior. The basic compensatory tracking task requires the participant to use a control handle to keep a vertically-movable cursor stationary on a display screen, centered between two fixed target marks. The cursor is continuously disturbed from inside the computer that runs the experiment, the disturbance varying randomly but smoothly in amplitude. About one third of the way through each run, a tone sounds to indicate the start of a period of spontaneous voluntary behavior, and two thirds of the way through, sounds again to end it. Runs last for 60 s, with a 2 s run-in period to allow control to be established before data recording begins. The screen is updated and a sample of handle position is taken 30 times per second, for a total of 1800 data points. The disturbance is generated

316

William T. Powers

and handle positions are measured with a precision of one part in 2000 relative to the maximum deviation from center, but cursor position is scaled down to fit on a screen with 200 lines of resolution. The participant is instructed to keep the cursor aligned with the target for the first part of each run. When the tone first sounds, the participant is to start making the cursor move in a smooth and regular pattern of up-down movements. These spontaneous voluntary movements of the cursor are to continue until the tone sounds again (about 2/3 of the way through the run), at which time the cursor is again to be maintained level with the target marks. Choice of the pattern of spontaneous cursor movements is left to the participant. Note that the spontaneous voluntary behavior is defined in terms of cursor motion (a perceptual variable) and not in terms of a regular handle movement (an action).

cursor

target

disturbance handle

F

Figure 1. Experimental setup. Handle movements are added to a disturbance generated inside the computer to position the cursor. Two stationary target bars are placed in the center of the screen. The cursor can move up and down between them.

Measurement of Volition

317

Figure 2. Results of experimental run. In the upper part of the figure, the solid line represents handle position, the broken line the magnitude of disturbance. Note the mirror symmetry in the first and third parts. In the lower part of the figure, the cursor position is shown. Deviations due to the disturbance appear throughout. In the center portion, the cursor rises slowly, then falls: the spontaneous part of the run, where the person is trying to make the cursor move in a slow regular way. The spontaneous part is delimited by vertical lines showing where tones occurred to signal start and end of spontaneous action. The drawing of Figure 1 shows the experimental setup with the effect of the disturbance also indicated schematically. Figure 2, about which more will be said later, shows a plot of the results. Every third data point is shown. The handle behavior is the solid line in the upper part of the figure; the disturbance amplitude is shown by the intermittent line and the center of the screen is represented by the straight line. In the lower part of the figure, the cursor behavior is shown, again with a straight line indicating the center of the screen. The disturbance was made just difficult enough to result in appreciable movements of the cursor. The two vertical lines indicate the times when the tone sounded. Because of the disturbance, the changes in cursor position in the middle part do not resemble the handle movements that created them. For this

William T. Powers

318

run, the participant (the author) was trying to make the cursor move in a stairstep pattern, first upward, then downward.

The Model of the Actor The participant's organization (relative to this task) is represented as a system containing an input function, a comparison function or comparator, and an output function. The input function converts the cursor position c into a perceptual signal p representing it inside the behaving system. The perceptual signal's magnitude varies as the cursor position varies, but with a slight exponential time-lag. Thus a step-change in the cursor position would result in a change in perceptual signal of the same numerical magnitude, but the perceptual signal would approach its final magnitude exponentially. The form of the input function is given by a computer program step (in the Pascal language) that computes the perceptual signal p from the cursor position, c: p := p

+ (c

-p)/S

The programming symbol 'I: =" (colon-equal) means assignment or replacement, not equality. To introduce a lag we subtract the old value of perceptual signal from the computed new value, which is just c. This p, is divided by a slowing factor S, and that fraction of difference, c the computed change is added to the old value of perceptual signal. The result replaces the old value of p. Thus p is allowed to change only a fixed fraction 1/S of the way from the old value to the new value on each repetition of this step. With no lag (S = l), the perceptual signal would be numerically equal to the cursor position at all times. This computation approximates an exponential lag with a time constant of tc = (1/30 s)/log,(S/S-l). The slowing factor S is one of the two adjustable constants in the model. The best value of S for the illustrated data proves to be about 5.50 (to the nearest 0.25), implying a perceptual time constant of 0.14 s. The method for evaluating parameters will be explained shortly. If p is initially zero, and c is constant at 100, then with a slowing factor S = 5.50, the successive values of p (obtained by executing the

-

Measurement of Volition

319

above program step over and over) are 0, 18.2, 33.1, 45.3, 55.2, 63.3, 70.0, ... 100.0. If the value of c changes during these computations, the value toward which the series is converging will be changing, so p will lag behind c by an amount that depends on how fast c is changing. This kind of lag is not a pure time-delay (or "transport lag") but simply a slowing or smoothing of the response of the input function. The comparison function subtracts the value of the perceptual signal from the value of a reference signal r. It is the varying value of r during the spontaneous voluntary phase that we are attempting to estimate by the procedures outlined below. The outcome of the subtraction is an error signal e, the magnitude and sign of which continuously indicate the mismatch between the perceptual and reference signals. The comparator is represented by the program step e

:= r

- p.

This sense of the subtraction was chosen to let all other constants be positive. The output function receives the error signal and converts it into a value of handle position, h; that is, h = f(e). The particular function chosen makes handle velocity depend on the magnitude of error. If handle velocity is a constant K times the error signal's magnitude, then the handle position is calculated by a program step that does a crude numerical integration (over the 1/30 s interval):

h := h

+ K * e.

This step is not an equation; it means that K times the error signal magnitude is added to the current value of handle position to obtain the next value (on the left of the ":=" sign). The constant K is the second adjustable parameter in the model. The asterisk is the program-language version of a multiplication sign, which is always explicit in a written program. The single constant K absorbs all other possible constants of proportionality in the model of the participant. These three steps result in a model whose dynamic properties approximate those represented by the Yransfer functions" obtained in similar experiments done by engineering psychologists (See Osafa-Charles, Agarwal, O'Neill, & Gottlieb, 1980, for the conventional forms).

320

William T. Powers

The Model of the Environment The handle position is sensed by an analogue-to-digital converter in the computer and is represented by a number that can range from -2000 to 2000. A second number is taken from a precalculated table of disturbances. The table is constructed by successive smoothings of a series of random numbers generated by an algorithm, and is scaled to a peak-to-peak amplitude of 2400 units. Adjusting the amount of smoothing changes the rapidity of variations in the disturbance amplitude, and so adjusts the difficulty of the task. The cursor position is determined every 1/30 s by sampling the handle position and adding to the result the next sequential entry from the table of disturbances. Thus the cursor position represents neither handle position nor disturbance alone, but only their sum. Using d for disturbance magnitude, we have the model of the environmental relationships in this experiment in the form of the final program step: c := h

-+ d.

This "model" of the environment correctly represents the actual environment, because the same program step is used to position the cursor when a human participant is moving the handle. When the model is running, the computer model of the participant is given the value of c directly where the real participant sees the cursor on the screen; the model gives back the value of h to the environment as a number where the participant moves a physical handle to generate that number. The same table of disturbances is used for both participant and model. This table can be changed easily to prevent memorization (by the person) of any patterns. All data given below are for a run in which a new disturbance pattern was experienced for the first time by the participant (although practice with other patterns preceded the live run). It should be emphasized here that the object of this exercise is to demonstrate a principle and a method, not to show research results with many human participants. Such applications will be developed, but for now only a single participant is needed--the author. The reader may, however, assume with confidence that these results will be typical of any well-practiced participant. Control-system experiments are highly reproducible after learning is finished.

Measurement of Volition

321

Running the Model We now have a model of both the participant and an environment, consisting of five program steps arranged below in the sequence appropriate for computation (the fifth step repeats the calculations). These steps are executed 1799 times, with an index i.(in brackets) advancing by one on each step. The index is used to access successive values of the disturbance, and also to point to locations in a table where the computed values of handle position are stored after being computed. Only handle position needs to be saved, as cursor position can be reconstructed exactly from c := h d. We set the model’s reference signal r to zero at first, indicating that the model is attempting to keep the cursor at the zero position (corresponding to the position of the target marks on the screen).

+

Initialization:

Step Step Step Step Step

r := 0 i 0 p := 0 h[i] := 0

.-

1. c := h[i] + d[i] 2. p := p + (c - p)/S 3. e := r - p 4. h[i + 13 := h[i] + K * e 5. Increment i; if i < 1799 go to step 1.

When a variable reference signal is used, a table r[i] is substituted for the fixed value r. When the human being is doing a run, steps 2, 3, and 4 are replaced by a step that displays the cursor position on the screen and samples the physical handle position. For a model run, two variables, h and p, must be set to initial values. The initial values are 0, a safe value because the same 2 s runin period is used for the model and for the participant, and allows plenty of time for any starting transient to disappear. The run-in is not shown; it is accomplished simply by starting i at 60 and running it downward to 0 before starting to advance it upward again; the stored values of h are overwritten. In this way there are just 1800 values of h in the final table with no extras to discard.

322

William T. Powers

Readers who are thinking of trying other models like this should be warned of a hidden difficulty. This model works primarily because of the time-integration in the output function. A digital computer model of a continuous closed-loop system, if constructed without any time integrations or other slowing factors, will not work properly because physical time is not properly handled. In the real system being modeled, the various functions all operate at the same time, not in sequence. Only when some suitable way of handling time is introduced can a model computed as a series of sequential steps give the right answers.

Evaluating the Parameters K and S The parameters K and S are evaluated using a very simple, yet satisfactory, heuristic procedure. Initially, the slowing factor S is set to 1 (no lag), the integration factor K is set to 0.1, and the model is given a trial run for comparison against the author's performance previously recorded, using the same series of disturbances. This occurs at high seed, taking about .1 s. The model's cursor behavior is then compared by subtraction with the author's cursor behavior and the sum of squares of the differences is computed. Only the first and last thirds of the data (before and after the tones, with a 1 s delay after the second tone) are used, because the model, at this point, can't generate a different pattern of behavior in the middle part.Then the output integration factor is stepped upward from 0.1 in increments of 0.005 as the procedure is repeated over and over. Each time the sum of squares reaches a new low, the values of K, S, and the summed squared error are saved. When the new squared error exceeds the minimum squared error by 3 per cent, the best parameter values are saved as the best values for that series. Then S is incremented by 0.25 and the entire sequence is repeated with K beginning 0.02 units below the previous best value. This procedure ends when the minimum squared error is 3 per cent greater than the value that went with the "best of the best" values of the parameters. The 3 per cent criterion for ending all runs was found by trial and error. While this method is not elegant, it is simple and takes less than one minute on a 10 Mhz IBM-AT-compatible microcomputer. More elegant statistical approaches do not give as good results because the statistical distribution of errors is not close enough to the usually assumed Gaussian distribution. Figure 3 shows a run of the model with the reference signal constant at 0, the optimum value of S, 5.50, and the optimum value of K,

Measurement of Volition

323

0.220. This model behaves reasonably well in the first and third parts of the run, although the behavior differs from the real run in the central part because of the constant zero reference signal.

Deducing the Reference Signal We now have a model with parameters that make it reproduce the participant’s behavior in the first and third parts of the experimental run. This model has a reference signal of zero, meaning that the model is maintaining the cursor near the center of the screen where the target marks are, or would do so if the cursor were displayed. We have adjusted the parameters to make the model show nearly the same variations in cursor position and handle position that the subject produces in the first and last thirds of the run, where the target position is zero. In the middle part, the spontaneous action of the participant makes the two cursor traces very different.

Figure 3. Results of model run with optimum perceptual lag (S = 5.50), optimum integration factor (K = 0.220), and reference signal set to zero. These values give the best fit (smallest least-squares difference) of model and participant cursor behavior in the first and third parts of the run. The model reproduces many features of the real data in Figure 2, but is generally smoother in its action.

324

William T. Powers

Figure 4. Deducing the reference signal. The upper trace is the person's error signal, the first time derivative of the actual handle position divided by the integration factor K. The middle trace is the person's perceptual signal deduced from the actual cursor position, assuming the same lag as in the model (see text). The bottom trace is the sum of the top two and represents the deduced reference signal. Notice the appearance of high-frequency variations in the reference signal.

To deduce the reference signal in the model, we apply the model's functions to the data taken from the participant. For each data point, we infer the error signal from the observed handle position, and the perceptual signal from the observed cursor position. Because the comparison process is defined as e = r p, we can calculate that r must be p e. The hypothetical perceptual signal can be obtained from the observed cursor position and the calculation representing the model's input function, by the program step p := p (c p)/S. The hypothetical error signal can be obtained from the observed handle position, the integral of the error signal: the error signal inside the participant would be dh/dt divided by the integration factor K. As a program step, we have e[t] := (h[t] h[t-l])/K. Having found e and p, we then calculate the

-

+

+ -

-

Measurement of Volition

325

value of r = p + e. The need to take a derivative is the main reason for recording handle position with such high resolution. Figure 4 shows, from top to bottom, (dh/dt)/K, p, and r as deduced from the participant’s data and the model’s parameters and functions. The reference signal contains a variation higher in frequency than any variations seen in the handle or cursor traces of Figure 2. We could remove the high-frequency variations in r by a smoothing method that discriminates strongly against high frequencies, but we will accept them as real and see what the consequence is.

Completing the Model

As a check to see if the derived reference signal does in fact result in the right model behavior, we can use the pattern just obtained in the bottom trace of Figure 4 as the reference signal for a model run. The same program steps outlined above are used, but instead of initializing r to zero, we now use the result from Figure 4 as a table of referencesignal amplitudes and do the computations with r[i]. The index i picks out successive values of reference signal just as it picks out successive values of disturbance. The result of a model run is shown in Figure 5. Comparison of this result with that of Figure 2 shows that the real run is duplicated. The correlation between handle positions, model and real, is .99841, and between cursor positions is .99991 (n = 1800). This does not indicate that we have made an extraordinarily accurate prediction, but only that the method of deriving the reference signal does generate just the signal needed to account for the observed behavior-in other words, that there has been no computational error and no cumulative rounding effect of consequence. In Figure 6 we have, from top to bottom, the participant’s cursor trace, the final model’s cursor trace, the deduced reference signal used for the model run, and a version of the deduced reference signal smoothed with a four-pole low-pass filter. The bottom trace in Figure 6, the smoothed version of the reference signal, shows a stairstep pattern more clearly than the cursor traces do, either for the model or for the participant. This is, presumably, a record of the intended positions of the cursor. The smoothed version of the reference signal shows a best estimate of the reference signal with the rapid variations removed. The same amount of smoothing applied to the actual cursor position makes almost no noticeable difference in its shape, because the smoothing cuts out only the highest-frequency variations.

326

William T. Powers

Figure 5. Model run using deduced reference signal from Figure 4. Handle-tohandle (model vs. real) correlation is 9 8 4 1 (n = 1800). See Figure 6 for cursor-to-cursor comparison.

Thus the reference signal could not be obtained simply by smoothing the cursor trace.

Reviewing the Rationale Let us review the strategy. We first matched a model with real behavior for portions of the run in which the participant is presumed to have a reference signal constant at zero. In doing this we evaluated two constants, a perceptual lag constant and an output integration factor. This produced a model that could match the participant’s behavior with normal accuracy outside the region of spontaneous voluntary behavior. Then we used those constants under the assumption that the participant is organized as the model is. In the model, the simulated

Measurement of Volition

327

Figure. 6. Comparison of cursor behaviors. Top trace is real cursor behavior; center trace is model cursor behavior; next trace is the deduced reference signal used to run the model; lowest trace is the smoothed version of the deduced reference signal. Cursor-to-cursor correlation is 99991 (n = 1800). This correlation shows that the reference signal was deduced and applied consistently. handle position is the time-integral of the internal error signal; hence the participant’s assumed error signal is proportional to the first derivative of observed handle position. The model’s perceptual signal is the lagged model cursor position; hence we assume that the participant contains a perceptual signal that is a similarly lagged actual cursor position. Still applying the model in a straightforward way, we then add the error signal to the perceptual signal to deduce the participant’s reference signal. Finally, we run the model using that deduced reference signal (unmodified) to see if the resulting behavior matches that of the participant, to check that the derived reference signal does lead to reproducing the actual behavior--that is, to see if the derivation was correctly done.

328

William T. Powers

i

I'

Figure 7. Experimental run when eyes are closed during middle part of run, and handle is moved by feel in a series of steps upward, then downward again. The model allows us to see the behavior of a variable, the reference signal inside the person, that is not directly visible from outside.

An Interesting Variation The model, at least as it stands, cannot distinguish apparent from real intentions: it will always compute a reference signal. Because of the way the derivation is performed, this reference signal will always make the model reproduce the observed variables correctly. As a preliminary way of investigating this problem further, I performed the experiment with slightly different instructions. Instead of the person at the controls moving the cursor in some regular pattern during the middle part of a run, the person now closes his or her eyes at the first tone and opens them at the second tone, continuing to move the handle (blindly) in a pattern something like the

Measurement of Volition

329

Figure 8. Comparison (like that of Figure 6). The reference signal is now spurious in the middle part. Cursor shows no regular pattern, high-frequency variations are missing from center part of unsmoothed reference signal trace. See Conclusions and Discussion. pattern in which the cursor was supposed to move (the same disturbance is used). The result of an experimental run is shown in Figure 7. Figure 8 shows the eyes-closed result that corresponds to Figure 6 the actual cursor behavior at the top, the model's cursor behavior next to the top, the "deduced" reference signal next to the bottom, and the smoothed version of the reference signal at the bottom. The result of carrying out all our manipulations of data is a model that exactly reproduces the behavior throughout the run. But we know that the model can't apply during the middle interval--there is no

330

WZZiam T. Powers

perceptual signal representing cursor position. What we now have is a model that reproduces the cursor behavior throughout the run on the assumption that the cursor behavior was intentional. In other words, if the person had intended the cursor to move as it actually moved, this model would show the reference signal representing that intention. In Figure 8, the high-frequency noise in the deduced reference signal disappears during the middle part of the run, although it is present before and after when the eyes are open, and tracking is actually occurring. We are still using the first derivative of handle position in computing the reference signal, so we know now that the noise does not originate in the output function or in the measurement. There is no similar noise in the perceptual signal at any time, so we have evidence that when tracking is really occurring the noise is probably associated with the reference signal. Examining Figure 7, we see that the only stair-step regularity that appears is in the handle trace; the cursor trace shows no obviously regular pattern. Without physically disturbing the control handle, we cannot prove that the regularity in handle movements is intentional (in the terms of this theory), but clearly if such disturbances were used, we could apply this same analysis just as we have done here and deduce a reference signal for handle position. Thus we could build downward toward lower levels of a hierarchical model. Similarly, by making the tracking skill part of a task involving control of more general variables, we could build upward toward higher levels of a hierarchical model.

