UV
NATURAL MINDS
UV
THOMAS W. POLGER
Natural Minds
Natural Minds
Thomas W. Polger
A Bradford Book The MIT Press...
77 downloads
1384 Views
1MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
UV
NATURAL MINDS
UV
THOMAS W. POLGER
Natural Minds
Natural Minds
Thomas W. Polger
A Bradford Book The MIT Press Cambridge, Massachusetts London, England
© 2004 Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher. This book was set in New Baskerville by SNP Best-set Typesetter Ltd., Hong Kong, and was printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Polger, Thomas W. Natural minds / Thomas W. Polger. p. cm. “A Bradford book.” Includes bibliographical references and index. ISBN 0-262-16221-0 (hc.: alk. paper) 1. Mind–brain identity theory. I. Title. B105.M55P65 2004 128¢.2—dc21 10 9 8 7 6 5 4 3 2 1
2003048826
for Amil, Elizabeth, Frank, and Selma
Contents
List of Figures
ix
Acknowledgments
xi
Introduction 1 Minds, Brains, and Multiple Realizability
xiii 1
2 The Illusion of Contingent Identity
39
3 Varieties of Functionalism
71
4 Realization and Generic Functionalism
111
5 Functional Realizations
139
6 Functional Properties and Mechanistic Explanations
181
7 Dennett’s Challenge
213
8 Minds, Brains, and Persons
241
Notes
247
References
271
Index
289
List of Figures
Figure I.1 The natural method. Figure I.2 Naturalized metaphysics. Figure 1.1 Cortical plasticity. Figure 1.2 A simple carburetor. Figure 1.3 Brodmann’s brain “maps.” Figure 1.4 Two views of multiple realizability. Figure 1.5 Multiple realizability for identity theories. Figure 2.1 Experimental techniques of neuroscience. Figure 2.2 fMRI images of Polger’s brain. Figure 3.1 The anatomy of machine functionalism. Figure 3.2 The logical space of functionalisms. Figure 5.1 Machine table for addition. Figure 6.1 The hierarchy of mechanisms. Figure 7.1 The Zombie Scorecard.
Acknowledgments
I have had the pleasure of discussing my ideas with many people. I am especially grateful for the guidance of Owen Flanagan, Güven Güzeldere, David Sanford, Robert Brandon, Dale Purves, Fred Dretske, and Bill Lycan. Steve Geisz, Brook Sadler, Ana Santiago, Ken Sufka, Frédéric Bouchard, and Eddy Nahmias were my philosophical family as I set out on this project, and I thank them for putting up with me. I benefited from extensive discussions of this project with Michael Lynch, and with my colleagues and students at the University of Cincinnati, especially John Bickle, Marica Bernstein, Chris Gauker, Bob Richardson, and Rob Skipper. For an incredible intellectual and personal experience, I am indebted to David Chalmers, David Hoy, and all the participants in the National Endowment for the Humanities summer institute on consciousness and intentionality at the University of California, Santa Cruz, during the summer of 2002. Work on this book was supported by the Charles P. Taft Memorial Fund at the University of Cincinnati, which also enabled Matt Woodruff to help me prepare the final manuscript. Thank you all. This book is based on my doctoral work at Duke University. A version of chapter 1 appeared as “Putnam’s Intuition” in Philosophical Studies 109 (2) 2002: 143–170, used with kind permission of Kluwer Academic Publishers (copyright 2002). A version of chapter 7 appeared as “Zombies Explained” in Dennett’s Philosophy: A Comprehensive Assessment (Ross, Brook, and Thompson, eds., MIT Press,
xii Acknowledgments
2000). I thank the publishers of those volumes for permission to use my work herein, as well as Tom Stone, Judy Feldmann, and the staff at the MIT Press for their assistance. This project would never have come to fruition without the wisdom, good spirits, and enthusiasm of Owen Flanagan. I have had the fortune to discuss all of the ideas in this book with him, and nearly all the words. Owen is my mentor and collaborator, and above all my friend. Heather Quinley was my guardian angel on the concluding leg of this adventure and has kept me as sane as anyone finishing a book could be. I hope that she will let me write another book someday, but it is really too much to ask. My family has supported me in innumerable ways. They taught me the importance of many things in life, books especially. My parents and my sister, Sarah, are still not quite sure what to think about having a philosopher in the family. Nor am I.