Discussion and Conclusions About the present results we can say at least this: when a person deliberately makes the cursor move in some clearly-conceived way, the model will allow us to deduce a reference signal behavior that the person will agree represents the intended movement of the cursor more closely than the actual cursor movement represents it (when we smooth out the highest-frequency variations in reference signal). We still have to rely on the person to tell us that the cursor movements really were intentional, and that the deduced pattern is close to the intended pattern. While that is legitimate information, it would be better to obtain it some other way. The only way to do so is to expand the model to include more kinds of behavior and more levels of behavior--to find other ways of observing what we assume is the same phenomenon.

Measurement of Volition

331

The ideal way to test this model might be through recording neural signals in appropriate parts of the central nervous system of the participant. In the present state of technology, however, doing this by nonintrusive and safe means is beyond us. We are in much the same position as astronomers were before space travel became possible. When a telescope is pointed toward the tiny dot of a planet in the sky, we can see or photograph an image that shows a disk with markings on it. By referring to optical theory, and by analogy with observing objects on Earth that we can inspect by other means, we can infer that there really is something out there corresponding to the image. This inference, however, is unverifiable, because the same image could be generated in many ways other than by a planetary body located millions of miles away and illuminated by the Sun. All we can be reasonably sure of is that a collection of wavefronts of light enters the telescope and is subjected to consistent transformations caused by the optical elements: any phenomenon capable of creating the same wavefronts at the open end of the telescope would produce the same appearance in the eyepiece or on film. A computer-generated hologram, for example, could reproduce the image exactly. We have now sent spacecraft to Mars, for example, and their cameras confirm, generally, the fuzzy outlines we see from Earth. Or do they? Are we not in the same position as before when we look at the images generated by the cameras? All we can really say is that the assumption of a real body, given the laws of optics and extrapolation from phenomena on Earth, is consistent both with the spacecraft pictures and the Earth-based telescopic pictures. It would seem that we will not get final confirmation until a human being orbits Mars or lands on its surface. Even then, the problem will not be solved, the inference will not become a fact beyond all doubting. All we could say is that the wavefronts of light reaching the pupils of the astronaut’s eyes, transformed by the optical properties of the lens and interpreted by the computations in a human retina, create a perceived result consistent with the idea that a real body exists, and also consistent with the spacecraft pictures and the Earth-based pictures. With each step we take toward certainty, certainty itself recedes. In short, we are faced with the same problem that greets all sciences that rely on models of reality for their understanding of nature-physics, chemistry, astronomy, geology, neurology, psychology, and so on. We assume models that seem to serve as instruments for observing

332

William T. Powers

formerly invisible objects. Then we try to find alternate ways of observing the same thing, which always turn out to be alternate models or alternate ways of applying the same model. When inconsistencies arise, we modify the models to remove them, or even invent new models when the old ones can't be made to work any more without changing their fundamental nature. The nearest we get to certainty that the models are true pictures of reality is a subjective conviction that what we see makes sense, looks simple, repeats itself, changes as it ought to when circumstances change. We seem to be seeing reference signals here through a n instrument called control theory. There are, no doubt, alternate explanations for what we see here. There always will be. The best we can do is expand our experiments and look for alternate views of the same phenomenon, either to increase our conviction that we are seeing something real, or to force us to change the model.

References Osafa-Charles, F., Agarwal, G. C., O'Neill, W. D., & Gottlieb, G. L. (1980). Applications of Time-Series Modeling to Human Operator Dynamics. IEEE Transactions on System, Man, and Cybernetics, SMC-10, 849-860.

Marken, R.(1982). Intentional and Accidental Behavior: A control theory analysis. 1982. Psychological Reports 50, 647-650. Marken, R. (1983). "Mind reading:" A look at changing intentions. Psychological Reports 53, 267-270.

PSYCHOLOGICAL PERSPECTIVE

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hershberger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

335

CHAPTER 14 SOME EXPERIMENTAL INVESTIGATIONS OF VOLITION George S. Howard and Paul R. Myers Why do human beings act as they do? This fundamental question, regarding the wellsprings of human behavior, represents a foundational issue for most psychological research. But, one could easily argue, the quest for enlightenment regarding the ultimate basis of human action also represents the fundamental quest of scholarship in the humanities. Humanists have proffered answers to the question of why people act as they do for thousands of years. Therefore, since scientific psychology is little over one hundred years old, one should not be surprised that discussions conducted in the humanities and the other sciences over the past several millennia have greatly influenced psychologists' research efforts in their inaugural century. Psychologists did not need to start at %quare one" in deciding what it meant to be a human being, or in deciding what it meant to be a science. We borrowed heavily from both the scientific and humanistic intellectual traditions. While much of what was borrowed proved helpful, some was not. It seems that the role of volition (or self-determination) in the genesis of human action is one domain which has been ill-served by the extant perspectives.

The Free Will vs. Determinism Debate We believe that a false dichotomy has grounded the free will vs. determinism controversy, and has dominated thinking in this domain for the past 2500 years. From the beginning, thinkers pitted the wrong combatants against one another. Determinism is the thesis that all actions result from some cause or causes. The opposite of free will is not determinism, it is nonagentic mechanism. Free willists can also be determinists! They are determinists who recognize the importance of a special category of cause-self-determination. Thus, instead of seeing the relevant issue as free will (or self-determination) versus determinism,

336

George S. Howard and Paul R Myers

Figure 1 represents a more appropriate conception which posits two independent dimensions: The most important point to be made is that one could be a determinist and a nonagentic mechanist (this represents the typical assumptive stance of scientific psychologists), or one could be a determinist who also is a free willist (one who believes in the causal force of agentic self-determination). It is this latter position that will be recommended in this chapter. Further, the studies presented herein not only speak to the ability of humans to self-determine, but they are also scientific and deterministic because they probe the causal factors involved in the formation of human behavior. Finally, all scientists must be determinists, as acausality is the enemy of scientific analysis (one should note that Indeterminacy and Uncertainty are quite distinct concepts from acausality). Although the position espoused herein diverges from the traditional construals of the free will vs. determinism antinomy, some important distinctions can be garnered from that controversy. For example, the most fruitful construal of the notion of free will turns upon the following question: When a person chooses and then performs an action, might that person have done otherwise, ceterisparibus (that is, if all other things had been equal). This description speaks to the issue most of us believe to be at the heart of free will. Namely, could the action have been self-determined to occur differently even if all characteristics of the situation had been identical except the agent's choice to behave differently. But the ceterisparibus assumption can never be met because time marches on and as the presocratic philosopher Heraclitus said, "One can never step in the exact same river twice." There is no way of knowing for sure, once a person makes a choice and acts upon it, that he or she in truth might have chosen to do otherwise. But we believe that these problems in satisfying the conditions of the ceferis paribus assumption

Readers familiar with the sex-role orientation literature will recognize that a similar conceptual move was made with great success in this domain. Masculinity and femininity were traditionally viewed as opposite ends of the same bipolar dimension. Thus, for example, the more masculine one became, by definition, the less feminine the person would be. Masculinity and femininity are now seen as separate bipolar dimensions. One could be: high on both dimensions (androgynous); high on one dimension but low on the other (either highly masculine or highly feminine); or low on both dimensions (undifferentiated).

Experimental Investigations of Volition

Determinists All actions result from some cause(s).

Free Willists Belief in Agency! People are largely the cause of their own actions.

<====>

337

Nondeterminists Things are uncaused; they just happen randomly, spontaneously, etc. Nonagentic Mechanists Our actions are the result of mechanisms (e.g., environmental, physiological, genetic, cultural) which are completely coercive.

Figure 1. A reconceptualization of the core issues of the free will vs. determinism controversy. occur only because thinkers restrict their consideration to single choices and acts. And, in fact, this is precisely where free will advocates foundered in the free will vs. determinism debate. Thus, while freedom of choice is perceived by the acting agent as an inner certainty, for many years this human power could not be demonstrated with compelling rigor. Conversely, advocates of determinism are continually heartened by the evidence of science suggesting the numerous nonagentic mechanisms (biological, environmental, cultural, etc.) implicated in human action. Thus, of late, the free will vs. determinism debate has gone quite badly for free will advocates.

An Operationalization of the Ceteris Puribus Assumption It is generally conceded that among the most important strategies developed by modern sciences are those that deal with experimental control. Control over extraneous variables that might obfuscate the relationship of interest; control of plausible rival interpretations to the experimental hypothesis. Among the many successful control procedures

338

George S. Howard and Paul R Myers

developed, two forms of control that emerge are: control though elimination, and control through equalization. The development of vacuums represents an example of control via elimination. By reducing the density of the medium through which objects fall to near zero, one can study the effects of gravity in a "purer", less contaminated, manner. But some variables cannot be eliminated from a study. If one is interested in the relationship between a particular chemical fertilizer and crop yield, then the amount of water and light that plants receive are critical factors to be controlled. But one cannot eliminate water and light without killing the plants, therefore, a control through equalization strategy is indicated.

Between-subject and Within-subject Methods of Equalization Random assignment of experimental subjects to research conditions represents the optimal equalization control strategy. For example, if I had access to one hundred fields for a study of the effects of a particular fertilizer on crop yield, I could randomly assign the fields to one of two groups of fifty fields. Then one of these groups of fields would receive the experimental fertilizer while the other group of fields would receive either no fertilizer or the standard fertilizer treatment (usually determined by whichever comparison represents the most interesting theoretical or practical contrast). The rationale behind the random assignment procedure is that two groups will be created which are equal on the average on all conditions (e.g., sunlight, water, insects, quality of soil, etc.) save the independent variable (e.g., fertilizer versus no fertilizer) and any other variable inadvertently correlated with the independent variable (e.g., consider the implications of the following bad experimental procedure-suppose the experimenter weeded the fertilized plots each morning and the not-fertilized plots each afternoon, and assume the experimenter was far more efficient at weeding in the morning). Assuming the correlated variable problems are all handled properly, it follows that differences on the dependent variable (e.g., crop yield) between the two groups of plots can logically be attributed to the independent variable (e.g., fertilizer versus no fertilizer) since the effect of all other possible causes are controlled by the formation of two equivalent groups through the random assignment procedure. Behavioral scientists are virtually univocal in acknowledging the power of equalization via random assignment in isolating discrete causal relationships. But many variables of interest to social scientists (referred to as subject-variables--e.g., intelligence, motivation, personality traits, etc.) cannot be randomly assigned in a

Experimental Investigations of Volition

339

non-trivial manner. Take intelligence as an example. One cannot randomly assign subjects to be highly intelligent or low in intelligence. Thus, intelligence cannot be studied as an independent variable by employing random assignment--although random assignment can be used to control for intelligence when the effects of other variables (e.g., teaching styles) are being investigated. Yet another strategy for controlling extraneous variables experimentally involves employing within-subject designs. In such studies each subject is exposed to more than one condition of the independent variable. The famous "Pepsi challenge" capitalizes upon within subject strategies, since each subject tastes and rates the flavor of both Pepsi and Coke. In such studies, each subject serves as his or her own control and thus all subject variables are equivalent for the two groups when the ratings of Pepsi and Coke are made. Of course, proper design procedures (e.g., counterbalancing the order of presentation of stimuli) must also be employed to eliminate specific threats (e.g., novelty effects; carryover effects, etc.) to valid inference. The new operational definition of volition (or will, or behavioral freedom, or self-determination, or personal causation) described below represents a hybrid approach that employs the logic of random assignment in a within-subject design.

A Method for Studying Self-Determined Behavior The crucial aspect of the new methodology for investigating volitional behavior is that it requires the active cooperation of the research subject. If, for whatever reason, the subject chooses not to cooperate fully with the experimenter, a serious underestimate of that subject's degree of volitional control will be obtained. One of the first procedures developed involves the experimenter dividing the total time for the experiment into a large number of equal-length time blocks. The experimenter then randomly assigns each of the time blocks to either %y to or 'I try not to conditions. For example, if one considered subjects' ability to control between-meals snacks, as a part of their ability to lose weight, the instructions for each subject on half the studies' days would be to ''eat as many snacks as you wish" whereas on the other half of the days, subjects would be instructed to "try not to eat any snacks." Differences in mean number of snacks consumed on "eat" versus ''not eat" days is a reflection of the subject's ability to volitionally control snacking behavior. The studies to be reviewed below probed subjects' capacity to volitionally control their actions. Is this teleological account the only --'I

--I1

340

George S. Howard and Paul R Myers

interpretation that can be given for the data reported herein? Definitely not. One might choose to argue for a Humean-type, efficient cause explanation. For example, if subjects eat more peanuts on "try to eat" days than they do on "try not to eat" days a critic might claim that such evidence does not imply volitional control of that action. Rather, the critic might assert that subjects have been socialized to play "good subject" roles in scientific studies. Thus, the difference in amount of peanuts eaten on "eat" versus "not eat" days is best attributed to conformity by the subject to the demands of the experiment, rather than being evidence of volitional control? While at first blush this might appear to be a serious criticism of this group of studies, closer inspection reveals that such interpretive difficulties are endemic to all research (although this fact is rarely acknowledged by psychological researchers). The problem is referred to by philosophers of science as the underdetermination of theory by evidence (Hansen, 1958; Kuhn, 1962; 1977). In its weakest form, the underdetermination thesis suggests that the meaning of a research finding is never transparent. Does this mean that there are no criteria whereby scientists can make objective (if fallible) judgments as to the probable meanings of research findings? Not at all. Cronbach (1982) points toward a partial solution to this dilemma by expanding the traditional notion of the concept of the '"validity"of an experiment: Validity depends not only on the data collection and analysis but also on the way a conclusion is stated and communicated. Validity is subjective rather than objective: The plausibility of the conclusion is what counts. And plausibility, to twist a cliche, lies in the ear of the beholder (p. 108). (Italics added.) The position espoused herein is that humans possess capacities that enable them to behave volitionally. If that position is correct, then

This conformity rival hypothesis is an example of a "variable inadvertently correlated with the independent variable" point made above. The only difference being that the weeding example mentioned depicted an instance of bad experimental procedure which might easily have been remedied through better control procedures. However, the unfortunate coupling of volition with conformity is fundamental to this operationalization procedure and is not amenable to design control. Thus, the plausibility of this rival interpretation had to be specifically tested in several experiments to be described below.

Experimental Investigations of Volition

341

individuals might be expected to be able to control their behavior in meaningful ways. But note that another investigator might see that same behavior as being under the control of stimuli, both internal and external. Such an individual might actually design the exact same studies as will be reviewed here (and, hopefully, would have obtained a similar set of results). But the account of the meaning of the findings would have been quite different--and perhaps equally plausible. No amount of evidence ever ’proves’ a theory, and relatedly, there can be multiple theoretical interpretations of any body of evidence. Hence, one is forced to a position like Cronbach’s where the scientific community evaluates the plausibility of various competing accounts of empirical findings. This chapter argues for the plausibility of a volitional account of a growing body of research evidence. However, other interpretations of these findings are also plausible.

The Evidence for Self-Determination The caveat on the interpretation of research evidence aside, what are the data that suggest the importance of volitional control in human action? Using subjects’ self-control as a warrant for volitional behavior, Howard and his colleagues (Howard, 1988; Howard & Conway, 1986; Howard, Curtin, & Johnson, 1988; Howard, DiGangi, & Johnson, 1988; Howard, Youngs, & Siatczynski, 1988; Lazarick, Fishbein, Loiello, & Howard, 1988; Steibe & Howard, 1986) conducted a series of studies that considered what proportion of a particular eating behavior was due to particular external, efficient-cause influences, and what part was due to volitional control. Several studies dealt with subjects’ ability to control eating peanuts. Peanut eating was chosen because it provides a noncontroversial dependent measure, and because eating peanuts is an activity subjects tend to enjoy, but which they should be able to control. Here (Howard & Conway, 1986, Study 1) the effect size (partial Eta squared) for volitional control was S6, while the effect size for whether the food was kept in sight or out of sight was .13. Comparable figures for the second study were .57 for volition and .16 for whether the subject received a written reminder or not. Finally, in a third study (Howard, Youngs, & Siatczynski, 1988, Study 1) the effect size for volition was .53 while the effect size for a written reminder was .03. Therefore, in studies on the control of eating behavior, volition appears to be about five times more influential than certain (sight and reminder) external, efficient-cause

342

George S. Howard and Paul R Myers

influences (i.e., the average effect size for volition as .56, while the average effect size for the efficient causes was .11). Great care should be taken in interpreting the above findings. The major point is that we now are able to assess the influence of volition in eating in a rigorous, empirical manner. Beyond that, interpretation becomes difficult. For example, the ratio of the effect size of volition to the effect size of external factors should be viewed with extreme caution. Had we considered other and/or more nonvolitional factors in our studies, that ratio would likely have been reduced. Therefore, if one really were interested in volitional versus nonvolitional factors in eating behavior, these studies would represent but a first step toward developing a more complete understanding of the phenomenon that might be obtained by investigating additional external and/or organismic variables. Conversely, one might consider whether various self-control enhancing techniques might actually increase the volitional-to-nonvolitionalratio in our account of the phenomenon. Have we tested the magnitude of the force of free will in these studies? Here we would respond with a resounding "no". Imagine a hypothetical subject, who was told on certain days to try to eat as few peanuts as possible (a "don't eat" day) who thought the following: I know I'm not supposed to eat today, and I could resist if I wanted to (Assume this is to be true for now). But I'm not that invested in the results of this study, and I've studied hard today, I feel I deserve a treat. I'm going to eat some nuts because I feel like it. In such a case, the subject might have exercised his/her free will, but the act of eating the nuts would serve to decrease the volitional component of behavior observed in the study. Insofar as such decisions do actually occur in our studies, the magnitude of effect of volitional control observed in this research represents an inappropriately conservative estimate of the effect of free will. That is, the exercise of free will might actually reduce the magnitude of the volitional component observed.

Criticisms of a "Volitionat'Interpretation A second set of investigations probed the plausibility of the volitional account of the findings of the peanut eating studies, as opposed to an efficient cause explanation of the data. Specifically, it was suggested by some that the differences in the amount of peanuts

Experimental Investigations of Volition

343

consumed on days when subjects were instructed to "eat" versus when they were told to "not eat" did not represent evidence for volitional control. Rather, critics saw these differences as confirming that peanut-eating behavior was under the control of the experimenter, since the subjects obeyed the experimenter's instructions. Or as one commentator put it, 'The psychologist reader, steeped in behavioristic lingo, 'sees' in the experimental instruction to eat or not eat a nonvolitioml control or manipulation directing behavior." Two objections, which are really two ways of wording the same problem, were raised. These explanations suggest that the subject is either trying to be a good subject (Orne, 1962; Weber & Cook, 1972) or is trying to avoid being "socially sanctioned'' for being an inconsistent, and therefore, a poor subject (Hayes, 1987). First, we wish to address the contention that subjects are responding to demand characteristics, that is, that subjects had determined the experimenter's hypotheses and were behaving in such a manner as to confirm the hypotheses (and thus be good subjects). Note that the force of this position evaporates if one sees subjects volitionally choosing whether to adopt "good subject" or "bad subject" roles (see Howard, Youngs, & Siatczynski, 1988; Study 2). But, it also is worth mentioning that Howard & Conway (1986) note enormous differences among subjects, such that several subjects did not show any evidence of trying to play the role of "good subject." Next, Hayes' (1987) reply focuses on research on "social standard setting" which speaks to our findings. Hayes believes subjects are able to behave in particular ways in experimental settings because they are "public" settings. But in "private" (read, nonexperimental) contexts, these subjects would be unlikely (perhaps even unable) to behave in the same manner. Although Hayes' research highlights an interesting factor in this domain, what remains unclear is exactly how important the public versus private nature of the data is in our studies. Consider the following thought experiment. You set a jar of peanuts on your desk and flip a coin each day to determine whether it would be an "eat" or "try not to eat" day. Then you choose a colleague to tell which set of instructions you are entertaining, and you show him or her your daily results (public phase). Our data suggest that you might consume about six times the weight in peanuts on "eat" days as on the "try not to eat" days. Next, inform your colleague that your personal experiment has been completed, but you continue the same set of procedures without letting anyone else know (private phase). We would

344

George S. Howard and Paul R Myers

not be at all surprised if you found that the effect of volitional control remained unchanged in the private phase. Nor would we be concerned if, for example, your average weight consumed on "eat" days was only (for example) four times the amount consumed on "try not to eat days." The decline (from a sixfold to a fourfold advantage for "eat" days) represents an estimate of the impact of the private versus public nature of the experimental context. When one of the present authors (Howard) conducted this experiment on himself, the advantage of "eat" days in the private phase was actually greater than the advantage of the "eat" days in the public phase. But we are completely confident that if Hayes conducted the same experiment upon himself, he would be perfectly able to show no effect of volition in the private phase--if he wanted to demonstrate such an effect. What would surprise us is if you (the typical reader) were unable to eat more peanuts on "eat" days than on "try not to eat" days in the private condition. Of course, there is a simpler way of thinking of Hayes' public versus private challenge to our studies of volition. If one considers the issue to be one of external validity, the findings might be summarized as follows: In public settings (such as therapy) subjects are able to achieve their agentic goals more satisfactorily than in private contexts (such as self-improvement efforts) (see Howard, DiGangi, & Johnson, in press). Thus, if one is interested in public, psychological activities such as psychotherapy, the public studies we presented likely possess greater external validity than the private research alternative.

Empirical Investigations of the Obedience Hypothesis Because of the possible plausibility of the conformity objection to the original volition studies, the following series of investigations explicitly attempt to test the volitional interpretation versus the "control via the experimenter's instructions" interpretation. In one study (Howard & Conway, 1986, Study 2) subjects sometimes received their daily "eat" or "not eat" instructions via a coin toss, and the experimenter was unaware as to what condition the subject was in for that day. Subjects still demonstrated a strong effect of volition. At other times, subjects simply chose and recorded whether that particular day would be an "eat" or a "not eat" day (and also did not let the experimenter know the condition they chose). In this condition, subjects also showed strong volitional control. Therefore, subjects showed volitional control in two different types of situations where the experimenter not only did not give the "eat"

Experimental Investigations of Volition

345

and ''not eat" instructions, but also the experimenter was not even aware of whether subjects were in an "eat" or %ot eat" condition on any day. In yet another study, Howard, Youngs, & Siatzcynski (in press, Study 2) had subjects choose and record each day whether they would "follow instructions" or "do the opposite" of the instructions given (the meta-volitional factor). Subjects were then told by the experimenter either to "eat" or "not eat" peanuts that day. As expected, the volition by meta-volition interaction was significant and accounted for 65% of the within-subject variance. That is, subjects ate many more peanuts on "eat" days than on 'hot eat" days (135 g. versus 10 g.) when they had chosen to follow the instructions. However, when they decided to "do the opposite" they consumed far more peanuts in the "not eat" condition than in the "eat" condition (120 g. versus 3 g.). These studies serve to lessen the plausibility of the "they were compelled to obey the experimenter's instructions" efficient-cause objection to the volitional interpretation of the above studies. T.he intriguing aspect of the studies of volition reviewed above is that through the random assignment of conditions of volitional control (e.g., "eat-not eat", "binge-not binge", "initiate conversations-do not initiate conversations", "exercise-do not exercise", etc.) to time blocks, all possible explanations for mean differences between the two conditions, save two, are rendered implausible. The two possible explanations are: 1) that these mean differences reflect the agents' power of self-determination (or behavioral freedom, or volition) in this particular instance; and 2) that subjects were compelled to obey the experimenter's instructions and could not do otherwise (or, similarly, they had been so thoroughly socialized into a "good subject" role and therefore could not choose to disobey). Thus, the crux of the difference between these two explanations involves whether subjects obey their own directives or the directives of the experimenter. Howard (in press) addressed the above question of who causes the subject's behavior (the subject himherself or the experimenter through the experimental instructions to the subject) by collapsing the distinction between the subject and the experimenter. Thus, the author served as both experimenter and subject for the study. Enormous volitional control of alcohol consumption was evident in this study. But if as both experimenter and subject he was merely conforming to the experimental instructions, then he was conforming to his own commands--but this is precisely the character of volition.

346

George S. Howard and Paul R Myers

Volition as a Factor in Therapeutic Interventions We will now turn to applied and/or therapeutic investigations and applications of the above designs. Many psychologists have observed that self-determination is a critical factor in counseling and psychotherapy. Schultz (1977) concludes, after drawing from the collected theories of Gordon Allport, Carl Rogers, Erich Fromm, Abraham Maslow, Carl Jung, Viktor Frankl and Fritz Perls, the following: Perhaps the only point on which they agree fully is that psychologically healthy persons are in conscious control of their lives. Healthy persons are capable of consciously, if not always rationally, directing their behavior and being in charge of their own destinies (p. 143). Thus, it has long been held by therapists that many of their clients' troubles stem from their inability to control either their environment or their response to the environment (Mahoney & Thoresen, 1974). The process of therapy often involves the process of reestablishing such control. After having demonstrated a methodology for empirically assessing volitional effects in a series of studies on peanut eating, one reviewer wondered whether the concept of volition generalizes beyond the shell of a peanut. Indeed, Howard and his colleagues have considered the role of volition in domains that are of more interest to practicing psychologists. Many practitioners (e.g., clinical, counseling, industriaVorganizationa1, and school psychologists) are often involved in behavior that their clients have difficulty controlling. At first blush it might appear that volition would not be a useful construct to aid these applied psychologists in their ministrations. For example, simply telling a schizophrenic to "stop hearing those voices," or instructing a depressed client to "feel less depressed," would seem to be singularly unhelpful. However, a slight modification of the procedures employed in the "peanuts studies" of volition has yielded some interesting findings in several clinically important domains. Rather than attempting to volitionally control the problem behavior directly, these studies adopt a strategy similar to that practiced by most experienced clinicians. Namely, subjects were encouraged to exercise volitional control over the conditions that serve to maintain the problem behavior. We would like to suggest that the following studies be viewed as the beginning of a bridge building process that will establish a direct link between research and the practice of therapy from a volitional perspec-

Experimental Investigations of Volition

341

tive. The following sections highlight the issues addressed by volition research to date. Frequency of Heterosexuul Interaction. In one study (Howard & Conway, 1986; Study 3) college students, who wished to increase the frequency of their heterosexual social (heterosocial) interactions, were recruited for a study in which they were encouraged to exert volitional control over three factors related to heterosocial interactions: (a) the number of conversations initiated with members of the opposite sex; (b) the amount of time spent in places where social interactions frequently occur (e.g., dining hall, student center, parties, etc.); and (c) the frequency of positive self-statements students make about themselves and their social skills. Control of all three factors was structured in the same "try to --/try not to --I' paradigm used in the peanut studies. It should be noted that, while few subjects in the previous "peanut studies" doubted their ability to control peanut consumption, the subjects in this study identified heterosexual social skills as an area of concern. The data revealed that on the group analysis level, subjects were able to control all three of these conditions related to the number of heterosocial interactions, and that in so doing they were extremely effective in achieving their goal of having more (and more satisfying) heterosocial interactions. Control of Snacking and Exercise. Lazarick, et al. (1988; Study 1) considered the degree to which eating and exercise habits are under an individual's volitional control (for control of exercise see also, Howard, DiGangi, & Johnson, 1988). Subjects were divided into two groups, those who wanted to lose weight and those who did not particularly care to lose weight. After an initial baseline observation period, subjects were given a container of vegetables and one of four sets of instructions: (a) To snack on as many vegetables as he or she liked but try not to exercise; (b) try not to snack on the vegetables and also try not to exercise; (c) snack on vegetables and exercise as much as he or she wished; and (d) try not to snack on the vegetables but exercise as much as he or she wished. The order of presentation of the four conditions was counterbalanced across subjects. The results indicated that subjects could control both snacking on vegetables and exercising. Group differences were nonsignificant for the "desire to lose weight" variable. It should also be noted that the volitional control over snacking and exercising did not translate into weight loss as clearly as was hoped.

348

George S. Howard and Paul R. Myers

Control of Time Spent Researching Vocational Information. In another study reported by Lazarick et a1 (1988; Study 2), the researchers considered how pursuing two types of information, vocational and personal, impacted on career indecisiveness. Subjects were given two information packets. The vocational packet contained information and exercises on the meaning of work, inventories pertaining to work values, talents, interests, etc. The personal p a c k t covered topics like anxiety, goal setting, self-confidence, self-identity, etc. Subjects were then assigned to one of three conditions: (a) search for personal information, (b) search for vocational information, and (c) try not to search (purposely avoid attempting to discover about oneself or the world of work). The dependent measure was time spent engaged in the searching activity. The study's results indicate that not only do subjects have a considerable amount of control over their behavior, but they were able to follow the "try not to search'' instruction even though they reported dissatisfaction with this condition. Control of Singeing Behavior by Bulimics. Two studies (Lazarick et al., 1988; Study 3; Steibe & Howard, 1986) in this series have obtained strong evidence for the efficacy of a volitional treatment of binge eating and bulimia by using a "try not to/ act normally" paradigm. The findings demonstrate that frequency of binge eating episodes can be reduced and subjects can replace high caloric foods with vegetables. Thus, binge eating is modifiable in the short run by efforts of will, but other issues remain unresolved making long term control unlikely. Control of Social Consumption of Alcohol. Finally, Howard (1986), Howard (1988) and Howard, Curtin, & Johnson (1988) showed, with a slight variation in methodology, that social drinking is, indeed, under volitional control. In Howard, Curtin, and Johnson (1988, Study l), subjects exerted control over their consumption of alcohol. The study had two phases. First, the subjects were to follow a baseline, monitoring period. The second phase was the target hitting period in which they were to drink only as many glasses of alcohol as were indicated by the predetermined targets. Inaccuracy scores for the target hitting period could then be compared to inaccuracy scores from the monitoring period. The latter were determined by comparing 'hormal" drinking patterns to a set of randomly selected targets. To operationalize volition, then, subjects converted a significant message (FREE WILL), via Morse Code, into target numbers. That is, a 2 drink target represented a "dash", a one drink target depicted a "dot", a 0 drink target represented a space between letters, and two consecutive

Experimental Investigations of Volition

349

0 drink targets signaled a break between words. The subjects' volition was measured by how close they came to spelling "FREE WILL" and by how much they improved their accuracy from the baseline period to the target hitting period. [It should also be noted that half of the subjects knew that they were spelling the words "FREE WILL" (meaning condition), while the other half only knew the target numbers (less meaning condition). The role of meaning will be discussed below]. The results were striking. For example, those in the high meaning condition achieved 100% accuracy and those in the low meaning condition, while unable to spell the words precisely, improved their accuracy from baseline to intervention by up to 86%. Thus, these results invite another round in the controversy over whether the increased use of alcohol over time by nonaddicted individuals is best understood as a form of progressive physical addiction, a breakdown of volitional control, or both. Future research in this area with special populations and with the control of diverse environmental variables should prove both interesting and fruitful.

Implications of the Volition Studies Peter van Inwagen (1983) has pointed out that there have been many debates about free will versus determinism over the years. We believe that the recent findings on volition have nothing to say about many of these debates (e.g., when free will is conceived of as representing simply the absence of physical constraint). However, when one ponders the notions of mechanistic determinism and self-determination articulated above, we believe the evidence of strong volitional control in certain areas of human action represents strong warrant for belief in the thesis of self-determination, and sounds death-knell for the thesis of complete mechanistic determinism. Given the data suggesting the implausibility of the conformity explanation, and the fact that the random assignment procedure equated the groups (i.e., ?ty to --'I; 9ry not to -") on the average on all other possible explanations, the enormous differences found on the dependent measures can only be attributed to the causal force of self-determination. Such findings represent strong support for free will advocates' claim that these subjects might indeed have chosen to do otherwise, ceteris paribus. There are, of course, important implications also for the human sciences. The standard view among social scientists is that science can not provide evidence of human freedom. But this stance does not

350

George S. Howard and Paul R Myers

necessarily mean that social scientists reject the reality of human freedom. The sociologist Peter Berger (1963), for example, adopts the following position: Freedom is not empirically available. More precisely, while freedom may be experienced by us as a certainty along with other empirical certainties, it is not open to demonstration by any scientific methods...Every object of scientific scrutiny is presumed to have an anterior cause. An object, or an event, that is its own cause lies outside the scientific universe of discourse. Yet freedom has precisely this character...The individual who is conscious of his own freedom does not stand outside the world of causality, but rather perceives his own volition as a very special category of cause, different from the other causes that he must reckon with. This difference, however, is not subject to scientific demonstration...There is no way of perceiving freedom, either in oneself or in another being, except through a subjective inner certainty that dissolves as soon as it is attacked with the tools of scientific analysis (pp. 122-124). While Berger's claims were originally factually correct, the recent methodological refinements presented herein now make possible demonstrations of precisely what Berger claimed to be impossible. That is, we are now able to provide scientific evidence of self-determination or behavioral freedom in human actions. Such findings do not preempt traditional psychological investigations into how nonagentic factors influence human behavior. Rather, the fact that less-than-perfect self control is evidenced by almost all subjects in the studies reviewed above suggests that further research needs to probe the ways in which biological inheritance, current physiological states, environmental arrangements, developmental experiences, cultural norms, and so forth serve to modulate one's ability to self-determine. There are many nonagentic factors that exert coercive influences in our lives. These influences represent conditions that serve to either increase or decrease the likelihood that persons can fine tune their control of their actions. It behooves even the radical free will advocate to understand the ways in which "will" is bounded by various nonagentic influences.

Experimental Investigations of Volition

351

Finally, we knew since the time of Gregor Mendel that many physical characteristics resulted from one's genetic inheritance. But science is far from satisfied with such a gross understanding of a phenomenon. With the unraveling of the chemical structure of the gene by Watson, Crick, and others, a much more satisfactory scientific understanding was achieved because we now have far greater insight into how (the mechanism whereby) genetic transmission actually occurs. The studies reviewed above might be analogous to Mendel's crude insights in that they simply demonstrate that in certain circumstances people can volitionally control their actions to a certain degree (with all other possible explanations of these findings reasonably well controlled methodologically). But we will remain unsatisfied until we understand precisely how it is that (i.e., what is the mechanism whereby) a person achieves the ability to self control his or her actions. A promising start in answering this question has been made by Howard, Curtin and Johnson (1988) who find meaningfulness to be strongly related to subjects' ability to demonstrate self-determination in their actions. Howard (1986, 1989) has explored some of the ways in which a psychology of self-determined, meaningful action might profitably be pursued. From this ''human science" perspective, psychologists would be equally interested in the "reasons for" and the "causes of'' human behavior. A complete explanation of human action would include the role of the subject's understanding of what is taking place--and how that generic understanding was causally efficacious in the genesis of the behavior of interest. This broadened, integrated perspective should be exciting for practitioners in psychology, because it weds the agentic stance of the applied psychologist to a solid program of research aimed at assessing the impact of self-determination on human actions. The scientific psychologist will be intrigued because this holistic approach can sometimes account for enormous amounts of the variance in human behavior--sometimes up to 100% of the within-subject variability (see, Howard, Curtin & Johnson, 1988) in important domains of human action (e.g., alcohol consumption).

References Berger, P. L. (1963). Invitation to sociology: A humanistic perspective. New York: Doubleday & Co. Cronbach, L. J. (1982). Designing evaluations of educational and social programs. San Francisco: Jossey-Bass.

352

George S. Howard and Paul R Myers

Hansen, N. R. (1958). Patterns of dkcovery: A n inquiry into the conceptual foundations of science. Cambridge: University of Cambridge Press. Hayes, S. C. (1987). Contextual determinants of ”volitional action”: A reply to Howard and Conway. American Psychologist, 42, 10291030. Howard, G. S. (1986). Dare we develop a human science? Notre Dame, IN: Academic Publications. Howard, G. S. (1989). A tale of two stories: Excursions into a narrative approach to psychology. Notre Dame, IN Academic Publications. Howard, G. S. (in press). Can science furnish evidence of human freedom?

Nature. Howard, G. S., & Conway, C. G. (1986). Can there be an empirical science of volitional action? American PJychologist, 41, 1241-1251. Howard, G. S., Curtin, T. D., & Johnson, A. J. (1988, August). The hardening of a ”sofr”science. Invited Address; Mathematical and Statistical Models of Behavior Track; Science Weekend; APA Convention; Atlanta. Howard, G. S., DiGangi, M. L., & Johnson, A. (1988). Life, science, and the role of therapy in the pursuit of happiness. Professional Psychology: Research and Practice, 19, 191-198. Howard, G. S., Youngs, W. H., & Siatczynski, A. M. (in press). Reforming methodology in psychological research. Journal of Mind and Behavior. Kuhn, T. S. (1962). The structure of scientific revolutions. Chicago: University of Chicago Press. Kuhn, T. S. (1977). The essential tension. Chicago: University of Chicago Press. Lazarick, D. L., Fishbein, S. S., Loiello, M. J., & Howard, G. S. (1988). Practical investigations of volition. Journal of Counseling Psychology, 35, 15-26. Mahoney, M. J., & Thoresen, C. E. (1974). Self-control: Power to the person. Monterey, CA. Brooks/Cole. Orne, M. T. (1962). On the social psychology of the psychological experiment: With particular reference to demand characteristics and their implications. American Psychologist, 17, 776-783. Schultz, D. (1977). Growthpsychology:Models of the healthypersonality. New York Van Nostrand. Steibe, S. C., & Howard, G. S. (1986). The volitional treatment of bulimia. The Counseling Psychologist, 14, 85-94. van Inwagen, P. (1983).An essay on free will. Oxford: Oxford University Press. Weber, S. J., & Cook, T. D. (1972). Subject effects in laboratory research: An examination of subject roles, demand characteristics, and valid inference. Psychological Bulletin, 77, 273-29s.