Introduction
I am a conscious organism. Human beings are conscious organisms. Somehow or other we are living creatures that have experiences. Moreover, I suppose, we have our experiences in virtue of the kinds of creatures that we are. That is, the fact that we are conscious is a fact about us as organisms. I am interested in understanding how it is that we are conscious organisms. I want to know: What is the nature of the world and of us as creatures such that we are conscious organisms? One part of the answer now seems clear. At least for human beings, consciousness has something to do with having a brain. In this book I defend the view that our minds are our brains. 1 Naturalism and Neurons Perhaps somewhere there are creatures who are conscious but whose experiences have nothing to do with brains or anything like brains— perhaps their experiences have to do with inflatable cavities in their feet or with silicon chips in their bellies. We’ll have to think carefully about how much credence to give to these sorts of science-fiction cases. But for human beings, and for other terrestrial creatures that seem to be conscious, it’s a good bet that experiences have something to do with brains, or at any rate with neurons. With his trademark how-could-anyone-think-otherwise candor, John Searle writes: The famous mind–body problem, the source of so much controversy over the past two millennia, has a simple solution. This solution has been
xiv Introduction
available to any educated person since serious work began on the brain nearly a century ago, and, in a sense, we all know it to be true. Here it is: Mental phenomena are caused by neurophysiological processes in the brain and are themselves features of the brain. To distinguish this view from the many others in the field, I call it “biological naturalism.” Mental events and processes are as much part of our biological natural history as digestion, mitosis, meiosis, or enzyme secretion. (1992: 1)1
Searle’s contention that mental phenomena are biological processes like any other leaves many questions unanswered. Just being told that consciousness happens in brains doesn’t tell us much. At first glance, the idea that my mental states are (somehow or other) states of my brain is hardly less vexing than the idea that my mental states occur (somehow or other) in my immaterial soul or in the rings of Saturn. How is it possible that my sensations are biological events or processes? Consciousness remains a great mystery. We marvel as much today as William James did when he described conscious experience as a “great blooming buzzing confusion.”2 Although many of us share this sense of wonder, the fact is that we do not all agree about what it is that makes consciousness so perplexing. Some philosophers and scientists think that the remaining questions are only symptoms of the complexity of brains and the nascence of the brain sciences. There is a mystery about consciousness, they admit, but the enigma regards an extraordinarily complicated biological organ rather than any philosophical puzzle. Others think that consciousness, even if we were to discover everything there is to know about brains, presents us with a philosophical difficulty that is not solvable by us, or perhaps even by much smarter creatures. Some think that understanding consciousness will force us to revise, in one way or another, our fundamental understanding of the universe. Still others think that we do not even know where to begin to seek understanding of consciousness. My approach acknowledges that we need to know more about brains in order to understand consciousness fully. Perhaps some problems about consciousness will dissipate or vanish altogether when we know more about brains. But neurobiology, cognitive neuroscience, and the lot will not—cannot—go all the way on their own.
xv Introduction
The discovery that consciousness happens in the brain gives focus to our fascination. It leads to questions about brains, to be sure. But it also leads to questions about the world of which we are a part. At least some of those questions are philosophical questions. This is a work of philosophy. Not neurophilosophy (Patricia Churchland), or psychophysiology (Austen Clark), or heterophenomenology (Daniel Dennett). This is philosophy. It is metaphysics—though it is naturalized metaphysics, to be sure. Naturalism in philosophy is all the rage. But naturalism is not new. Naturalized philosophy is strangely like decaffeinated soft drinks: Kola beans have little caffeine in them to begin with; caffeine is added to caffeinated colas.3 Hence manufacturing “decaffeinated” cola does not involve (as the term implies) taking anything away; it merely requires not adding caffeine. Conversely, “naturalizing” metaphysics sounds like it might require adding something unsavory to our otherwise wholesome philosophy. But naturalized philosophy was the only kind of philosophy for much of the history of ideas, more or less. “Naturalizing” philosophy, then, is a matter of simply leaving put what is sometimes taken away. I have announced that I will pursue naturalized metaphysics, but I do not take myself to be advancing a general account of naturalism in philosophy. Nevertheless, it may be useful to note some of the main features of the naturalistic approach. My naturalism follows something like what Owen Flanagan (1992, 2000) calls the natural method. Flanagan’s idea is that the best way to formulate a theory about consciousness is to strike a balance between the competing claims of first-person experience, psychology, neuroscience, evolutionary biology, and so forth (see figure I.1). The tactic is “to see whether and to what extent the [many] stories can be rendered coherent, meshed, and brought into reflective equilibrium” (Flanagan 1992: 11). The natural method is a strategy for inference to the best explanation in theorizing about minds and consciousness. Like Descartes’s methodical worrying about a deceiving demon, thinking in terms of the natural method helps us to be scrupulous about our assumptions and what we take for granted. At the start of the day we invite all claims, and then we sort through them to discover a point of balance. At the end of the day some sources may
xvi Introduction
PHENOMENOLOGY
U R AL M AT
PHILOSOPHY
ET
)
)
EX
(E
H OD (
THE N
EVOLUTIONARY BIOLOGY
X P A DE D N
PSYCHOLOGY & PSYCHIATRY
P A N DE D
ANTHROPOLOGY & SOCIOLOGY NEUROSCIENCE
Figure I.1 The natural method. Adapted from Flanagan (1995b). Used by permission.
prove to be more useful than others, but we do not presuppose this to be so from the outset. Thus far I follow Flanagan. But his goal is to advance specific, broadly empirical accounts of consciousness and its varieties, for example, of dream experiences (1995b, 1996, 2000). In contrast, in this book I am concerned to consider underlying philosophical issues of fairly general and traditional sorts. Naturalistic metaphysics aims to secure the philosophical foundations of naturalistic explanations of mind and consciousness, such as those advocated by Flanagan. The natural method takes philosophical claims to be as malleable and revisable as any others. Each discipline that contributes to the method is subject to feedback in the project of seeking reflective equilibrium—feedback both from the larger project and from within its own traditional disciplinary boundaries. As the overall picture begins to come together, we may find that we should reshape some of the puzzle pieces. Some of the pieces are philosophical; and of those, some are metaphysical. Naturalized metaphysics, then, is the
xvii Introduction
MORAL THEORY
PHILOSOPHY OF SCIENCE
TU R A
D MET
HYSICS
NA
E IZ
AP
L
METAPHYSICS PHILOSOPHY OF MIND EPISTEMOLOGY
PSYCHOLOGY NEUROBIOLOGY CLINICAL PSYCHOLOGY
COGNITIVE NEUROSCIENCE
Figure I.2 Naturalized metaphysics. Adapted from Flanagan (1995b). Used by permission.