VOLITIONAL ACTION, W.A . Hersh berger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

353

CHAPTER 15 CONTROL THEORY AND PSYCHOLOGY: A TOOL FOR INTEGRATION AND A HEURISTIC FOR NEW THEORY Michael E. Hyland There are several views about how control theory relates to "conventional" psychology. One view, and a view which I supported several years ago, is that control theory has got hold of "the real truth" and the other theories in psychology are simply wrong. If this view is accepted, then much of the research in psychology can be ignored because such research starts from incorrect theoretical assumptions. It is common experience that in most disputes, whether theoretical or not, each of the opposing parties is partly in the right and partly in the wrong. After several failures to convince others that control theory was correct and their own approach incorrect, I began to wonder whether this simple solution about who was right and who was wrong might be flawed. There are two reasons why a simplistic rejection of all other psychological theories in favor of control theory can be questioned. The first reason is that psychological theories are very heterogenous; they include behaviorist, physiological, phenomenological, cognitive, motivational, ' and many other theories. Thus, a simple rejection of "psychological theories" as a generic category fails to recognize the great diversity of approaches in psychology. In fact, many of the demonstrated advantages of control theory compared to other theories is where the other theory is based on behaviorist principles. Behaviorism lost its position of influence in main stream psychology many years ago, and a demonstration of the disadvantages of behaviorism is neither new nor peculiar to control theory. Indeed, arguments against behaviorism form a standard part of introductory theoretical psychology (e.g., Hyland, 1981). Nor will the argument that control theory is better than behaviorism be seen as relevant by the many psychologists who have long since found it unnecessary to argue that their approach is better than behaviorism. A second reason for questioning an approach which rejects "all other psychological theories" results from an examination of a particular

354

Michael E. Hyland

group of theories in psychology, namely, theories of motivation. If a comparison is to made between control theory and a particular theory in psychology, the most useful comparison must be theories of motivation because motivation theories, like control theory, were introduced specifically to explain purposive behavior. A careful reading of motivational theories, particularly in their early statements, shows that motivational constructs are often formulated in ways which show some cognizance of ideas which are more fully developed in control theory. For example, Hull (1943, p. 15) wrote "When...conditions deviate appreciably from the optimum a state of need is said to exist. Murray (1938, p 123-124) wrote "a need ...is a force...which organizes...action in such as way as to transform in a certain direction an existing, unsatisfymg situation." According to Lewin (1936, p. 156) "The reaching of a goal can at the same time mean the release of tension", and Tolman (1932, p. 272) explained how drives are the consequence of a disturbing stimulus which causes a physiological disturbance. Without wishing to read too much into these early theoretical statements, it would seem that primitive control theory ideas were present even before Wiener's classic formulation in 1948, and therefore it may be wrong to see control theory as being in opposition to the more conventional motivational theories. I have developed an alternative to the view that supposes that control theory is right and everyone else wrong. This alternative view is that control theory provides a new language for representing the processes underlying purposive behavior, but that many of these processes are described in conventional psychological theories (Hyland, 1986, 1987, 1988). If control theory is another language for describing theoretical ideas expressed in other parts of psychology, then the burden must lie with control theorists to show that their language is useful. As a reviewer of one of my early papers put it--the author must actually show that it is worth the reader's while learning this new and somewhat complex language. Why is the language of control theory worth learning? The advantages of control theory stem from the fact that it provides a microanalysis of theories expressed in other areas of psychology--much in the same way that physics provides a microanalysis (ie., explanatory level of description) of chemistry. By providing a microanalytic level of description, control theory makes two theoretical contributions. First, it provides a common language which shows how different theories are related and how they are different. Specifically, control theory describes a general framework where each of the individual theories forms part of

Heuristic Control Theory

355

that framework. Second, as control theory is more holistic than individual theories of motivation, new theoretical ideas and predictions emerge which cannot be derived from the less holistic individual theories. In this chapter I begin by describing how I have used control theory as an integrative framework for understanding theories of motivation. The idea that control theory is at a different level from conventional psychological theories has been suggested independently be Klein (in press) who refers to control theory as a "meta-theory'' and who suggests that Yhe current, component theories would be consigned to the role of 'middle range' theories." Klein's work, unlike my own, emphasizes the empirical integration possible from control theory. The second part of the chapter is concerned with new theoretical insights which follow from the holistic orientation of control theory. Two topics are considered: consequences of differing types of reference criteria and mode theory, the latter being a new approach to behavioral inconsistency in personality.

Motivational Control Theory I have used the label motivational control theory (Hyland, 1988) to describe the use of control theory as a special language of description which contributes to the field of motivation in psychology. The basic structure of motivational control theory is a hierarchical arrangement of control loops, and the simplest form of this structure is shown in Figure 1. Additional components can be added on, depending on the range of theories considered. Figure 1 shows the conventional arrangement of a comparator comparing a reference criterion and perceptual input such that detected error leads either to other lower level reference criteria or to behavior which has the effect of reducing detected error. There are, however, two additions to this conventional arrangement. The first is the ampZifier which is responsible for signal amplification between the comparator and behavior. The degree of signal amplification is called error sensitivity as it expresses the extent to which behavior is sensitive to error. An everyday example of the effect of error sensitivity on behavior is given by Hyland (1988, p. 644): Suppose that two people share a room and have the same goal: that the room should be clean. That is, both people have the same reference criterion about the level of cleanliness that is desired. However, one of these people

356

Michael E. Hyland finds dirt a salient issue, whereas the other does not find dirt very upsetting. The former is likely to put more effort into keeping the room clean. The latter, the one who does not find dirt so upsetting, will not put as much effort into keeping the room clean, but will appreciate the work done by the more fussy partner.

In common sense terms, error sensitivity relates to the intensity of behavior. That is, it relates to the effort which a person invests in achieving a goal. Goals which are more salient (i.e., where there is a greater sensitivity to error) will in general attract more effort and be pursued with greater intensity than goals which are less salient. The second addition to the conventional arrangement of a hierarchy of control loops is the selector. Detected error in a superordinate control loop can often be eliminated in a variety of ways. For example, feelings of low self-esteem can be eliminated by making money, getting publications, being loved by other people, and so on. As the saying goes, there are many ways to skin a cat. However, the person may need to select just one subordinate goal as a way of achieving the superordinate goal. Such choices, and the rules governing such choices, are made in a component called the selector. It is important to emphasize how the introduction of a selector affects the hierarchy of control loops. Instead of supposing that detected error from a superordinate loop automatically elicits a reference criterion of a subordinate loop, we must now allow for the fact that the relationship between superordinate and subordinate level loops is a variable. This variable (i.e., the particular way superordinate and subordinate loops are related) affects the organization of the hierarchy. Thus the organization of the control hierarchy is itself a variable.

Theories of Motivation Expressed in Terms of Motivational Control Theory There are three distinct motivational traditions in the psychological literature, and each tradition relates to a different aspect of the hierarchical system expressed in motivational control theory. One tradition focusses on variation in error sensitivity, a second focusses on variation in reference criteria, and a third focusses on variation in the organization of a control hierarchy. Each of these traditions will be described and then related to motivational control theory.

Heurktic Control Theoiy

357

@

[m3

= comparator =Amplifier

E l

= Selector

I

I

memory

I 7 1

ENVIRONMENT 1

H I

L

BEHAV1oR

I

Figure 1. Two levels in a hierarchical control system are shown, where RC = Reference Criterion, DE = Detected Error, and PI = Perceptual Input. The selector and the amplifier are both affected by inputs from memory and from the environment.

Murray (1938) presented a view of personality based on individual differences in purposive behavior. Murray suggested that humans share some thirty needs, and that these needs were aroused by aspects of the environment. People differ in terms of the arousability of their needs. So, for example, if someone had a highly arousable need for affiliation, that person would spend more time affiliating than someone with a highly arousable need for achievement who would spend more time achieving. Murray’s ideas were incorporated into several later theories, notably those of McClelland (1951,1961) who substituted the term motive for need, and Atkinson (Atkinson, 1957, 1981; Atkinson & Birch, 1970, 1974, 1978). These different theories share the idea that purposive behavior can be understood in terms of a limited number of motives and that

358

Michael E. Hyland

individual differences arise from differences in the strength of motives. Further, these theories are interactionist (Atkinson, 1981) in that behavior is believed to result from an interaction between personality and the environment: Variability in behavior results from (a) environmental conditions and (b) a personality component which reflects the arousability (i.e., sensitivity) to those environmental conditions. In terms of motivational control theory, the tradition started by Murray is one where the theorist focusses on a limited number of reference criteria which are assumed to be shared by many or all people. Individual differences occur not because people have different reference criteria, but because they have different sensitivities to error. Variability in behavior results from (a) environmental conditions as reflected in the perceptual input and (b) a personality component which reflects variation between people’s level of error sensitivity on a particular control loop. The greater the detected error and the greater the error sensitivity, the greater the effort which is invested in a task. In contrast to an approach emphasizing the strength of a limited number of motives, Lewin (1935, 1938) suggested that individuals follow a potentially infinite number of different goals where goal following involves sequences or steps towards other goals. Lewin’s emphasis on the goals themselves was a precursor to Locke’s (1968) theory of goal setting, a theory which has become an important part of occupational psychology, (Locke, Shaw, Saari & Lantham, 1981). According to goal setting theory, -people set themselves goals or have their goals set for them, and work performance depends on the particular nature of those goals. From the perspective of motivational control theory, goal setting theory explains individual differences in terms of people having different reference criteria, rather than different error sensitivities in loops involving the same reference criterion. By focussing on the reference criteria, it is also possible to account for individual differences in effort invested in a task Since different goals may require different amounts of effort, effort is mobilized simultaneously with direction in proportion to the perceived requirements of the goal or task. (Locke et al., 1981, p. 132) A number of empirical predictions arise from goal setting theory which are entirely consonant with control theory predictions (Klein, in press), for example, setting difficult goals (i.e., higher standards) is

Heuristic Control Theory

359

expected to produce better results-the higher standards would, of course, produce greater detected error. The third type of motivational tradition in the psychological literature focusses on cognitions. According to Weiner (1972, 1980) people differ in terms of the attributions which are made for the cause of an event. One person, for example, may attribute the cause of success in examinations to ability and effort; another may attribute the cause of success to luck. The former individual will try to achieve the superordinate goal of passing an exam through effort and ability seeking activities, for instance, by studying. The latter individual will seek the superordinate goal of passing an exam, either by doing very little or, if he or she believes that luck can be enhanced, through luck enhancing activities such as prayer or helping others. Thus, depending on the attributions an individual has for the cause of the outcome on a superordinate goal, different subordinate goals will be selected to attain that superordinate level goal. From the perspective of motivational control theory, attributional accounts of motivation focusses on the organization of the control hierarchy. Individuals differ in the way that lower level reference criteria are selected to reduce the error of some higher level control loop; that is, people's selectors differ. In summary, different motivational traditions in psychology focus on different aspects of the processes underlying purposive behavior. These different aspects can be described colloquially as (a) "how much you want it," (b) "what you want," (c) "how you go about getting it." In motivational control theory, the "how much you want it" or the intensity of goal seeking is explained in terms of some joint function of detected error and error sensitivity. The "what you want" is explained in terms of the reference criterion." And the "how you go about getting it" is explained in terms of the selector as the selector determines the organization within the control hierarchy. Motivational control theory, therefore does not supplant any of the other approaches to motivation, but it does provide a framework which shows how these other approaches are related. In contrast to advocates of a particular theory of motivation who often believe their theory to be correct and others incorrect, motivational control theory shows that all motivational traditions are correct--though each tradition is only a partial statement of the truth. The different theories of motivation are shown through motivational control theory to be complementary rather than in competition.

360

Michael E. Hyland

Motivational Control Theory as Innovation Motivational control theory does not show other theories of motivation to be wrong but this does not mean that it is only a summary of those other theories using a more general language. There are several ways in which motivational control theory can introduce novel ideas in current motivational understanding.

Reference Criteria Hyland (1988) suggests that there are four different types of reference criterion. These are (a) end states, for example, building a house; (b) rate of progress towards an end state, for example, rate of progress in building; (c) doing and being goals, for example, achieving, self actualization or affiliating; and (d) emotions and other internal mental states--e.g., pleasure, inebriation. In traditional literatures on motivation, the assertion that people seek goals or are motivated obscures the very important differences which arise between these different sorts of reference criteria. I will consider two ways in which a comparison between different types of reference criteria has consequence. First, consider end state goals on the one hand and emotions and internal mental states on the other. Satisfactory attainment of an end state (i.e., reduced detected error) is likely to have emotional consequences which may themselves be desired. Thus completing a task does not just reduce detected error in a control loop where the reference criterion is an end state but may also reduce detected error in another control loop where the reference criterion is an emotion. Completing a task may reduce detected error on more than one control loop. Let us assume that individuals vary in terms of the relative error sensitivities of end state based loops and emotion based loops. That is, some people are more concerned with achieving concrete ends whereas others are more concerned about achieving pleasurable states and avoiding dysphoric states. Many pleasurable mental states, however, can be achieved through means other than working on a task. For example, the state of inebriation, if it is sought, requires little effort given the appropriate resources. Thus, if individuals vary in terms of the relative error sensitivities of end state based and emotion based loops, then individuals should also vary in terms of how much they seek pleasure and how much they seek end states. Thus a personality disposition of pleasure seeking may reflect relatively greater error sensitivity in emotion based control loops compared to a personality disposition of work

Heuristic Control Theory

361

orientation which may reflect relative greater error sensitivity in end state control loops. A pleasure seeking versus work oriented dimension is to be expected from Hyland’s (1988) classification of different types of reference criteria and, moreover, has face validity. Yet this dimension does not feature in any theory of motivation with one exception to be considered later, namely reversal theory (Apter, 1982). A second example of the importance of considering the type of reference criterion concerns the predicted and observed relationship between task difficulty on the one hand and preference and performance (these two tend to be related) on the other. Theories of work motivation, such as that of Vroom (1964), predict and show that people prefer and do best on easy tasks. On the other hand, theories of achievement motivation, such as that of Atkinson (1957), predict and show that people do best on tasks of intermediate difficulty. The predictions of these two types of theories are treated as being competitive and the differing observations as irreconcilable. However, the different predictions and observations between the two theories arises because each considers a different type of reference criterion. If the reference criterion is an end state, common sense dictates that people will prefer the easiest way to attain that end state. Hyland (1988) gives the example of chopping a pile of logs. People don’t try to make the task of chopping wood more difficult by standing on their heads. People generally try to achieve end states the easy way. On the other hand, if a person has a goal to be a successful mountaineer, then a climb up a steep hill may well be less preferred than a more challenging and more difficult climb, as the climb up a steep hill will not contribute to a sense of being a good mountaineer. Where goals are doing or being goals, then the easiest task is not necessarily the best way of reducing error between the detected error and reference criterion. Thus, where doing or achieving goals form the reference criterion tasks (e.g., the pursuit of excellence) then moderately difficult tasks are preferred because (a) they are sufficiently difficult to provide a sense of accomplishment and (b) they are not so difficult that the person is likely to fail. On the other hand, where an end state forms the reference criterion, then an easy task should be preferred because the easy task increases the likelihood of successful completion. In summary, the type of reference criterion is an important factor in the way a control hierarchy works. Attention to the different types of reference criteria can lead to potentially important personality comparisons (e.g., pleasure seeking versus end state oriented) and can prevent

362

Michael E. Hyland

pseudo-empirical controversies (the effect of task difficulty on performance).

Mode Theory Personality theory is based on the assumption that behavior is consistent over time and consistent between situations. Language and common sense suggests that behavior does indeed exhibit such consistencies, and such consistencies are described with the many trait descriptions which are found in the English language. Research, however, has demonstrated relatively low levels of consistency and this has led some researchers (e.g., Mischel, 1968) to reject the common sense and classical trait view. Such researchers present a view where behavior is the consequence of an interaction between a stable person variable and the situation. This interactionist view attributes inconsistency of behavior to situational variability, though there is still a stable person characteristic which can be described as personality. There is yet another explanation for the observed inconsistency of behavior. It is that (a) person characteristics as well as situations are inconsistent over time, (b) the way these person characteristics vary over time shows a regular pattern, and (c) the common sense view of personality derives from an averaging of behaviors over time. This alternative view, first advanced in reversal theory (Apter, 1982, 1984), will be elaborated within a motivational control theory framework to form a new theory which I call mode theory. In terms of current motivational approaches to personality, individual differences in behavior can result from three different kinds of theoretical variation: variation in terms of types of reference criterion, variation in terms of error sensitivity, and variation in terms of the organization of a control hierarchy. From a traditional interactionist perspective, each of these different types of variation is assumed to be constant over time for a given individual but to vary between individuals. For example, people high in achievement motivation are assumed to have higher error sensitivities on achievement oriented loops compared with people low in achievement motivation, and this difference between high and low achievers is assumed to be constant over time. Mode theory is based on the two assumptions. The first is that the error sensitivities of control loops are not constant but vary over time. The second is that variation in the error sensitivities of different loops are interdependent. The first of these assumptions is consistent with

Heuristic Control Theoty

363

theoretical ideas in other branches of psychology; the second assumption is specific to mode theory. Let us assume that error sensitivity varies, and that this variation is the consequence of three factors. First, there is a natural variation which is time based such that error sensitivity of control loops exhibits time based rhythms. Thus, instead of assuming that the theoretical entity contributing to personality is fixed, we assume that it varies over time. The biological advantage of such spontaneous variation in error sensitivity is that the organism engages in a variety of behaviors over a period of time irrespective of whether there is any situational change. Such a possibility of motivational variation was recognized by Tolman (1932, p. 274) many years ago who asserted that "Appetites are cyclical; aversions are relatively constant." Second, variation in error sensitivity is affected by behavior (as opposed to goal attainment). This possibility has been suggested in previous accounts of motivation, specifically as Hull's (1943) reactive inhibition and Atkinson and Birch's (1970, 1974) consummatory force. The suggested form of this relationship is that repeated activity to seek a goal can reduce the error sensitivity of that goal, and again, the biological advantage of such an effect is that it increases the variability of behavior over time. Third, variation in error sensitivity is affected by aspects of the situation, of which the most important are the presence or absence of a goal object and indicators of the likelihood of attaining a goal--the value and expectancy in expectancy-value theory. Thus, Tolman (1932) suggested that the presence of a goal object can increase the motivation for that goal object, and a good example of this effect is provided by research on eating behavior (Schacter & Rodin, 1974). Some individuals (external control) tend to feel hungry at the sight or smell of food; so, for example, the sight of a cream bun increases error sensitivity of a control loop where the reference criterion is eating the cream bun. That the likelihood of attaining a goal affects error sensitivity is suggested in several recent motivational theories. For example, Carver and Scheier (1986, 1987) suggest that individuals "disengage" from activities for which there is a low probability of success. The biological advantage of such disengagement is that the organism does not waste effort in trying to seek an unattainable goal. Assuming that error sensitivity is affected by (a) natural rhythms, (b) behavior, and (c) the situation, mode theory makes one additional

364

Michael E. Hyland

assumption; it is simply that there are causal relations between the error sensitivities of different control loops. The idea that causal relations occur between error sensitivities of different loops was proposed in its simplest form by Hyland (1988) who suggested that error sensitivity of a superordinate level loop should affect the error sensitivity of a subordinate level loop. This suggestion is needed to explain the fact that if a person is particularly concerned to achieve some higher level goal, he or she will invest more effort in the lower level goals which form the means for attaining that higher level goal. Mode theory goes beyond the hierarchical causal relationship of error sensitivities to suggest that error sensitivities are causally related in modes. A mode is a grouping of control loops where the grouping is based on some unspecified functional commonality between the loops. For example, loops where the reference criterion is an emotional state might form a mode which can be contrasted with loops referring to end states forming another mode. Following this example further, mode theory suggests that when one emotion-based loop increases in error sensitivity (for whatever reason) this will tend to increase the error sensitivities of other emotion-based loops. Thus, such an individual will tend to vary between modes of function, an emotion-based mode and an end statebased mode. The consequence of a motivational system operating in modes is that a person’s psychological processes vary between states, an idea which is suggested, purely descriptively, in reversal theory (Apter, 1982, 1984). According to reversal theory, individuals are assumed to reverse between states and where such reversals are to some extent independent of the situation. In reversal theory, most research has focussed on two alternative states, the paratelic and telic states, which can be characterized as a pleasure seeking and work oriented states. According this view, people tend to be in a pleasure seeking state (when they are playing) and tend to be in a work oriented state when they are working. Mode theory’s suggestion of emotion-based and end state-based reference criteria forming different modes provides an explanatory level description for reversal theory’s paratelic and telic states. However, there are important differences between mode theory and reversal theory which go beyond the fact that mode theory uses the language of motivational control theory and therefore provides a more atomistic, explanatory account. In reversal theory, the states which an individual is in are discrete dichotomies: For example, the individual is

Heuristic Control Theory

365

either in the paratelic state or the telic state. By contrast, in mode theory, the extent to which an individual is functioning in one particular mode is a continuous variable so that, for example, the error sensitivities of emotion-based loops can vary: This is analogous to suggesting that the paratelic-telic distinction is a continuum rather than a dichotomy. A second difference between mode theory and reversal theory is in terms of the type of modes or states which are assumed to exist. Reversal theory provides a listing of the states which are involved in reversals (Apter, 1988). By contrast, no such listing is provided in mode theory. Indeed, mode theory suggests that the modes themselves are variables. For instance, one individual may have a particular pattern of causal relationships between the error sensitivities of control loops whereas another individual may have a completely different pattern of causal relationships. The possibility that modes are themselves variables leads to a new approach to the problem of behavioral inconsistency in personality. As part of their rebuttal of Mischel’s critique of trait theory, Bem and Allen (1974) suggested that trait theorists had misunderstood Allport’s original formulation of trait theory by using traits nomothetically instead, as Allport did, of using them idiographically. In an idiographic use of traits, it is assumed that not all traits are relevant to all individuals. Instead, for each individual there are a limited number of traits on which the individual is cross-situationally consistent. Bem and Allen’s advocation of the idiographic use of traits was purely descriptive and lacked an explanation of why traits should be idiographic, an explanation provided by mode theory. A mode represents a particular group of control loops whose error sensitivities covary. Let us suppose that for a given individual, aggressiveness-oriented loops form a mode. That is, control loops which are characterized as “aggressive means of attaining an end” have causally connected error sensitivities. Then, for that individual, level of aggressiveness will tend to be cross-situationally consistent because the person’s tendency to aggression or lack of aggression will manifest itself in a variety of goal seeking situations. For example, if the individual has high levels of error sensitivity on these aggressiveness-oriented loops, then he or she may act aggressively to colleagues at work, as well as being aggressive in the home. Of course, this individual will not always be aggressive because situations, behaviors, or time-based oscillations could result in periods of low levels of error sensitivity on aggressive-oriented

366

Michael E. Hyland

control loops: An aggressive individual is often but not always in an aggressive mode. Now consider a second individual whose aggressive-oriented loops do not form a mode. For this second individual, aggressiveness will not be cross-situationally consistent as the error sensitivity of one aggressiveoriented control loop will not covary with the error sensitivity of another aggressive-oriented control loop. Thus, for example, this individual may adopt aggressive means for attaining superordinate goals at work, but be very nonaggressive at home. Thus mode theory provides a mechanism whereby people are cross-situationally consistent on some but not all traits. Further, even when a person is consistent on a trait, there are inconsistencies in behavior due to changes in error sensitivity brought about by natural time-based rhythms, behavior, and the situation. Mode theory's application to the idiographic-nomothetic trait distinction has several interesting theoretical properties. The first is that it bridges the gap between two types of concept: traits and goals. According to this view, traits are simply descriptive labels for particular types of goal or particular types of goal attainment. For example, a "methodical" person is someone who tends to have high error sensitivity on loops with goals which can be classified as "methodical" and where the error sensitivities of such loops are causally related. A "successful" person is someone who has managed to reduce detected error on a variety of control loops, and typically where these control loops have reference criterion which are positively valued in the culture in which the person lives. According to the mode theory perspective, a trait is a category which is applied by an observer for an actor's behavior. A trait exists in the observer's head rather than being part of the actor's psychological processes, and this is an important difference with the conventional approach to traits where traits are assumed to have some existential status within the actor (e.g., Royce & Powell, 1983). A second theoretical property is that mode theory explains rather than merely describes why some people are cross-situationally consistent on some traits but not on others--which is a weakness in Bem and Allen's (1974) account. As a trait is a descriptive category of many loops, consistency occurs only when the error sensitivity of these loops are consistent. Such consistency arises when the loops in question form a mode such that their error sensitivities covary. A third theoretical property of mode theory is that it suggests, like reversal theory, that people are intrinsically variable. Instead of suggest-

Heuristic Control Theory

367

ing that the person component to behavior is constant and behavioral variation is due only to the way the situation interacts with this person component, mode theory suggests that the person component is itself variable. People enter into different modes of functioning, and they behave as though they have different personalities depending on their mode. For example, a person who is often aggressive will exhibit all the properties of a nonaggressive person when he or she is not in an aggressive mode.

Summary A central argument of this chapter is that current theories of motivation are not incompatible with control theory. However, control theory provides a new language of description which, because it is explanatory, has two major advantages over other theoretical languages for describing purposive behavior. First, control theory shows conventional theories of motivation are not competitive but complementary to each other. That is, control theory provides a common theoretical framework in which existing theories of motivation can be placed. Second, control theory provides a more holistic framework than any other theoretical approach, and motivational properties exist at the more holistic level. This chapter presents for the first time how this holistic approach can be used in the development of a new theory, mode theory.

References Apter, M. J. (1982). The experience of motivation: The theory of psychological reversals. New York: Academic Press. Apter, M. J. (1984). Reversal theory and personality: A review. Journal of Research in Personality, I8, 265-288. Apter, M. J. (1988). Reversal theory as a theory of the emotions. In M. J. Apter, J. H. Kerr & M. P. Cowles (Eds.), Progress in reversal theory (pp 43-62). Amsterdam: North-Holland. Atkinson, J. W. (1957). Motivational determinants of risk taking behavior. Psychological Review, 64, 359-372. Atkinson, J. W. (1981). Studying personality in the context of an advanced motivational psychology. American Psychologist, 36, 117-128. Atkinson, J. W., & Birch, D. (1970). The dynamics of action. New York: Wiley.

368

Michael E. Hyland

Atkinson, J. W., & Birch, D. (1974). The dynamics of achievement-oriented activity. In J. W. Atkinson and J. 0.Raynor (Eds.), Motivation and achievement (pp. 271-325). Washington, DC: Winston. Atkinson, J. W., & Birch, D. (1978). A n introduction to motivation. Princeton: Van Nostrand. Bem, D. J., & Allen, A. (1974) On predicting some of the people some of the time: The search for cross-situational consistencies in behavior. Psychological Review, 81, 506-520. Carver, C. S., & Scheier, M. F. (1986). Functional and dysfunctional responses to anxiety: The interaction between expectancies and self-focussed attention. In R. Schwarzer (Ed.), Selfrelated cognitions in anxiety and motivation (pp. 111-141). Hillsdale, NJ: Erlbaum. Carver, C. S., & Scheier, M. F. (1987). Origins and finctions of emotion in a control-process model of action control. Unpublished manuscript, University of Miami. Hull, C. L. (1943). Principles of behavior: A n introduction to behavior theory. New York: Appleton-Century-Crofts. Hyland, M. E. (1981). Introduction to theoretical psychology. London: Macmillan. Hyland, M. E. (1986). Events, states and control theory. Cognitive Systems, I, 343-351. Hyland, M. E. (1987). A control theory interpretation of psychological mechanisms of depression: Comparison and integration of several theories. Psychological Bulletin,102, 109-121. Hyland, M. E. (1988). Motivational control theory: An integrative framework. Journal of Personality and Social Psychology, 55, 642-651. Lewin, K. (1935). A dynamic theory of personality. New York: McGraw-Hill. Lewin, K. (1936). Principles of topological psychology. New York: McGrawHill. Lewin, K. (1938). The conceptual representation and the measurement of psychological forces. Durham, NC: Duke University Press. Locke, E. A. (1968). Toward a theory of task motivation and incentives. Organizational Behavior and Human Performance,3, 157-189. Locke, E. A., Shaw, K. N., Saari, L. M., & Latham, G. P. (1981). Goal setting and task performance: 1969-1980. Psychological Bulletin, 90, 125-152. Klein, H. J. (in press). An integrated control theory model of work motivation. Academy of Management Review. McClelland, D. C. (1951). Personality. New York: Sloane. McClelland, D. C. (1961). The achieving society. Princeton: Van Nostrand. Mischel, W. (1968). Personality and assessment. New York: Wiley.

Heuristic Control Theory

369

Murray, H. A. (1938). Explorations in personality. New York: Oxford University Press. Royce, J. R., & Powell, A. (1983). Theory of personality and individual differences. Englewood Cliffs, NJ: Prentice-Hall. Schacter, S., & Rodin, J. (Ed.). (1974). Obese humans and rats. Washington, D C Erlbaum/Wiley. Tolman, E. C. (1932). Purposive behavior in animals and men. New York: Appleton-Century-Crofts. Vroom, V. H. (1964). Work and motivation. New York: Wiley. Weiner, B. (1972). Theories of motivation: From mechanism to cognition. Chicago: Rand-McNally. Weiner, B. (1980) Human motivation. New York: Holt, Rinehart, & Winston. Wiener, N. (1948). Cybernetics: Control and communication in the animal and the machine. Cambridge, MS: M.I.T. Press.

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hersh berger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

371

CHAPTER 16

THE BEHAVIORAL ILLUSION MISPERCEPTION OF VOLITIONAL ACTION J. Scott Jordan and Wayne A. Hershberger This chapter concerns the perception of volitional action and describes an experimental investigation of what William Powers (1978) has called a behavioral illusion: Whenever we observe an animal’s behavioral output varying systematically as a function of an environmental disturbance, we are inclined to assume that the observed behavioral relationship reflects properties of the animal. Since this assumption is, in fact, unwarranted, Powers has called the inclination an illusion. Powers was concerned principally with the conceptual inclinations of students of behavior, such as Skinner (1971, 1987), but the term illusion suggests a perceptual inclination that applies to us all. That is, it suggests that we all are inclined to misperceive the behavioral output of another person to be the other person’s doing (volitional action), when it is not--and by implication misperceive or even overlook their actions which are volitional. To qualify as a perceptual illusion the inclination would have to be replicable across observers, and amenable to environmental manipulation. We found that it was both. However, before proceeding to a consideration of our research, let us first explain its theoretical context. To behave voluntarily is to control, and to control is to maintain a correspondence between what is and what should be by offsetting disturbances to that correspondence. This control process is illustrated in Figure 1 in its canonical form. That is, the flow chart of any control system controlling a single parameter can, in principle, be reduced to a single negative feedback loop such as that depicted in Figure 1. The arrow labelled reference value specifies what should be. The arrow labelled input represents what is. The difference between the controlled input (what is) and the reference value (what should be) is labeled error. The error signal drives the control system’s output so as to offset the effects of environmental disturbances, thereby keeping the controlled input equal to the reference value. The feedback loop is negative; the error, in effect, negates itself.

372

J. Scott Jordan and Wayne A. Hershberger

(organism or mechanism)

- - - - - - - - (input)-1 L-

- - - - - - - - - - - - -\ f-(output)- - - -

--

(environment)

L

I

4+%

Compensatory

Figure 1. A canonical control loop mapped onto the interface (dashed line) between an organism (or mechanism) and its environment. The negative feedback loop of a canonical control system supports two emergent, lineal, cause and effect relationships. They are represented in Figure 1 by the two blocked arrows labeled Intentional actions and Compensatory reactions. The former cause and effect relationship exists between the reference value and the controlled input. Since the controlled input is the system’s doing, and the reference value is the system’s intended value of input, this functional relationship represents the intentional, or voluntary behavior of the system. The other cause and effect relationship exists between the environmental disturbance and the output. Since this involves the system’s automatic, compensatory reactions

The Behavioral Illusion

373

to environmental disturbances, this functional relationship comprises the system’s involuntary, or reflexively elicited, behavior. The behavioral illusion involves the mistaking of compensatory reactions for volitional actions. Powers demonstrated the case with the following task. He seated a subject in front of a computer terminal, handed him a joystick, and instructed him to use the joystick to control the position of a cursor on the computer screen (Cursor A), keeping it close to a fixed cursor (Cursor B). Although the vertical position of Cursor A was continuously being disturbed by a distal disturbance (a random number generator in the computer) the subject had little difficulty maintaining A close to B. Powers then altered the feedback function that related the position of Cursor A to the position of the joystick and had the subject perform as before. The subject continued to maintain a stable input pattern, keeping Cursor A next to Cursor B, but his output (manipulation of the joystick) immediately changed, reflecting the altered feedback function. After the feedback function had been altered, the subject continued doing the same thing, keeping ,A next to B. Only the environmental disturbances did anything different, in eliciting different compensatory outputs. That is, a disturbance’s effects on a control system’s output “is determined strictly and quantitatively by the inverse of the feedback function [whereby the system’s output affects its input] and is, therefore, a property of the environment and not the subject” (Powers, 1978, p. 432). So, when Powers changed the feedback function, he was also changing the subject’s involuntary behavior; that is, it was Powers, and not the subject, who was making the change. The subject did not change what he was doing, which was to control the position of Cursor A, by automatically offsetting the effects of environmental disturbances with compensatory output. That never changed. However, we are inclined to view the subject as having varied his behavior. The illusion is twofold. We regard the changing output as the subject’s changing his behavior, although it is not; and, we disregard the absence of a change of controlled input as being the subject’s behavior, although it is. We investigated the behavioral illusion using procedures similar to Powers. However, we asked our subjects to perceive rather than to perform. Specifically, we asked our subjects to watch as one of us (Jordan) controlled the horizontal position of a letter (x) on the screen of a computer terminal, using a computer mouse to offset computergenerated disturbances. The lateral movements of the mouse were reflected on the screen in terms of a second letter (0)whose horizontal

374

J. Scott Jordan and Wayne A. HershbeBer

position corresponded to the position of the mouse. The subjects were instructed that, after watching for a while, they would be asked to say which of the two letters best represented what they had observed the experimenter to be doing. The behavioral illusion would manifest itself as a preference for the letter 0. In a pair of related experiments, we found that this preference varied as a function of two independent variables, one being the continuity of the experimenter’s volitional action (Experiment l), and the other being the subject’s opportunity to work at cross-purpose with the experimenter (Experiment 2). But even under the most favorable of viewing conditions, the behavioral illusion persisted in the form of a slight preference for the letter 0. The particulars of our two experiments were as follows.

Experiment 1 Method Subjects. Subjects were 80 male and 80 female students enrolled in introductory psychology at Northern Illinois University. They were all volunteers and earned bonus course credit for participating. Within sex, subjects were randomly assigned to one of two experimental conditions (C or D, described below). Apparatus. The experimenter controlled the horizontal position of a letter x on the screen of a computer terminal (Digital Equipment Corp. PDP-11) by using a computer mouse (Radio Shack D31) to offset computer-generated disturbance to the horizontal position of the letter x. A letter o also appeared on the screen either above or below the x. The horizontal position of the o was a reflection of the mouse movements the experimenter made in order to control the letter x. Both cursors were able to move in a horizontal fashion only. Both could be moved right or left by rolling the computer mouse right or left respectively, across a horizontal shelf immediately beneath the terminal screen. For half of the subjects the letter o was above the letter x; for the other half, o was below x. Procedure. Subjects were run in small squads of five to eight individuals each. During such a session, the experimenter seated himself in front of the computer terminal and told the subjects to gather around in such a way that everyone was able to see the computer terminal and the computer mouse. The experimenter then told the subjects that the letters x and o would appear on the screen, that they were to view each of the two letters and the experimenter for 3 minutes, and that upon

The Behavioral Illusion

375

completion of the session they were to determine what the experimenter was doing, in other words, which of the two letters best represented the experimenter’s behavior. After watching the experimenter for three minutes, and writing on a secret ballot the letter which they regarded as the one best representing the experimenter’s behavior, subjects were asked to give a brief written explanation of why they chose that particular letter. Finally, they were asked to complete the two-item questionnaire described below. Conditions. For some squads, the experimenter continuously controlled the letter x for three minutes. This was Condition C, the continuously engaged condition. In other sessions the experimenter told the subjects that during the second minute of the three-minute session, he would in no way interact with the computer screen. In fact, he sat with his hands in his lap during the second minute. This was called Condition D, the discontinuously engaged condition. Questionnaire. The questionnaire read as follows: For items 1 and 2, circle the answer you feel to be correct. Circle only one answer for each item.