metaphysics that complements the natural method (figure I.2). It is sometimes supposed that genuinely metaphysical theses are immune to this sort of revision. I don’t know if anyone ever actually pursued the austere but hobbled metaphysics that is critiqued by Carnap (1935) or Rorty (1979). Perhaps it is, as the critics allege, an impossible pursuit. But I don’t see why we must accept such constraints on metaphysics, or think that studying metaphysics requires them. Naturalized metaphysics is metaphysics: It seeks to provide answers to basic questions about the nature of the world and its occupants. But the naturalist sees no reason to think that metaphysics must be conducted in ignorance of what other disciplines, philosophical and empirical, tell us about the world. Naturalism avails itself of whatever conceptual materials are available and eschews prior constraints on what will prove useful. Like the natural method itself, naturalized metaphysics is a basically pragmatic approach. So the “naturalism” of naturalized metaphysics is in the first case methodological. But that does not mean that today’s naturalism is without content of its own. The natural method is a live process. As
xviii Introduction
we make progress toward an overall picture some specific theses emerge, adding substance to the method of naturalism. One such constraint is what we might call the No Ghosts Rule : there is no need to appeal to immaterial stuff in order to explain mental phenomena. This is a prohibition against substantial (substantival) forms of dualism about the mind. But the No Ghosts Rule is not a background assumption. Rather, it is the outcome of pursuing the natural method up to the present point. It records the current state of our knowledge about minds. If there is a first principle behind the No Ghosts Rule it is not a metaphysical stricture but a methodological one. Considerations of parsimony discourage us from explaining minds in a way that is disconnected from our explanations of other phenomena. The natural method converts the observation that immaterial substances are not appealed to by any theories and explanations concerning other parts of the natural world into the injunction that immaterial substances not be introduced specially to explain facts about minds. The rule is therefore restricted to specific questions about minds, leaving open that some other questions may demand appeal to immaterial substances. If, in coming to reflective equilibrium, we were to find a general reason to admit immaterial substances into our ontology, we could do so. The No Ghosts Rule merely prohibits a familiar but ad hoc solution to questions about minds. Despite worries about how to define or characterize naturalism, in most cases in philosophy of mind it is clear whether a theory appeals to nonnatural stuff. As with obscenity, we know nonnaturalism when we see it. If we were to discover that there are ghosts or other objects composed of a substance that is not physical or material, then we would know that naturalism is false.4 Of course, the No Ghosts Rule puts the naturalists’ burden on explaining what counts as physical or material. But notice that, on the present view, the question of what counts as physical or material—like the rule itself—is internal to the natural method. So although I provide no general answer to the question, we know what process we would engage in to formulate an answer: We would study physics. Thus the rule does not tie the hands of philosophers in their investigations, much less empirical scientists. It may be that we are faced with problems that require a more
xix Introduction
detailed theory of what counts as natural, physical, or material. In that case, a detailed articulation of naturalism would begin from the uncontroversial paradigm cases that help to set its limits and provide the needed contrasts. Physics and biology are natural sciences if anything is; if physics and biology are not naturalistic, then naturalism is simply incoherent.5 A nonnatural theory would be one that cannot be brought into reflective equilibrium with physics and biology. A nonnatural metaphysics would be one whose ontology is incompatible with that of physics and biology. The constraints created by my kind of naturalism are fairly mundane, all things considered. Because my goal is to adjudicate among philosophical views that all aim to be broadly naturalistic, locating the precise boundaries of naturalism is not crucial for me. But I will also appeal to claims that are not universally accepted even among naturalistic philosophers of mind. Some of these theses are about minds in particular, and some are more general. 2 Guiding Prejudices A creature is conscious if it has conscious states. (What it takes to have a conscious state is not so straightforward, we shall see.) A state is a conscious state if it exhibits the Nagel-property—if there is something that it is like to have or be in that state. I borrow this locution from Thomas Nagel’s famous (1974) paper, “What Is It Like to Be a Bat?” and I find it quite evocative. Some philosophers claim that this “what-it-is-like” locution is useless and fails to so much as gesture in the direction of distinct phenomena. If the complaint is that the what-it-is-like locution does not explain consciousness, then that is fair. It does not. But talking in this way seems to be a perfectly serviceable way of pointing out the phenomena that are of interest. There is something that it is like to see a red apple or smell tobacco or taste a martini. What it is like to see a red apple or smell tobacco or taste a martini is different from what it is like to see a green apple or taste tobacco or hear a symphony. As long as one does not suppose that in addition to bodies and brains there are also little objects (perhaps in our heads), “what-it-is-likes,” I see no special problem with setting out in this way.