1) Letter x: A) Represents what the operator is doing intentionally, voluntarily. B) Represents what the operator is doing involuntarily. C) Does not represent what the operator is doing, voluntarily or involuntarily. 2) Letter 0: A) Represents what the operator is doing intentionally, voluntarily. B) Represents what the operator is doing involuntarily. C) Does not represent what the operator is doing, voluntarily or involuntarily. Two versions of the questionnaire were prepared, one with letter x as item 1, and one with letter o as item 1. Half of the subjects in each condition saw one version; the other half saw the other version. Scoring of questionnaire. Answers to items 1 and 2 form a response pattern. Response patterns were coded in terms of which of the three answers a subject chose for letter x and letter 0, where v = voluntary

376

J. Scott Jordan and Wayne A. Hershberger

behavior, i = involuntary behavior, and n = neither voluntary nor involuntary behavior. For example, the response pattern (xv, oi) means that letter x represents voluntary behavior, and letter o represents involuntary behavior. The nine possible response patterns were grouped into four mutually exclusive categories. Category 0 comprised three response patterns: (xv, ov), (xn, ov), and (xi, ov). In all three of these patterns, letter o is taken to represent voluntary behavior, when in fact it represents involuntary, compensatory responding to disturbance. Cutegory I comprised the response patterns (xi, oi), (M,oi), and (xi, on). Each of these response patterns acknowledges one of the letters as representing involuntary behavior, but acknowledges neither letter as representing voluntary behavior. Category X included response patterns (xv, oi) and (xv, on). In both response patterns, the term voluntary has been applied to letter x, and not to letter 0. In the (xv,oi) pattern, letter o has been recognized as involuntary responding to environmental disturbance; thus both letters have been correctly identified. In the (xv,on) pattern, o is not seen as involuntary behavior, but has not been confused with voluntary behavior. Category N consisted of response pattern (xn, on). Such a response pattern denies that either letter represents the experimenter’s behavior, voluntary or involuntary. For instance, the horizontal movements of both letters might be perceived as being determined exclusively by the computer program with the experimenter moving the mouse merely to match the movements of letter 0. In this case, the mouse movements would be seen as having no influence on the position of either letter, and, therefore, neither would represent what the experimenter was doing.

Results Of all the subjects in Experiment 1, 84% (n = 134), chose letter o as best representing the experimenter’s behavior. This total comprised 70 Condition C subjects, and 64 Condition D subjects. This preference for the letter o is significantly different from chance at the .01 level (binomial). The difference between Conditions C (88% choosing 0, n=70) and D (80% choosing 0,n=64) is not large enough to be statistically significant. The subjects’ response patterns on the questionnaire painted a similar picture. Of the total 160 subjects, 123 (77%) fell into Category 0, 21 (13%) fell into Category X, 13 (8%) fell into Category I, and 3 (2%) fell into Category N.

The Behavioral Illusion

377

We performed a multivariate analysis of all these data by crossclassifying subjects in terms of the letter they had chosen as best representing the experimenter’s behavior (x or 0) and the category into which their responses to the questionnaire fell (0, X, I, or N). A multivariate ChiSquare analysis of the resulting frequency data yielded the significant effect illustrated in Figure 2. Inspection of the figure reveals that a subject’s selection of letter x or o as best representing the experimenter’s behavior is significantly associated with both the questionnaire category a subject falls into and the experimental condition (C or D) a subject is in, ~ ~ ( = 3 20.17, ) p < .001. Of the subjects falling into Category 0 who also selected o as best representing the experimenter’s behavior, more were in Condition C (63, or 79%) than in Condition D (57, or 71%), whereas, of the subjects falling into Category X who selected x as best representing the experimenter’s behavior, more were in Condition D (13, or 16%) than in Condition C (6, or 8%). (For the remaining 6 cells of the cross-classification, the number of subjects from the two conditions, C and D, were relatively equal.)

80

70

Continuous Discontinuous

3 60

E

0

a, 9 50 CT)

E

40

L

a,

$

30

5! 20 10

0

00

lo Ix Xo Xx No Cross-Classification Category

Ox

Nx

Figure 2. The number of subjects in the two experimental conditions of Experiment 1, are shown as a function of the cross-classification category they fell into.

378

J. Scott Jordan and Wayne A. Hershberger

The behavioral illusion appears robust. The vast majority of the subjects overlooked the experimenter’s volitional action (controlling the position of the letter x) and considered the experimenter’s compensatory reactions, elicited by the computer-generated disturbances, to be voluntary behavior. However, the significant x2 indicates that the illusion is attenuated by the opportunity to observe alternate periods of volitional activity and inactivity. For example, of the subjects who identified their chosen letter as representing voluntary behavior, 91% in the continuous condition selected letter 0,whereas only 81% in the discontinuous condition did so. A second experiment was conducted to see whether active observers able to work at cross-purposes with the experimenter in the task might also be less subject to the illusion.

Experiment 2 Method Subjects. Subjects were 40 male and 40 female students who were enrolled in introductory psychology at Northern Illinois University. They were all volunteers and earned bonus course credit for participating. Apparatus. All was the same as in Experiment 1 except that in this experiment, subjects were given a joystick (Radio Shack TRS-80) with which they could influence the horizontal position of either letter x or 0,depending on a switch setting determined by the subject. The switch was rotary with three dial settings labelled x, 0,and n. If the setting was on n, the joystick had no effect on either letter. Procedure. Within sex, subjects were randomly assigned to one of two experimental conditions, Condition C or Condition D. All was the same as Experiment 1 except that subjects were run one at a time and could influence the position of either letter, whenever they chose to do so, by using the aforementioned joystick and selection switch. Subjects were seated to the right of the experimenter who was sitting in front of the computer screen. The experimenter then read the following passage to the subject:

I am about to command this computer to run a program. When I do, two letters, x and 0,will appear on the computer screen to your front. The control box to your left has a clearly labelled switch that determines which of the two cursors on the screen the joystick you are holding influences. Both letters can move from left to right only, so there is no need to move your joystick in a

The Behavioral Illusion

379

vertical or diagonal manner. Note that I have a computer mouse. Your task is to determine what I am doing, in other words, which letter on the screen best represents my behavior? Remember, at any time you can turn the selector switch in order to change which letter your joystick influences. The program will run for three minutes, and then you will tell me what I was doing, in other words, which of the letters best represents my behavior, the x or 0. The switch setting was started on n so that no bias would be introduced due to the initial setting of the switch. The subjects in the discontinuously engaged condition were told, just as in Experiment 1, that during the second minute of the experimental session the experimenter would in no way interact with the computer screen.

Results Of all 80 subjects in Experiment 2, 49 (61%) selected letter o as best representing the experimenter’s behavior. This percentage is significantly different from 50% at the .01 level (binomial). The subjects’ response patterns on the questionnaire painted a similar picture. Of all 80 subjects, 53 (66%) fell into Category 0, 25 (31%) fell into Category X, and 1 (1%) fell into each of the other two categories. In order to assess the effects of passive versus active subject participation, the data from Experiments 1 and 2 were combined, resulting in four between-subject cells (active versus passive subject participation crossed with continuous versus discontinuous experimenter participation), with the subjects in each cell cross-classified, as in Experiment 1, in terms of the letter (x or 0 ) selected as best representing the experimenter’s behavior, and the questionnaire category (0, I, X, or N) into which they fell. A multivariate ChiSquare analysis of the resulting frequency data revealed statistically significant effects of both independent variables on both dependent measures. -3 illustrates how one dependent variable, the letter chosen as best representing the experimenter’s behavior (x or o), varied as a function both of the subjects’ participation [active vs. passive: ~’(1)= 14.91,p , < .001] and of the experimenter’s participation [continuously vs. discontinuously active: X’(1) = 3 . 8 8 , ~c .05]. (The latter effect, although not significant in either experiment taken individually, becomes so when the data from the two experiments are combined, due to the increase in power resulting from the larger sample size.)

380

J. Scott Jordan and Wayne A. Hershberger

100 80

Subjects Selecting 0 Subjects Selecting x

a

3 0 a,

cn .c

60

0

E 40 a, 2 2 20 0

Passive Passive Continuous Discontinuous

Active Continuous

Active Discontinuous

Experimental Condition Figure 3. The subjects in each experimental condition shown as a function of the letter they chose as best representing the experimenter’s behavior. Figure 4 illustrates essentially how the two independent variables jointly effect the two dependent variables combined as a multivariate measure. That is, Figure 4 shows the percentage of subjects in each of the four treatment conditions who identified their chosen letter (x or 0) as being voluntary behavior. The effects of each independent variable are significant; for subject participation (active vs. passive), ~ ~ ( =3 8.98, ) p c .025; and for experimenter participation (continuous vs. discontinuous), ~ ~ ( =3 10.88, ) p c .005.

Discussion Figures 3 shows that each of the two participation factors reduces the number of subjects selecting letter o as best representing the experimenter’s behavior, and increases the number of subjects selecting letter x. The joint effects of the two independent variables are additive; there is no multiplicative interaction. Figure 4 shows that each of the two participation factors also reduces the percentage of those of subjects, who having chosen the letter 0,identify

381

The Behavioral Illusion

100 . u) I -

Subjects Selecting o Subjects Selecting x

80

0

.-a,

n 3 60

cn w-

0

a,

40

2 a,

a 20 n U

Passive Passive Continuous Discontinuous

Active Continuous

Active Discontinuous

Experimental Condition

Figure 4. The percentage of subjects who identified their chosen letter as representing voluntary behavior, shown as a function of experimental condition. that letter as representing volitional action, while at the same time, increases the percentage of subjects, who having chosen the letter x as best representing the experimenter’s behavior, identify that action as volitional. The joint effects of the two independent variables are additive; there is no interaction. Though each of the two independent variables attenuated the behavioral illusion, both factors taken together did not fully eradicate the illusion. Even in the Active Discontinuous Condition (the least illusory), a majority of the subjects selected letter o as best representing the experimenter’s behavior and a majority of them identified that behavior as voluntary. The behavioral illusion is not only a replicable phenomenon, it is remarkably robust. An analysis of the information available in all four experimental conditions suggests that a subject’s selection of letter o as best representing the experimenter’s behavior is directly dependent upon the behavioral illusion, that is, mistaking the experimenter’s arm movements (compensatory output) as the experimenter’s doing. That is, it was the letter on the screen that best

382

J. Scott Jordan and Wayne A. Hershberger

correlated with arm movements that appeared to be the basis of the subjects’ selection. In the Continuous Condition, the subjects saw letter x remain near the center of the screen, while the position of letter o varied greatly, correlating 1.0 with the position of the experimenter’s computer mouse. This positive Correlation between the positions of the mouse and the letter 0,coupled with an extremely small, if any, correlation between the mouse and the letter x, provided the information upon which subjects appear to have based their selection of either x or o as best representing the experimenter’s behavior. In the Discontinuous Condition, subjects experienced the same positive correlation between the movements of the letter o and the movement of the mouse, but only during the first and third minutes of observation. During the second minute the experimenter stopped controlling the position of x; thus, during the second minute, the letter o remained stationary (because the experimenter was no longer offsetting disturbances to the position of x) while the position of letter x varied considerably as a function of the random-number generator. This opportunity to observe a discontinuity in the experimenter’s control of the position of letter x allowed subjects in the discontinuous condition to observe a second correlation, this one being a negative correlation between the activity of the mouse and the letter x; that is, during the first and third minutes, letter x remained relatively passive near the center of the screen as the mouse was moved actively about; and, conversely, x moved actively about the screen during the second minute when the mouse was not moved at all. Apparently, this negative correlation between the movements of the mouse and letter x decreased the proportion of subjects choosing letter o as best representing the experimenter’s behavior. Nevertheless, the more salient positive correlation between the movements of the mouse and letter 0, appears to have determined the letter chosen by most subjects in the gr0UP. The letter chosen also depended upon the subject participation variable. In the passive condition, subjects simply observed the experimenter’s behavior. Thus the only information available to them about the experimenter’s behavior, were the two correlations described above. In the active condition, subjects were given a selection switch and a joystick with which they could alter the position of the letter of their choice, either x or 0,depending upon their switch setting. Whenever the subject selected the letter x, there was a negative correlation between the subject’s joystick movements and the experimenter’s mouse movements. Whenever the subject selected the letter o there was a zero correlation between the subject’s joystick movements and the experimenter’s mouse movements. It was observed during this active

The Behavioral Illusion

383

condition that subjects would accentuate the former correlation by moving their joystick rapidly and alternately, from the extreme left to the extreme right. When subjects were moving the letter x in such a manner the experimenter was forced, in order to keep letter x in the center of the screen, to move his computer mouse to the same degree, but in the opposite direction. When subjects were moving the letter o in such a manner, no correlation existed between the subject’s joystick movements and the experimenter’s mouse movements because the experimenter was not controlling the position of the letter 0. These two correlations made available to the subjects the fact that they could only influence the experimenter’s behavior if they moved the letter x. This information, available in the active but not the passive condition, appears to have made active subjects less susceptible to the illusion than passive subjects. During the second minute of the active discontinuous condition, just as in the passive discontinuous condition, the experimenter stopped controlling x. It was observed during this second minute that subjects would, at first, move their joystick rapidly and alternately from the extreme left to the extreme right. Since the experimenter was not involved in the control task, the movements of both letters correlated positively with the subject’s joystick movements, the letter o correlating 1.0, and the letter x somewhat less than 1.0 due to the random number generator. Subjects would then stop moving the joystick until the experimenter was once again engaged in the control task. Although the active subjects in the discontinuous condition could see that they exercised greater influence over the position of letter x when the experimenter was inactive, this difference was a matter of degree rather than kind; that is, this observation did not provide a new kind of information unavailable to the active subjects in the continuous condition. Essentially all the new information available to the active subjects was provided in full by their opportunity to work at cross-purposes with the experimenter while he was continuously engaged in the control task. This is indicated by the lack of any significant interaction effect between the subjectand experimenter- participation variables. It could be argued that the subjects were never wrong in selecting either x or o as best representing the experimenter’s behavior inasmuch as compensatory output is a form of behavior; and, since the letter o correlated best with the experimenter’s compensatory output (i.e., hand movements), they were correct to select o as the letter best representing the experimenter’s behavior. However, such an argument overlooks the fact that the question posed for the subjects was “What is the experimenter doing?,” not “What is the experimenter’s hand doing?” More exactly, what is at issue here

384

J. Scott Jordan and Wayne A. Hershberger

is the notion of intentionality or the difference between voluntary and involuntary actions and the question of which type better represents a pe$orrrzer’s behavior. Such questions are readily resolved by a consideration of the subjects’ own judgments, reflected in their responses to the questionnaire, as to the volitional nature of the behavior represented by each letter on the screen. Since 203 (85%) of all subjects identified their chosen letter as representing voluntary behavior, as opposed to involuntary behavior, it is apparent that subjects in general not only accept the notion of intentionality, but also assume that behavior which is voluntary, better represents what the experimenter is doing. Subjects did not see letters x and o as equally representing the experimenter’s behavior; rather, they saw one of the letters in some way being a better representative of what the experimenter was doing, and that letter tended to be the letter they identified as representing voluntary behavior. If one considers only those 203 subjects who identified their chosen letter as representing voluntary behavior, the overall picture, illustrated in Figure 4, does not markedly change from that represented in Figure 3 which includes all 240 subjects. What is interesting is that Figure 3 refers to a lower-order, univariate relationship (that is, it only illustrates the effects of the two independent variables on the letter chosen as best representing the experimenter’s behavior), whereas Figure 4 refers to a higher-order, multivariate relationship (that is, it illustrates the effects of the two independent variables on the relationship between the letter chosen as best representing the experimenter’s behavior and the letter chosen as representing voluntary behavior). This further supports the notion that subjects assume that if one letter better represents the experimenter’s behavior, it must be the letter that represents voluntary behavior. Thus, the behavioral illusion comprises a twofold error: that which is voluntary is overlooked, whereas that which is involuntary is seen as voluntary. The present results and theoretical analysis should be of interest to social psychologists concerned with the theory of attribution developed by Fritz Heider and others (Heider, 1944; Kelley, 1973; Kelley & Stahelski, 1970). These theorists are concerned with the layman’s perceptions of intentionality in the actions of others. However, attribution theorists have carefully skirted the scientific question of man’s volitional action (i.e., what is its nature?) by addressing the question of the layman’s beliefs (superstitions?) about volition, whatever “volition’l might be, including, perhaps, nothing at all. Hence, attribution theory does not address the question central to the present research, which was the veridicality of such perceptions or beliefs.

The Behavioral Illusion

385

Summary and Conclusions A behavioral illusion (Powers, 1987) was investigated. A totaI of 240 college students in 4 experimental conditions watched the experimenter control the position of a letter (x) on a computer screen by using a mouse to offset computer generated disturbances. Another letter (0)reflected mouse movements. Subjects watching the experimenter continuously control the position of x tended to perceive the position of o rather than x as being the experimenter’s doing. However, the incidence of this misperception was reduced when subjects viewed discontinuities in the experimenter’s conduct, or worked at cross-purpose with the experimenter each of which provided a glimpse of the operator’s intentions in the form of visible correlations; for example, subjects were able to disturb the operator only when they tried to influence x. When an individual controls the value of an environmental variable by offsetting would be disturbances with compensatory output, the compensatory output, rather than the controlled variable, appears to constitute the individual’s behavior, even though that output is driven entirely by the disturbance. Conversely, the controlled variable goes unrecognized as the individual’s doing. This chapter was based upon a master’s thesis (Jordan, 1988), portions of which have previously been presented as a convention paper (Jordan & Hershberger, 1988).

References Heider, F. (1944). Social perception and phenomenal causality. Psychological Review, 51, 358-374. Kelley, H. H. (1973). The process of causal attribution. American Psychologist, 28, 107-127. Jordan, S . J. The behavioral illusion: The misperception of voluntary behavior. Unpublished master’s thesis. Northern Illinois University, DeKalb, IL. Jordan, S. J., & Hershberger, W. A. (April, 1988). The misperception of voluntary behavior. Paper presented at the Sixteenth Annual Meeting of the Midwestern Psychological Association, Chicago, Illinois. Kelley. H. H., & Stahelski, A. J. (1970). From moves in the prisoner’s dilemma game. Journal of Experimental Social Psychology, 6, 401-419.

386

J. Scott Jordan and Wayne A. Hershberger

Powers, W. T. (1978). A quantitative analysis of purposive systems: Some spadework at the foundations of scientific psychology. Psychological Review, 85, 417-435. Skinner, B. F. (1971). Beyond freedom and dignity. New York: Bantadvintage Books. Skinner, B. F. (1987). What ever happened to psychology as the science of behavior? American Psychologist, 42, 780-786.

VOLITIONAL ACTION, W.A. Hersh berger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

387

CHAPTER 17 VOLITION AND SELF-REGULATION: MEMORY MECHANISMS MEDIATING THE MAINTENANCE OF INTENTIONS Julius Kuhl and Miguel Kazen-Saad In this chapter we first discuss a theory specifymg the mechanisms underlying the maintenance of intentional representations in memory and then summarize several recent experiments from our laboratory dealing with this topic. Volition (or self-regulation) is here defined as a mechanism that supports the maintenance of information related to the current intention and resolves conflicts between cognitive and motivational preference hierarchies. This maintenance function protects the current intention (i.e., a cognitive preference) against competing action tendencies supported by tempting emotional preferences (Kuhl & Kazen-Saad, 1988).

The Process of Action Control In their natural environments organisms are usually confronted with the problem of having to process an excessive amount of information and of simultaneously satisfying multiple needs or goals. How do they manage to accomplish that? An important theoretical question concerns the mechanisms mediating the handling of multiple goals, specifically, the problem of selecting and maintaining one goal at a time and avoiding premature behavioral change. Although it is intuitively obvious that for different organisms different control mechanisms may serve a similar purpose, it is not as clear what these mechanisms are nor how we can distinguish them from one another. We would like to propose three basic mechanisms, differing in their level of complexity and flexibility, as being responsible for the management of information overload, multiple-goal handling, and goal maintenance: lateral inhibition, attentional selectivity and orienting, and volitional control. We also propose the following criteria to distinguish among these mechanisms: (1) the eliciting condition, (2) the direction of control, and (3) the complexity of control (this last criterion can be conceived of as the number of operational units involved in its

388

Julius Kuhl and Miguel Kazen-Saad

functioning). In this section we discuss the three maintenance mechanisms in terms of the above criteria. The first, most basic, control mechanism is lateral inhibition (cf. Dorner, Schaub, Staudel, & Strohschneider, 1987). It is a simple mechanism (low complexity) triggered by an external stimulus source (eliciting condition); it proceeds in a bottom-up way (direction of control); and, once it is active, competing stimuli have a lower probability of gaining control over the system.' At the output level it functions according to a simple contention-scheduling principle; that is, the action schema having the strongest activation will inhibit the execution of competing action schemas (cf. Norman & Shallice, 1985). The second, more complex, maintenance mechanism is attentional selectivity and orienting (cf. Neumann, 1987, and Posner, 1978, 1980), and represents the transition between bottom-up and top-down processes. This mechanism is normally triggered by an informational aspect of an external stimulus but can also be guided internally. Attentional selectivity is usually elicited by features of the stimulus (such as its novelty, surprisingness, or complexity, cf. Berlyne, 1960) that have been preselected by an active top-down process (also called set, expectancy, hypothesis, or orienting). This internal orienting mechanism can be directed both to simple visual or auditory signals or towards the physical, phonetic, and semantic codes of letters and words and its effects are manifested by benefits or costs associated with expected or unexpected stimuli, respectively (Posner, 1978, Chap. 7). Since the eliciting conditions triggering attentional selectivity and orienting include the informational contents of the stimulus, this mechanism is much more "intelligent" and flexible than simple lateral inhibition, which is blind to those stimulus features. The third and most sophisticated mechanism mediating the maintenance of active goals is volitional control, which according to our conception is found only in human beings (Kuhl, in press) and shows a clear developmental trend (Kuhl & Kraska, in press, Mischel & Mischel, 1983). Since one of the goals of the present chapter is to elaborate and expand upon this mechanism, it will be described in greater detail than the previous ones. The eliciting condition for volitional control is characterized by the mode of representation of the intention and a

' Potzl, in 1917 (see Erdelyi,

1984), described a patient with an occipitallobe injury in whom this lateral inhibition mechanism was impaired. In contrast to normal people, this patient was sometimes able to see his eye's blind spot and had double images (physiological diplopia), causing him great distress.

Volition and Self-Regulation

389

specific type of conflict. The mode of representation which characterizes volitional maintenance is based on a "self-model." This model is capable of simulating future states of the organism. For our purposes, the self model's ability to anticipate future emotional consequences of "simulated'' action alternatives is of particular importance: The functional significance of a current action alternative based on an earlier decision process involving this anticipatory self-simuiation is different from the functional significance of action alternatives based on currently active emotional processes. The former can be represented on a cognitive or (more precisely) symbolic level even if the emotional consequences that had been anticipated during the decision process are not activated on a later occasion. In this case, execution of the cognitively intended action is more difficult than performance of the action supported by currently active emotions (since emotions have a stronger impact on the control of actions than cognition, see Frijda, 1986). According to our view, this particular conflict between a cognitive and an emotional preference can not be resolved by automatic attentional control. Maintenance of the cognitive preference (i.e, the intention) requires a particular mode of processing which is characterized by a sequential mode of control (as opposed to parallel), and whose function is to increase the strength of the intended action schema and to decrease the strength of competing action schemas. The direction of control is top-down, and sometimes contextfree (as opposed to lateral inhibition and attentional selectivity, which are highly context-dependent and are affected by continuous impact of environmental stimuli, see Toda, 1982, Chap. 10). Volitional maintenance produces an even higher uncoupling of maintenance functions from situational change than automatic attentional processes do, as will be explained in later sections.

Three Meanings of Volition2 We will present a process model of volitional control and then describe recent experimental findings from our laboratory on volitional processes. Before doing that, however, we would like to spend a little time on the definition of basic terms. The concept of volition has been traditionally defined in at least three different ways. Its first meaning refers to the currently "intended" or willed action, that is, to the currentZy This section is a slightly modified version of the model as described in Kuhl and Kraska (in press).

390

Julius Kuhl and Miguel Kazen-Saad

dominant action schema. According to this definition, terms like "volitional efficiency" or "weakness of the will" are meaningless because by definition the organism always performs the volitionally "intended" action (provided it has the necessary skills and opportunity to do so). We do not find this concept very useful because it does not add anything to the explanation of observed behavior and it excludes by definition rather than by experimental evidence the concept of volitional control (self-regulation). The second meaning of volition refers to what we call cognitive preference, that is, a deliberate (and frequently conscious) intention based on a cognitive representation of an action plan (cf. the concept of volition in Neumann & Prinz, 1987). In our terminology we have reserved the term intention for this meaning of the term volition. Note that we distinguish between an active (dominant) action schema and an intention. The latter term is reserved for the case in which the individual has formed a symbolic representation of the dominant schema. In the literature, the term intention is used in both ways (sometimes even within the same article), that is, referring to the symbolic representation of the dominant action schema or to the action schema itself. One may reconcile the two meanings by calling the dominant action schema (which guides ongoing behavior) the implicit intention and its symbolic representation the explicit intention (cf. the distinction between implicit and explicit memories, Schacter, 1987). An intention in this sense is not to be identified with volitional control (although it may be considered a prerequisite for it). We use the term volition in its third meaning: a set of mechanism mediating the maintenance of an "explicit intention", especially when the latter is incongruent with the currently dominant action schema, that is the "implicit intention." These mechanisms (to be explained later) include: (1) freezing, (2) the generation of a superordinate maintenance goal (commitment) which may be mediated by similar processes as the generation of any other goal, and (3) volitional strategies that directly or indirectly modify the strengths of action schemas until they are congruent with the current intention. Cognitive (i.e., intentions), emotional (e.g., needs), and executional (e.g., habits) preferences can be incongruent among one another despite their close interactions because each of them depends, at least partially, on a unique set of determinants (cognitive anticipation of future consequences, past emotional experiences, and habit formation, respectively). Although volitional control may involve the use of fundamental computational mechanisms such as selective attention, strategic processing, and/or consciousness, it is not identical with any of

Volition and Self-Regulation

391

them. Volitional control requires the development of a particular set of metagoah specialized on the detection and resolution of conflicts between cognitive and emotional or executionab'preferences. One type of conflict requiring the use of volitional control will be discussed now. Figure 1 illustrates the three subsystems that constitute the efferent or "motivationaF' part of the cognitive, emotional, and executional (action) systems with a typical volitional problem: a conflict between a cognitive preference (to work) and an executional preference (to play). The interactions within and between the three subsystems are described by single arrows indicating effects of normal strength or double arrows indicating strong effects. It is assumed that emotional processes have a greater impact on the action system than cognitive processes do (see impulsivity in Figure 1). The relative activational strength of the information concerning the two action alternatives in our example (i.e., work and play) are indicated by small and big circles, respectively, within each subsystem. Figure 1 illustrates the case in which an individual has a strong emotional preference for playing. In other words, on an emotional level, past experiences with playing are activated that are more pleasant than those associated with working. This emotional preference for playing is entirely congruent with an executional preference for playing. That is, the action schema for playing is more strongly activated than the one for working. Executional preference is defined in terms of the executional strength of competing action schemas (Norman & Shallice, 1985) encoding procedural knowledge (Tulving, 1985). An executional preference for playing may be a combined result of the strong impact of the emotional preference for playing on the executional system and of the acquisition of a strong habit to play in the situation at hand. In general, habit formation directly affects the relative strengths of action schemas whereas emotional preferences are affected by the quality and intensity of past emotional consequences of performing the activity in question. During early childhood, congruence between emotional and executional preferences is the rule rather than the exception. Later in childhood, children begin to develop the ability to generate cognitive representations of their own intentional states. The development of cognitive representations of one's intentions is the basis for new abilities but also for new problems. One new ability is the self-regulatory maintenance of a cognitive preference (i.e., the cognitive representation of an intention) even when it is incongruent with emotional and executional preferences. This case is illustrated in Figure 1: The person

392

Julius kkhl and Miguel k e n - S a a d

INPUT PROCESSING ("Cognition"1

HETAPROCESSING Volition \

OUTPUT PROCESSING I Hotivation I

ICcmmitrnent "

I

MOTOR SYSTEM

I

Figure 1. A model of volition (see explanation in text). has a cognitive preference for working whereas her/his emotional and executional preferences favors playing. The cognitive preference for working may be based on a past decision to finish homework before going out to play with friends. Note that at the time when this decision was made, cognitive and emotional preferences may have been congruent. That is, when considering the various consequences of playing vs. working, she/he formed the intention to work and felt emotionally better about it than about playing. Later, while working, the incongruence between cognitive and emotional preferences depicted in Figure 1 may have developed as a result of unpleasant experiences associated with working while the cognitive preference remained unchanged or was less affected than the emotional preference.

Volition and Self-Regulation

393

Commitment can be defined in terms of a superordinate goal directed at the maintenance of a decision (see Figure 1). Without volitional intervention, behavior consistent with the commitment is unlikely to "win the competition" against action alternatives supported by currently active emotional states. The above prediction can be derived from the impulsivity assumption already described. According to the model, volitional support for the maintenance of the intention (i.e., the cognitive preference for working) is initiated when a comparison between the three subsystems results in a discrepancy between cognitive and emotional or between cognitive and executional preferences (see Figure 1: CEI = Cognition-Execution Incongruence). Although young children may learn to activate volitional support €or an intention if any degree of cognitive-emotional or cognitive-executional incongruence is encountered, a more parsimonious strategy may be developed in later years. This strategy is based on a finer-grained appraisal of the degree of difficulty of enactment of carrying out an intention. Specifically, difficulty of enactment (d,) can be estimated in terms of the relative strength (T,) of the action schema that corresponds to the cognitive preference compared to the sum of the strengths of all competing action schemas (Ti), that is: d, = 1 - (T, / ZTJ. At this later stage of development, the volitional maintenance of the current intention is activated on& if the difficulty of enactment exceeds a critical value (see Figure 1). When the critical value (d,) is exceeded, the individual may recall the commitment and generate a "maintenance goal" to activate volitional strategies which support the maintenance of the cognitive preference. When an individual emotionally prefers playing to working and the action schema for playing is dominant, a cognitive preference for working would have no chance to be enacted unless he/she has the capacity to inhibit the dominant action schema from being carried out. Since the operation of volitional strategies requires time, the executional system has to be "frozen" until the action schema corresponding to the current intention (to work) has been made dominant. Freezing is the first step of a volitional intervention cycle. It creates a "time window" for the operation of volitional strategies. The facilitating and inhibiting effects of these strategies on the activational strengths of cognitive, emotional, and executional preferences is illustrated in Figure 1. The strategy of attention control (Kuhl, 1983, 1985) enhances the cognitive preference for working and inhibits the cognitive representation of playing by channeling attentional resources

394

Julius Kuhl and Miguel Xazen-Saad

accordingly (see Norman & Shallice, 1985, for a theory of volition based on a similar mechanism). The second strategy illustrates motivation control which enhances the activation of the emotional preference for working and reduces the emotional attractiveness of playing. This can be achieved by focusing on positive incentives, such as being able to maintain an intention in general or being able to enjoy the good feeling of having finished one's work. Beckmann and Kuhl (1984) obtained experimental evidence concerning the facilitation of decision-making and enactment through the use of a motivation control strategy. Both attention control and motivation control increase the probability that the individual would eventually succeed in performing the cognitively preferred (i.e., the intended) action despite the initial emotional and executional preference for an alternative action. The third strategy, execution control, directly affects the strength of the action schema corresponding to the current intention either through an automatic facilitation process or through complex strategies such as the deliberate development of a habit to perform the intended action in a specified situation at a particular time (e.g., doing homework right after lunch). Execution control is similar to Norman and Shallice's (1985) "Supervisory Attentional System." In contrast to their model, however, ours does not equate volition with selective attention (although the former includes the latter). Aspects uniquely associated with volition are: (1) anticipation of future emotional consequences of various action alternatives, (2) the symbolic representation of an intention (which is not necessarily congruent with emotional preferences or with the dominant action schema), and (3) sequential control of volitional mechanisms that support the current intention. VoZitional eficciency, that is, the probability that the intended action (e.g., working) is initiated and maintained until goal attainment, is a function of the relative strength of: (1) the initial strength of the incongruent emotional and executional preference (for playing), (2) the relative strength of the freezing mechanism (determining the length of the time window for volitional intervention), and (3) the impact volitional strategies have on the three subsystems involved. We are currently running computer simulations of the model to develop a more precise way of describing its implications.

The Incomplete-Intention Paradigm Historically, motivational psychologists have been more interested than cognitive psychologists in the study of individual and personality

Volition and Self-Regulation

395

differences mediating the maintenance of intentions. Traditional motivation theory starts with the assumption that motivational states persist in memory. Freud postulated the persistence of unfulfilled wishes, even in spite of conscious attempts to repress them. Lewin (1935) hypothesized the existence of tension systems underlying unsatisfied needs and conceptualized intentions as "quasi-needs." Zeigarnik (1927), working within Lewin's framework, developed an experimental paradigm to investigate the superiority of memory for interrupted over completed activities, which she demonstrated. These studies, and others related to interrupted tasks, were later criticized on methodological grounds (e.g., Butterfield, 1964; van Bergen, 1968). Working within a cognitive perspective, Anderson (1983) has alluded to the special status of the memory for intentions or goals using the metaphor of source nodes, which provide the energy for maintaining an intention active in memory from "within" rather than from external sources. We believe that his metaphor is more descriptive than explanatory. From our own perspective, volition research should be looking for explanatory mechanisms mediating this phenomenon. Until now there is no experimental paradigm available that would unequivocally establish the postulated special status of memory structures related to incomplete intentions. Our recent work is based on an attempt to overcome the limitations of previous investigations by applying techniques used in current cognitive research. Our main goal is to investigate in greater detail the mechanism mediating self-regulatory control, not just to describe its occurrence. The first step towards achieving this aim requires the development of experimental paradigms which enable us to measure the degree of maintenance of intentions. In our first experiment (Goschke & Kuhl, 1988), 28 subjects memorized a pair of scripts describing simple actions (set a dinner table, clean up a messy desk, empty a garbage can, dress up for leaving the house). One of the scripts described actions subjects later had to either execute themselves or observe the experimenter carry out (prospective script), whereas the other script was to be neither executed nor observed and served as a control (neutral script). Once subjects had memorized the steps of the two scripts, they were requested to count aloud backwards for a while, starting from a given number, before being told which of the two scripts was the one they had to execute (or observe) and which was not. Immediately thereafter, a word recognition test started and the subjects' task was to decide as quickly and accurately as possible as each of the words was presented whether it had appeared before in either of the two previously memorized scripts. The same

396

Julius Kuhl and Miguel k e n - S a a d

procedure was repeated with a second pair of scripts to allow for withinsubject counterbalancing of the execution and observation conditions. The repeated presentation of an item on a recognition test results in a shortened recognition reaction time (RT). This facilitation will decrease as the lag between the first and second presentations of the item increases. As a way of testing the special status in memory of intentionrelated words, we repeated some of the items on the recognition list with a lag of one or five words between presentations. Confirming our hypothesis, results indicated that in the observation condition both prospective and neutral words showed a big RT facilitation at lag one, which decreased at lag five. The results of the execution condition showed exactly the same pattern for the neutral words as the one found above, but for the prospective words there was an equal amount of facilitation in RT, independently of lag size. The pattern of results found in this experiment was surprisingly well replicated in a second similar experiment. These findings confirm the assumption that, once activated by a cue, information related to an incomplete intention is maintained longer in memory than neutral information. In addition to assessing the duration of the maintenance of intention-related information, we were also interested in its spontaneous availability. Spontaneous availability can be considered context-inadequate in our experimental situation, because subjects could not perform the postponed intention during the recognition phase. Memory availability of information of context-inadequate intentions may be an important phenomenon. It may help us distinguish between the two last levels of maintenance discussed in the introduction of this chapter. In contrast to attentional maintenance, volitional maintenance may perseverate even when the conditions for executing an intention are not longer present. This hypothesis is based on the idea that metacognitive representations (here: the symbolic representation of the current intention) and sequential processing increment the uncoupling between the current situation and overt behavior (Leslie, 1987; Smolensky, 1988). Recognition performance was used as an index of the spontaneous availability of concepts related to prospective or neutral scripts. In the observation condition there were no significant RT differences in recognizing words belonging to the prospective and neutral scripts, whereas in the execution condition the words belonging to the prospective script were responded to significantly faster than those of the neutral script. Exactly the same pattern as above was obtained when we analyzed

Volition and Self-Regulation

397

recognition errors for words of the prospective script. These errors tended to be lower than those of the neutral script, but only in the execution condition. Although the above findings may suggest that intention-related words are more accessible in memory for all people, it should be mentioned that most of the variance is accounted for by an individual difference variable related to volitional control. Kuhl's (1983) action control theory postulates that in state-oriented people, symbolic representations of uncompleted intentions seem to perseverate producing thoughts (or ruminations) about past, present, or future states related to the uncompleted intentions. On the other hand, action-oriented individuals focus their attention on a currently active action plan, which is represented in a "procedural" rather than a symbolic code (cf. Tulving, 1985). According to a recent elaboration of the theory (Kuhl, 1987; Kuhl & Beckmann, in press), action-oriented individuals do not differ from stateoriented ones in their degree of self-regulatory efficiency (i.e., maintenance of intentions) but rather in the mode of control which mediates that maintenance. Whereas state-oriented people employ volitional maintenance mediated by a symbolic representation of their intentions (i.e., declarative knowledge structures), sometimes even in intention-irrelevant situations, action-oriented individuals seem to maintain a code of the action component of the intention (i.e., procedural knowledge structures). Since recognition data seems to index declarative rather than procedural memory (Tulving, 1985), the above hypothesis predicts better recognition performance for words referring to the semantic representation of an intention in state-oriented people compared to action-oriented ones whenever the intention has to be postponed, as in the case of the present experiment. Figure 2 shows that this was indeed the case in the RT for intention-related words under the execution condition. A very similar interaction was found for recognition accuracy in terms of error rates and also when we analyzed the data using a signal-detection measure of accuracy (d'). Summarizing this experiment we can say that when an intention is activated, information related to the to-be-executed intention is maintained longer than neutral information (cf. Bock & Klinger, 1986). When the execution of the intention has to be postponed, state-oriented subjects seem to keep the symbolic representation of the postponed intention more active than action-oriented ones, even during a period in which they can not carry it out. The results just described are consistent with our hypothesis that state orientation is characterized by the overmaintenance

398

I100 1060 1000

Julius Kuhl and Miguel h e n - S a a d

REACTION TIME (mnrcl PrDPDeCtir. Word

Neutral W4M

1030

88% I

960

aoo 050 000

760

ACTION [Obr BTATE [ O h ) ACTION [Exec] 8TATE (Exec] PER$ONALITY AND EXPERlMENTAL CONDlTlON

Figure 2. Mean RT for correctly recognized words from the prospective and neutral scripts as a function of personality and experimental condition of the symbolic representation of context-inadequate intentions. This interpretation is consistent with the results of other experiments (Kuhl, 1983, p. 268; Steinsmeyer-Pelster & Schurmann, in press) in which actionoriented subjects showed better availability of intention-related information than state-oriented ones during a period in which the intention could be carried out. The experiment to be described next was designed to test the model’s assumptions (see Figure 1) concerning the degree of intentional maintenance needed when subjects are exposed to several levels of difficulty of enactment .