xx Introduction
To say this much is not to presume that consciousness is either singular or homogeneous. What conscious states have in common is that there is something it is like for their bearers to have them. Beyond that, there may be great diversity. The Nagel-property might be a single property, or it might not. There are many varieties of consciousness, and much remains to be said about the structure of the relations between conscious states, and between conscious states and everything else. I assume that it makes sense to talk and to theorize about mental states or processes generally, and about conscious mental states and processes in particular. This claim, and the natural method itself, seems to presuppose that contemporary cognitive and brain sciences are not entirely misguided pursuits. They may be wrong about the details and far from complete. But they are not wrong that there are some phenomena of interest (e.g., belief, desire, and sensation), and they are open to naturalistic investigation. Like that of Flanagan (1992), my naturalism is therefore constructive rather than eliminative, mysterian, or agnostic. In this book I advocate a mind–brain identity theory: the view that minds are brains, that sensations are brain processes. I therefore assume that brains exist. That is, I assume that there are macroscopic, composite, physical objects among which seem to include: brains, bricks, trees, platypuses, laptop computers, and lumps of clay. I also assume that brains, bricks, trees, platypuses, laptop computers, and lumps of clay can be in states and undergo processes, and therefore that they and their states can have causal powers. In particular, I assume that bricks can cause windows to break (when appropriately applied), that silicon chips can cause computers to open doors (when appropriately activated), and that brains and their states have causal powers in the same ways as bricks and silicon chips. (This would hardly be worth mentioning were it not that some philosophers deny one or more of these claims. This seems to me rather silly, but I assure you that it is true.) I do not offer any philosophical accounts of either material composition or of causation, but I assume that those relations do not present any special problem for brains that they do not present for bricks. Since I will argue that minds are identical to brains, I will thereby argue that minds are at
xxi Introduction
least as real and causally efficacious as bricks. To get that far would satisfy most philosophers of mind. The view that sensations are brain processes is sometimes called “reductive”—proudly by its advocates and pejoratively by its critics. The identity theory that I advocate is not “reductionist,” and it neither requires nor recommends “reduction.” But my account is not “antireductionist” either. In chapter 6, I argue that it is a mistake to think of problems in philosophy of mind in terms of “reductionism” or its rejection. Clearly much more needs to be said about the difference I am now insisting on between identity theories and “reductionist” theories, a distinction that some will find perverse. For the moment I shall only register the distinction and point out that none of my arguments depends on adopting either “reductionist” or “antireductionist” stances. So that we don’t forget, I’ll continue to deploy the annoying device of scare-quotes to indicate my apprehension about the “reduction” debates. Finally, I take conscious experience to be the fundamental mental phenomenon. A theory of the nature of conscious experience is a theory of the nature of minds and mental states in general. Certainly there are many bodily and neurological states that are not conscious states, but are those states of mind? On my view all mental states, events, properties, or processes are conscious states, conscious events, conscious properties, or conscious processes. Some of these conscious mental states are also contentful states; they are states like beliefs and desires that have contents or are directed at objects. It is not a requirement, on my view, that all mental states be contentful states. But no state is a mental state if it is not a conscious mental state—a state that has the quality that there is something it is like to have that state. I therefore reject what is sometimes known as Brentano’s thesis, that “intentionality is the mark of the mental.” This puts me at odds with the vast majority of philosophers, even most of those who are interested in consciousness. Intentionality, in philosophical parlance, is the aboutness or contentfulness or directedness that is sometimes thought to be distinctive of language and thought. My thoughts, and these words, may be about or directed at, say, the woman at the party who was drinking a martini.
xxii Introduction
The “mark of the mental” is the grail of philosophy of mind, a single characteristic that distinguishes mental states from nonmental states. That intentionality (or some related phenomenon, such as representation) is the central—perhaps the only—mental phenomenon, seems to be widely accepted within philosophical circles, not only in philosophy of mind but in the broader philosophical community. But this is precisely the claim that I resist, and the tide may be turning in my favor. A number of philosophers have recently defended views that are prima facie complementary to my approach. I have in mind not only Owen Flanagan (1992), John Searle (1992), and Galen Strawson (1994), but also Brian Loar (1981), Charles Siewert (1998), Terence Horgan and John Tienson (2002), and David Pitt (in press), among others. I don’t know whether I should adopt any of these particular versions of the view, nor whether any of those authors would welcome my metaphysical claims. But it seems that the received view is drawing renewed scrutiny. I will not develop my own arguments here for the thesis that consciousness is the basic mental phenomenon. That is a whole project—perhaps a whole career—in itself. You should take it as an unargued assumption of this book that consciousness, rather than intentionality, is the basic mental phenomenon.6 A fallback strategy for the reader is to interpret this book as though it were about conscious mental states only, and to suppose that I simply neglect intentional states. You can treat me as what Terence Horgan and John Tienson (2002) call a “separatist,” and take it that I have set aside the problem of intentionality for another day. In that case, you will read me as adopting a somewhat less radical position than the one I in fact endorse, for the view that sensations but not intentional states are identical to brain states has been favored at one time or another by, among others, Ned Block (1978), Sydney Shoemaker (1982), and Owen Flanagan (1992). Yet even so diluted, this interpretation of my view would still be controversial enough to be interesting, in particular because it stands in opposition to currently fashionable representationalist theories of consciousness advocated, for example, by William Lycan (1987, 1996), Michael Tye (1995, 2000), and Fred Dretske (1995). The diluted view is also in conflict with the Multiple Drafts model of Daniel Dennett (1991),
xxiii Introduction
and the antirepresentationalist view recently defended by Hilary Putnam (1999). Consciousness has enjoyed an astounding rebirth as a front-page topic in philosophical and scientific circles in the last decade or so. I believe that this is not a passing fad, that it is a substantial and important event in philosophy of mind. The event is not that philosophers now talk about consciousness but that we now admit that we talk about consciousness. (Let us not forget that sensations of pain and color—especially red, for some reason—have long been philosophers’ favorite cliché mental states.) All along philosophy of mind has been concerned with consciousness, even if it did not use the term.7 It is hard to miss the continuous discussion of “experience” and “qualia,” suggesting that consciousness was never gone, it just passed by other names. Seeing colors and feeling sensations were always among philosophers’ paradigm cases of mental events and processes. I shall take them to be the phenomena in need of explanation. 3 The Road Ahead In this book I advocate the view that conscious processes, events, states, or properties are type-identical to biological processes, events, states, or properties of brains—a version of identity theory. Moreover, mine is an extremely tough-minded identity theory because, unlike other current and past mind–brain identity theorists, I contend that minds are necessarily identical to brains. Among current identity theorists, few explicitly endorse this strong claim.8 The identity theory, so construed, has much to recommend it. Among its merits are that it is simple, it is unimpeachably naturalistic, and it provides a robust account of mental causation. Yet for some years now the theory has been considered dead in the water. To establish an identity theory is more than I could hope for, and certainly more than I can deliver. In this book my goal is simply to revive the theory, to get it back on the table as a live option. My strategy therefore depends a great deal on clearing the philosophical landscape and making it safe to be an identity theorist. In broad outline this book has three parts. The first is dedicated to answering two
xxiv Introduction
objections to the identity theory that claim to prevent it from getting off the ground. The second part is focused on critically examining the most prominent alternatives to identity theories: functionalist theories. The keystone of this discussion and of the book is the development and deployment of an account of the realization relation that is crucial to functionalist theories but usually taken for granted. In the closing portion of the book I show that identity theory can make use of some explanatory resources that are often supposed to be special to functionalism. Specifically, I argue that identity theory is compatible with a robust model of mechanistic explanation. The two major obstacles to identity theory that I focus on are problems that arise only or in a special way for identity theory and not for other naturalist or physicalist theories of mind. The most well known and widely discussed is Hilary Putnam’s argument from multiple realizability, which claims that the identity theory cannot be right because some creatures that do not have brains just like ours could—in fact, do—nevertheless have mental states. According to this critique, the identity theory fails because it is not general enough to cover all instances of mental states. So mental states cannot be identical to brain states. In chapter 1, I argue that the interpretations of multiple realizability that are problems for identity theory are not plausible independently of prior commitment to a substantial metaphysical thesis about minds. These sorts of multiple realizability therefore cannot be premises in arguments against mind–brain identity claims. And I show how an identity theory can accommodate those interpretations of multiple realizability that are independently plausible. A second problem for identity theory is Saul Kripke’s argument that the apparent contingency of the relationship between minds and brains cannot be explained away, so if identities are necessary we must abandon the identity claim. In chapter 2, I provide a new model for explaining away the contingency of scientific identities that accounts for the apparent contingency of the mind–brain relation. My model appeals to the rudimentary condition of our capabilities to individuate mental states and brain states. A welcome side effect of my model is that it also helps to dissolve the problem
xxv Introduction
presented by Joseph Levine’s “explanatory gap” argument, which is an epistemic variation of Kripke’s metaphysical argument. Levine contends that even if naturalism is correct, it yields inadequate explanations of minds. But if identity theory is viable then the explanatory gap can be collapsed. Together the considerations of chapters 1 and 2 demonstrate that identity theory is philosophically viable. However, the apparent success of Putnam’s alternative, functionalism, remains an obstacle to a full revival of identity theory. Versions of functionalism are the dominant views in contemporary philosophy of mind. I argue, in chapters 3 through 5, that functionalism is not as well situated as it claims to be. My contention is based in part on the observation that substantial differences between varieties of functionalism have escaped attention because they arose bit by bit as the theory evolved over time, and because there is an understandable tendency to confront a complicated doctrine in a piecemeal way. Only by taking the time to formulate an overall view of the body of functionalist doctrines and their history can we find ourselves in a position to assess its many claims. The general outline takes shape in chapter 3. There I begin to set out an extensive taxonomy of functionalism. I distinguish six varieties of functionalism according to the sorts of questions for which they are proposed answers: metaphysical, intentional, semantic, theoretical, explanatory, and methodological. Metaphysical functionalism is the particular form that competes with identity theory, so we want to be clear about what it claims and how it is related to other varieties of functionalism. Next I examine the notions of function to which each variation of metaphysical functionalism appeals. I contend that varying notions of function dictate the key characteristics of the theories that appeal to them—crucially, how the functionalists’ key realization relation is to be understood. In chapter 4, I defend an account of functional realization and sketch the work that it is supposed to do for functionalist theories. Then, in chapter 5, for each of a number of familiar notions of function, I explore the corresponding realization relation, measuring each against the characteristics that the advocates of functionalism themselves claim for it. I argue that no single