The Difficulty of Enactment Paradigm According to our theory (Kuhl & Kazen-Saad, 1988), the maintenance of an intention should be a function of the person’s commitment to the intention; that is, protection of an intention should occur only if

Volition and Self-Regulation

399

she or he is committed to perform it. In the above experiments we did not measure explicitly the degree to which subjects were committed to perform the induced intention because the experimental situation was set up in a way that maximized the probability that they would commit themselves to perform the required actions. In the present experiment we assessed subjects' commitment through a credibility measure. Subjects (N=24) were told they would have a conversation about "intimate matters'' with a person of the opposite sex after the first phase of the experiment was completed. Their first task was to rate a total of 100 trait adjectives on how much they would like to have a conversation with a person showing that trait, e.g., tolerant or critical, on a scale from 1 (not at all) to 7 (very much so). Once subjects had rated the adjectives, they were presented with a summary description of the personality of three possible conversation candidates, each consisting of the person's name and four personality traits (e.g., Susanne W.: sincere, inquisitive, alert, emotional). They were told the conversation partners' personalities had been previously assessed by standard personality tests and the computer had presented the four most typical traits, while in fact the traits were chosen according to ratings made by the subjects themselves. Two of the three candidates were assigned adjectives rated below-average in conversation attractiveness and the other one was assigned traits rated above-average on that dimension by the subject. At this point, subjects were asked to make two decisions: to choose the person they wanted to have the conversation with (the chosen partner) and to choose the person that would be the worst possible for that purpose (the rejected candidate). As could be expected, in every case subjects chose the candidate with the above-average rated adjectives, rejecting each of the other candidates about equally often. In the following and most important part of the experiment we pursued three goals: First, to investigate the temporal course of the "shielding" process. Second, to test the hypothesis that a shielding process is summoned up only under conditions in which a personal commitment has been made concerning the intention-relevant decision. Our third goal was to attempt a solution to a difficult methodological problem so far ignored in the history of the psychology of will (cf. Kuhl, 1983, Chap. 8): the confounding between two opposing effects of increasing the difficulty of enactment of maintaining an intention. On the one hand, the maintenance of the intention is rendered more difficult due to the formation of competing (alternative) action tendencies, which lowers the probability that the intention will be maintained. On the other hand, we can expect

400

Julius Kuhl and Miguel Kazen-Saad

that increasing the difficulty of enactment will result in the use of selfregulatory processes related to the intention, which will increase the probability that the intention be maintained. As long as we are unable to measure separately the contributions of the above mentioned factors, it will be impossible to make unambiguous predictions concerning the influence that an increment in the difficulty of enactment of an intention will have upon the self-regulatory efficiency of a person. In the present experiment we attempted to manipulate the difficulty of enactment of the intention through the presentation of additional information about the rejected candidate. The success in shielding the original decision was measured through the memory availability (efficient retrieval) of information associated with the chosen and the rejected conversation candidates. Each trial of this phase of the experiment consisted of two parts. In the first part we manipulated the difficulty of enactment: We presented on the computer screen new information (personality traits) about the rejected candidate which was either congruent (low difficulty of enactment) or incongruent (high difficulty of enactment) with the original unattractive description associated with this person. The rationale for the above manipulation is the assumption that volitional shielding processes will be activated if the subject is confronted with information challenging herhis original decision. The measurement of the shielding was done in the second part of each trial by assessing subjects' recognition of intention-relevant information. In this second part of each trial subjects were shown an adjective describing a trait and had to decide whether or not the adjective shown was one previously associated with one of the conversation candidates, either chosen or rejected. We varied the stimulus onset asynchrony (SOA) between the presentation of additional (congruent or incongruent) information about the rejected candidate (e.g., "Birgit K. is intolerant.") and sentences referring to the original descriptions of the chosen and rejected conversation partners (e.g., "Is Susanne W. sincere?"). Subjects had to verify this description as correct or not. We expected that exposure to intentionincongruent information about the rejected candidate would interfere with the retrieval of the old information concerning the chosen individual at short intervals. At longer intervals, with Ach's (1910) "difficulty law of motivation," we expected committed subjects not only to overcome this debilitating effect but even protect their intention by increasing the activational level of information related to their intention (i.e., to meet the chosen individual). In sum, the procedure was to present one additional trait describing the rejected candidate before showing a test sentence refer-

Volition and Selj-Regulation

401

ring to either the chosen partner or the rejected candidate. Half the time the additional trait of the rejected candidate was taken from the belowaverage pool of rated adjectives (congruent information) and the other half from the above-average pool (incongruent information). An example of an incongruent trial is (If Birgit K. was rejected and Susanne W. chosen): First Screen: "Birgit K. is" Second Screen: "Intelligent" - - - SOA - - Third Screen: "Is Susanne W. Tolerant?" The SOAs between the presentation of the additional trait (for the above example, intelligent) and the to-be-verified sentence ("Is Susanne W. tolerant?") were either 400, 800, or 1600 msec. As shown in Figure 3, "believers" (i.e., people who believed that the conversation candidates actually had the traits we displayed and that they would meet them afterwards during the experiment) showed interference in retrieval time at 400 msec of SOA and facilitation at 1600 msec of SOA after they were presented with incongruent information (p ' < .05), but showed no significant change after congruent information, which agrees with our hypothesis. Additional findings showed that nonetheless their recognition performance at every SOA was equally good after being presented with either congruent or incongruent information. Although the above pattern of results corresponds very closely to the predictions derived from the theory of action control, it would be possible to attribute these findings to the effects of nonvolitional processes, defined post-hoc. It is not easy, however, to find alternative explanations which account for the complete pattern of results. Especially difficult for nonvolitional alternative explanations would be to explain the fact that the shielding effect was obtained only with believers (high personal commitment) and not with nonbelievers (low personal commitment). By way of summary, we can say that subjects who had made a commitment to carry out a prospective intention and were exposed to information that challenged it, showed an interference in retrieving intention-related information at short SOA intervals, but, at longer SOAs, could overcome this interference by strengthening the memory availability of intention-related information. The requisite for this effect is, of course, previous commitment to the intention, which was not the case for nonbelievers. These findings confirm our assumption that information related to an uncompleted intention is maintained in memory, and they convey additional information concerning the memory mechanisms

402

Julius a h 1 and Miguel k e n - S a a d

14aQ

-

la60

X

lgOQ -

\ \

1260 -

1200

-

1150 -

11mI

I

ma

I

I

BM) 1600 BTIMULUB ONBET MYNCHRONY Imsbcl

Figure 3. Mean RT taken by "believers" to verify sentences referring to the chosen or rejected conversation candidates as a function of SOA and type of additional information presented. mediating this effect. Apparently, maintenance is especially enhanced when the enactment of an intention is challenged (i.e., in the incongruent condition) and it is mediated by an active process which requires a substantial amount of time (more than 800 msec).

The Overcommitment Hypothesis State orientation can be construed as overcommitment, that is, as the inability to disengage from context-inadequate intentions when performing a particular task. Action orientation, in contrast, implies this ability. State orientation increases the risk of developing a depressive

Volition and Self-Regulation

403

disorder since the overmaintenance of intentions increases the number of perseverating thoughts intruding into working memory which in turn interfere with the enactment of context-adequate intentions (Kuhl & Helle, 1986, in press; Kazen-Saad & Kuhl, in press). If our analysis is correct, we should expect that state-oriented people show a tendency to encode action related information in a highly committing. format. Specifically, we would predict that state-oriented people are more likely than action-oriented individuals to encode optional instructions(e.g., "if you have time, you can give me a call") as obligatory commitments or intentions ("I must call this person"). The following sentence recognition experiment (N=48) was designed to test this hypothesis. As a cover story, subjects were asked to imagine they worked in an office and had a child who celebrated herhis birthday on the following day and since they had to carry out their normal secretarial activities and prepare their child's birthday party at the same time they had to remember many different things to do, differing in importance. Subjects were told we were interested in studying how well they would recognize sentences describing the needed tasks. In the first phase of the experiment subjects learned a series of 32 sentences describing the activities required to do the above jobs, 16 of them started with "I must ..." (e.g., "I must buy a birthday cake") and the other 16 with "I can ...I' (e.g., "I can paint the balloons with stripes"). On a subsequent recognition test, some of the sentences remained unchanged, while others had a commitment change, either from optional activities ("I can...") to obligatory ones ("I must...), or vice versa. As shown in Figure 4, we can observe that action-oriented subjects show a significantly better performance in rejecting false commitments or obligations (i.e., changes from "I can..." to "I must...") than state-oriented ones (p < .Ol). For the opposite change (i.e., from previous commitments to optional activities) recognition performance was identical for all subjects. The recognition results were mirrored by the RT data where we observed that action-oriented subjects showed a tendency to reject false commitments faster than state-oriented subjects. The above results confirm our prediction concerning the lack of discrimination by state-oriented people between optional activities and obligatory commitments. It is also worth mentioning that under normal conditions action-oriented people are very fast and accurate in rejecting false commitments, which can be interpreted as an useful strategy they have to protect themselves from the accumulation of an excessive number of unfulfilled or degenerated intentions.

404

Julius Kuhl and Miguel k e n - S a a d

A PRIME k t i m Orirntod

I

-

.-

474

h b Oriented

-

P,74

Opt tonal: 'I CAN'

Obllfirtory: 'I MUST'

8ENTENCETYPE

(Random Performance 0,SI

Figure 4. Discrimination performance in sentence recognition as a function of personality and type of change made in level of commitment (from previous "I must..." to Optional: "I CAN," and from previous "I can..." to Obligatory: "I MUST'). (A PRIME is a nonparametric analog to d', the discrimination measure of signal detection theory).

Concluding Remarks We would like to end this chapter going back to the initial discussion concerning the three levels of control mediating the handling and maintenance of multiple goals. Although lateral inhibition and selective attention are fundamental control mechanisms, we believe that

Volition and Self-Regulation

405

the most important, and to date the least researched one, is volitional control (or self-regulation). Our theory proposes a process model (see Figure 1) specifymg the eliciting conditions and control mechanisms required for volitional functioning. The results of the experiments reported here are encouraging and represent our initial steps to test its validity. It is our belief that we need to take into account the role of individual differences (action and state orientations) together with cognitive mechanisms in order to fully understand volitional action.

References Ach, N. (1910). h e r den Willensakf und das Temperament. Leipzig: Quelle & Meyer. Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Cambridge University Press. Beckmann, J. & Kuhl, J. (1984). Altering information to gain action control: Functional aspects of human information processing in decision-making. Journal of Research in Personality, 18, 223-237. Bergen, A. van, (1968). Task interruption. Amsterdam: North-Holland. Berlyne, D. (1960). Conflict, arousal, and curiosity. NY: McGraw-Hill. Bock, M., & Klinger, E. (1986). Interaction of emotion and cognition in word recall. Psychological Research, 48, 99-106. Butterfield, E. C. (1964). The interruption of tasks: Methodological, factual, and theoretical issues. Psychological BufZetin, 62, 309-322. Dorner, D., Schaub, H., Staudel, T, & Strohschneider, S. (1987). Ein System

zur Handlungsregulation oder die Interaktion von Emotion, Kognition und Motivation (Projekt "Mikroanalyse" DFG 200/5-7. Report # 57). Bamburg, West Germany: University of Bamberg, Lehrstuhl Psychologie 11. Erdelyi, M. H. (1984). The recovery of unconscious (inaccessible) memories: Laboratory studies of hypermnesia. In G. H. Bower (Ed.), The psychology of learning and motivation: Vol. 18 (pp. 96-127). New York Academic Press. Frijda, N. H. (1986). The emotions. Cambridge, M A Cambridge University Press. Goschke, T., & Kuhl, J. (1988). Memoy for intention related knowledge: Taskirrelevant activation of @re goals. Manuscript in preparation. Osnabriick, West Germany: University of Osnabriick.

406

Julius Kuhl and Miguel k e n - S a a d

Kazen-Saad, M., & Kuhl, J. (in press). Motivationale und volitionale Aspekte der Depression: Die Rolle der Lageorientierung. In R. Straub, G. Hole, & M. Hautzinger (Eds.), Denken, Fuhlen, Wollen und Handeln bei depressiven Menschen. Bern: Lang. Kuhl, J. (1983). Motivation, Konflikt und Handlungskontrolle. Heidelberg: Springer-Verlag. Kuhl, J. (1985). Volitional mediators of cognition-behavior consistency: Selfregulatory processes and action vs state orientation. In J. Kuhl, & J. Beckmann (Eds.), Action control: From cognition to behavior (pp. 101128). Heidelberg: Springer-Verlag. Kuhl, J. (1987). Action control: The maintenance of motivational states. In F. Halisch, & J. Kuhl (Eds.), Motivation, intention, and volition (pp. 279291). Berlin: Springer-Verlag. Kuhl, J. (in press). Commentary on Logue, A. W. (1987). Integrating research on self-control. Behavioral and Brain Sciences. Kuhl, J., & Beckmann, J. (Eds.). (in press). Volition and personality: Action and state-oriented modes of control. Gottingen: Hogrefe. Kuhl, J., & Helle, L. (1986). Motivational and volitional determinants of depression: The degenerated intention hypothesis. Journal of Abnormal psycho lo^, 95, 247-251. Kuhl, J., & Helle, L. (in press). Depression and action control: An informationprocessing model and an experimental test. In J. Kuhl, & J. Beckmann (Eds.), Volition and personality: Action and state-oriented modes of control. Gottingen: Hogrefe. Kuhl, J., 8z Kazen-Saad, M. (1988). A motivational approach to volition: Activation and deactivation of memory representations related to uncompleted intentions. In V. Hamilton, G. H. Bower, & N. H. Frijda (Eds.), Cognitive perspectives on emotion and motivation (pp. 63-85). Dordrecht, The Netherlands: Kluwer Academic Publishers. Kuhl, J., & Kraska, K. (in press). Self-regulationand metamotivation: Computational mechanisms, development, and assessment. In R. Kanfer, P. L. Ackerman, & R. Cudeck (Eds.), Abilities, motivation, and methodology: The Minnesota Symposium on individual digerences. Hillsdale, NJ: Lawrence Erlbaum Associates. Leslie, A. M. (1987). Pretense and representation: The origins of "theory of mind." Psychological Review, 94, 412-426. Lewin, K. (1935). A dynamic theoly of personality. New York: McGraw-Hill. Mischel, H. N., & Mischel W. (1983). The development of children's knowledge of self-control strategies. Child Development, 54, 603-619. Neumann, 0. (1987). Beyond capacity: A functional view of attention. In H. Heuer, & A. F. Sanders (Eds.), Perspectives on perception and action (pp. 361-394). Hillsddale, NJ: Earlbaum.

Volition and SelfRegulation

407

Neumann, O., & Prinz, W.(1987). Kognitive Antezedenzien von Willkiirhandlungen. In H. Heckhausen, P. M. Gollwitzer, & F. E. Weinert (Eds.), Jenseits des Rubikon: Der Wlle in den Humanwissenschafren (pp. 195215). Berlin: Springer-Verlag. Norman, D. A., & Shallice, T. (1985). Attention to action: Willed and automatic control of behavior. In R. J. Davidson, G. E. Schwartz, & D. Shapiro (Eds.), Consciousnessand self-regulation:Advances in research: Vol. 4. New York: Plenum. Posner, M. I. (1978). Chronometric explorations of mind. Hillsdale, NJ: Lawrence Erlbaum Associates. Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32, 3-25. Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning Memoy, and Cognition, 14, 501-518. Smolensky, P. (1988). On the proper treatment of connectionism. Behavioral and Brain Sciences, 11, 1-74. Steinsmeyer-Pelster,J., & Schiirmann, M. (in press). Determinants and effects of action and state orientations. In J, Kuhl, & J. Beckmann (Eds.),

Volition and personality: Action and state-oriented modes of control. Gottingen: Hogrefe. Toda, M. (1982). Man, robot, and society. The Hague: Nijhoff. Tulving, E. (1985). How many memory systems are there? American Psycholo@t, 40, 495-501. Zeigarnik, B. (1927). iiber das Behalten von erledigten und unerledigten Handlungen. Psychologische Forschung 9, 1-85.

This Page Intentionally Left Blank

VOLITIONAL ACTION, W.A. Hershberger (Editor) 0 Elsevier Science Publishers B. V. (North-Holland), 1989

409

CHAPTER 18 LEVELS OF INTENTION IN BEHAVIOR Richard S. Marken and William T. Powers The nervous system can be modeled as a hierarchy of control systems (Albus, 1981; Arbib, 1972; Powers, Clark, & McFarland, 1960a, 1960b). Hierarchical models are motivated, in part, by the hierarchical appearance of the behavior produced by the nervous system (Lashley, 1951; Miller, Galanter, & Pribram, 1960). Complex behavior can be seen as the result of simpler behaviors that are themselves the result of even simpler behaviors, this reduction stopping only at the level of muscle tensions. Thus, going to the store is done by driving a car, which is done by turning a wheel, which is done, ultimately, by tensing muscles. While hierarchical models can be devised that reproduce a given behavior (Pew, 1966; Rosenbaum, Kenny, & Derr, 1983; Marken, 1986), it is difficult to show that the same behavior could not equally well be produced by a single-level model that is no more complex (Klein, 1983). Nevertheless, the concept of hierarchical organization in behavior is accepted by most psychologists. This acceptance is based largely on persuasive descriptions of brain structure and plausible-but untested-diagrams of neural organization (Davis, 1976). Laying out an orga