xxvi Introduction
version of functionalism has all of the characteristics, and it is questionable whether any could. While it is not impossible that someone could formulate a version of functionalism that succeeds, I caution against too much optimism. Some will resist this suggestion even while acknowledging the difficulties faced by functionalism. Without a decisive refutation of functionalism, it may be objected that some version must be correct, despite the deficiencies of the current candidates. In chapter 6, I consider two such arguments. First I consider versions of the argument that metaphysical functionalism about the mind must be correct because the world is “functions all the way down.” The second argument claims that metaphysical functionalism is entailed by a general commitment to autonomous psychological explanation. Seeing why these arguments fall short allows us better to understand what the functionalist wants to get out of a theory of mind. I contend that these arguments reveal that most functionalists seek a mechanistic theory of mind. It is here that I must make good on my promise to distinguish the identity theory from “reductionist” accounts of minds. Considering mechanisms and mechanistic explanations allows me to exhibit the resources available to the identity theory. I argue that identity theory is a robust and flexible mechanistic theory of mind, and that it can satisfy the kinds of explanatory demands that have sometimes seemed to favor functionalism. In fact I suggest that identity theory is at an advantage even on this count. To illustrate the resources of identity theory and the utility of distinguishing mechanism from functionalism, I confront Daniel Dennett’s zombie challenge: to show that the philosophical notion of “zombies”—creatures in some way like us but lacking any conscious states—is not incoherent. In chapter 7, I argue that the identity theory can allow for functional duplicates without entailing that consciousness is epiphenomenal in any way that should concern naturalists. I diagnose Dennett’s worries as stemming from a confusion of functionalism with mechanism. I conclude by briefly entertaining an admittedly counterintuitive consequence of my view, having to do with the possibility of isolating mental states in petri dishes. This strange case suggests an
xxvii Introduction
affinity between the identity theory and some contemporary theories of personal identity. It is too much to hope that I could both resurrect the identity theory and decisively settle the question in its favor. But I believe that I have made a strong case for a theory of mind that is clearly in the tradition of identity theories. Such a theory must be reckoned as a serious contender and is worthy of further consideration. There is a great deal to be said in favor of identity theory. Let me begin.
1 Minds, Brains, and Multiple Realizability
Questions about minds are as ancient as any in philosophy. But the mind–body problem as it exists today first appeared only in the middle of the twentieth century. The current mind–body problem does not concern the issue of whether minds are part of the physical world just as are aardvarks, hearts, rocks, cigarettes, heat, and the like. The problem is not whether minds are part of the natural world, but how they are. Although the idea that minds are part of nature is not itself new, until quite recently there were no theories that could make sense of the contention. As a result, it was difficult even to imagine how mental phenomena could be natural phenomena.1 Many factors contributed to the philosophical and scientific milieu that makes a naturalistic theory of mind credible. Advances in biology (e.g., discovery of the double-helix structure of DNA), psychology (e.g., replacement of behaviorism with cognitive psychology), mathematics (e.g., computability theory)—not to mention the advent of technologies suitable for manipulating the world as theorized by these new accounts, as well as a healthy dose of science fiction—all helped to set the stage for credible physicalist theories about minds. And, of course, the nascent neurosciences began to reveal that the complexity of the brain is of a scale not previously imagined. At last we could begin to see how minds might be sophisticated physical phenomena. As Owen Flanagan writes, “we do understand how physicalism can be true. It can be true if every mental event is realized in the brain” (1992: 93). So the current mind–body problem
2 Chapter 1
does not concern the general possibility of a naturalistic theory of mind. Instead, at issue is what the world—minds, brains, organisms, environments, and so forth—must be like if a naturalistic theory of mind is true: the mind–body problem—our mind–body problem—has been that of finding a place for the mind in a world that is fundamentally physical. The shared project of the majority of those who have worked on the mind–body problem in the past few decades has been to find a way of accommodating the mental within a principled physicalist scheme, while at the same time preserving it as something distinctive—that is, without losing what we value, or find special, in our nature as creatures with minds. (Kim 1998: 2)
Theories are plentiful. They compete, for example, on the basis of which best complement other natural theories, and which give the most satisfactory or satisfying accounts of a range of mental phenomena. Of course there are still some theorists who doubt that mental phenomena are exhausted by physical phenomena. But they are not typically vexed by what a physicalist theory of mind would look like.2 Instead, they simply contend that the world does not in fact work, perhaps could not work, in the way that physicalist theories of mind require. One might even believe that physicalism is correct up to a point, but that it leaves an unexplained remainder, as David Chalmers argues (1995, 1996a). So the important issues for a naturalistic philosophy of mind are how it requires the world to be, and whether the world lives up to those expectations. 1 Identity Theory and Putnam’s Intuition The most rudimentary naturalistic theory of mind is the mind–brain identity theory: minds are identical to brains. In J. J. C. Smart’s words, “Sensations are nothing over and above brain processes” (1959: 163). Identity theory was the first serious contender for a mechanistic theory of mind.3 The slogan “sensations are brain processes” is admittedly a bit crude. For one, it is unclear whether identity theories should be thought of in terms of the identity of organs, or states, or processes, or events, or properties.4 But the core idea is so simple that it can hardly be refined: mental states, processes, events, or properties just
3 Minds, Brains, and Multiple Realizability
are brain states, processes, events, or properties.5 The merits of identity theory are equally plain. Smart advocated identity theory on the grounds that it is parsimonious and ontologically modest; it does not require us to posit new sorts of properties, events, states, and so forth.6 Thus it is thoroughly naturalistic; it appeals only to the ontology of the natural sciences. And identity theory is the only theory with a robust explanation of how mental phenomena can cause physical phenomena: “It makes mental causation entirely unmysterious: Mental causation turns out to be a species of physical causation” (Kim 1996: 56). If mental phenomena are identical to physical phenomena, then no distinctive question about mental causation ever arises. This is no small matter. In spite of its merits, the identity theory was quickly displaced as the primary form of materialism. Its deposer is Hilary Putnam’s functionalism, the ontological thesis that mental states are not identical to physical states of brains but are instead realized or instantiated by physical states of brains. The main arguments against identity theory and for functionalism all originate from the single intuition that some creature could have sensations without having brain states like ours, or any brain at all. Putnam supposed—and many have followed him—that this intuition demonstrates the implausibility of identity theory: Consider what the brain-state theorist has to do to make good his claims. He has to specify a physical-chemical state such that any organism (not just a mammal) is in pain if and only if (a) it possesses a brain of suitable physical-chemical structure; and (b) its brain is in that physical-chemical state. This means that the physical-chemical state in question must be a possible state of a mammalian brain, a reptilian brain, a mollusc’s brain (octopuses are mollusca, and certainly feel pain), etc. At the same time, it must not be a possible (physically possible) state of the brain of any physically possible creature that cannot feel pain. Even if such a state can be found, it must be nomologically certain that it will also be a state of the brain of any extraterrestrial life that may be found that will be capable of feeling pain before we can even entertain the supposition that it may be pain. It is not altogether impossible that such a state will be found. . . . But this is certainly an ambitious hypothesis. (1967, in Putnam 1975c: 436)
Putnam takes the identity theorist to be committed to the view that every possible conscious creature must be capable of having states
4 Chapter 1
just like our brain states. As an alternative, he proposes that conscious states can be realized by various biological or nonbiological states, such that creatures lacking brains and brain states like ours may nevertheless be conscious. At least, it seems probable that mental states can have multiple realizations.7 Functionalism, it is argued, is superior to identity theory because the latter cannot accommodate the overwhelming likelihood that mental states are multiply realizable. Of course identity theory’s sparse form does not leave much opportunity for nuance: mental states are identical to brain states. But as I suggested earlier, a theory of mind rests also on what it says about the world. I contend that the world is as identity theory takes it to be rather than as functionalism takes it to be. Although there is little room to amend identity theory itself, much can be said about how identity theory views the world. In this case, I need to explain how identity theories view multiple realizability. My strategy is that of divide and conquer. I distinguish four interpretations of the multiple realizability intuition. I argue that the usual ways of construing multiple realizability are much stronger than can be supported by Putnam’s intuition alone and should not be admitted. And the plausible forms of multiple realizability do not impugn the prospects for a mind–brain identity theory. Arguments from multiple realizability purport to show that functionalism has the theoretical merit of being more general than identity theory. If two theories are comparable in other ways but one explains more than the other, the more widely applicable theory is to be preferred. Generality is a good thing. Functionalism, because it allows for multiple realizability, can explain the mentality of more kinds of things; specifically, it is not restricted to those creatures with brains like ours. But generality is a matter of degree. What degree of generality is appropriate to a theory of mind? That is, to what degree are mental states multiply realizable? There is no getting around the fact that identity theory cannot handle the wildly different realizations of mental states that functionalism so easily accommodates. In the end the identity theorist will have to settle for identities that are in some way restricted. The questions we must ask, then, are whether restricted identities are bad or compromise identity theory and
5 Minds, Brains, and Multiple Realizability
whether the unrestricted multiple realizability licensed by functionalism is desirable. We need to know how much generality is required if we are to assess whether or not identity theory is compatible with multiple realizability. The answer, to pervert Daniel Dennett’s (1984) expression, is that identity theory can accommodate the varieties of multiple realizability worth wanting. 2 Varieties of Multiple Realizability Worth Wanting Recently there has been renewed attention—largely critical—to multiple realizability. Some have aimed to defuse the threat that multiple realizability seems to present for various theses of explanatory “reduction” (Bickle 1998; Sober 1999; Shapiro 2000, forthcoming). Others have been concerned to resolve the difficulties that multiple realizability is alleged to introduce for explanations of mental causation (Kim 1998; Heil 1999). Still others have questioned the empirical evidence for multiple realizability, and whether it constitutes an obstacle to the practices of neuroscientists (Bechtel and Mundale 1999; Bechtel and McCauley 1999).8 For the most part the authors mentioned treat multiple realizability as a structural constraint on explanations of mind. Each attempts to show that within the constraints of multiple realizability we can nevertheless justify “reductive” explanation, account for mental causation, carry on neuroscientific practice, and so forth. Some, but not all, believe that their conclusions help to support some version of identity theory. Certainly structural features are an important aspect of arguments based on multiple realizability; but what makes multiple realizability such a thorn is that it enjoys a presumptive advantage that is based on the intuitive appeal of its content. That content, in turn, dictates the structural constraints. If we could find reason to question the content of the multiple realizability intuition, then perhaps we would not have to theorize within its structural constraints. And I think we have such reasons. Putnam’s intuition is that we should not expect every kind of conscious creature to have a brain with states just like our own. I find this intuition plausible. It seems to accord with the information that we get from biologists and cognitive ethologists. There is at least
6 Chapter 1
good reason to think that human beings are not the only conscious creatures, even on our little planet. But it is a far stretch from Putnam’s intuition to multiple realizability as it is usually entertained. Consider these samples from the spectrum of multiple realizability: Weak MR. At least some creatures that are not exactly like us in their physical composition can be conscious. SETI MR. Some creatures that are significantly different from us in their physical composition can be conscious.9 Standard MR. Systems of indefinitely (perhaps infinitely) many physical compositions can be conscious. Radical MR. Any (every) suitably organized system, regardless of its physical composition, can be conscious. The forms of multiple realizability most commonly encountered in the literature are what I have here dubbed standard MR and radical MR. Examples are numerous: [T]he mere fact that a creature’s physical states are radically different from ours does not in itself preclude them from being realizations of mental states. (Shoemaker 1981b, in 1984: 280) Creatures on this planet to which we unhesitatingly ascribe a range of mental characteristics differ biologically from us in endless respects. (Heil 1992: 133) [T]here are infinitely (indefinitely) many possible physical “realizations” of pain: C-fibers, silicon fibers, et cetera. (Bealer 1994: 187) Mental property M—say, being in pain—may be such that in humans C-fiber activation realizes it but in other species (think of octopuses and reptiles) the physiological mechanism that realizes pain may well be vastly different. And perhaps there are biological systems—at least no laws of nature rule out such a possibility—that are not carbon- or protein-based, and there can be electromechanical systems, like robots in science fiction, that are capable of having beliefs, desires, and even sensations. All this seems to point to an interesting feature of mental concepts: They include no constraint on the actual physical/biological mechanisms or structures that, in a given system, realize or implement them. (Kim 1996: 74–75) For more than three decades antireductionists have argued that radically different physical systems could realize identical mental kinds. . . . They often defend the premise of multiple realizability with thought experiments
7 Minds, Brains, and Multiple Realizability
involving silicon-based extraterrestrials, computers, androids, robots, and other brainless science fictional beings. (Bickle 1998: 114)
To be sure, if functionalism is correct then standard MR is likely, and perhaps even radical MR. But if a variety of multiple realizability is to be the basis for an argument against identity theory, then it will have to be a form of multiple realizability whose plausibility does itself not depend on the truth of functionalism. This is plain enough, for if the plausibility of multiple realizability depends on the plausibility of functionalism, then the argument from multiple realizability only repeats the assertion that functionalism is plausible. Let us ask: Does anyone have the pretheoretic intuition that mental states admit standard MR—that systems of indefinitely (perhaps infinitely) many compositions can be conscious? Of course, any intuitions are colored by some amount of theory. But functionalism is the particular theory about which we are now concerned; our intuitions regarding what is plausible have not been immune to the proliferation of functionalist theories. I contend that standard MR and functionalism go together. Standard MR is not part of any argument against identity theory that is independent of substantial metaphysical claims. Of course, there is nothing wrong with arguing against identity theory from the combination of functionalism and standard MR; but that would not be an argument that could generate a presumptive judgment against the plausibility of identity theory, as the argument from multiple realizability has been thought to. If the multiple realizability argument is to be a burden to identity theory then it must itself be plausible independent of functionalism. I am not going to rest my conclusion only on this “he-said-she-said” argument. But notice that the case is not as flimsy as it at first seems. It is extremely difficult to see how someone who endorses standard MR can resist the thesis of radical MR, that any suitably organized system can be conscious. As Fred Adams puts the point, “If the state is realizable by indefinitely many [physical-chemical] systems, then why not all? The natural reply is that some things simply are not put together in the appropriate way to be in pain. Hence, we doubt that trees and rocks and grass, etc., feel pain (otherwise we might stop mowing our lawns)” (1979: 159). If standard MR were itself
8 Chapter 1
plausible, its supporters seem to be committed to the stronger thesis of radical MR. I am not claiming that standard MR entails radical MR. It does not. But unless some explanation can be given that justifies constraints on realizers of mental states, the considerations that justify belief in standard MR will also tend to support radical MR. What Adams calls “the natural reply” to block the slip from standard MR to radical MR is not available because to explain why “some things simply are not put together in the appropriate way” requires invoking a substantial theory. But the multiple realizability arguments, if they are to be a problem for the identity theory, are supposed operate prior to adopting any such theory. We can put this problem in the form of a dilemma: Either standard MR is a consequence of prior metaphysical commitments, or it is not. If standard MR is the consequence of a prior metaphysical commitment, then it is question-begging with respect to the identity theory. So it cannot be a consequence of prior metaphysical commitment. But if standard MR is not a consequence of prior metaphysical commitment, then it looks as though it is a premise in a conceivability argument. Yet if it is a premise in a conceivability argument, then there seems to be no reason that radical MR should not also be conceivable. This will be so even if the evidence for the conceivability claim is that some things that are physiologically different from us (e.g., octopi) in fact have sensations (e.g., pain). That evidence, if it is evidence for anything at all, is also evidence for radical MR. It seems that the grounds for believing in standard MR also justify belief in radical MR. But belief in radical MR is more than dedication to theoretical generality. Radical MR is a substantial metaphysical thesis that is not plausible independently of the rejection of identity theory. Radical MR cannot be a premise in an argument against identity theory. If standard MR is not question-begging, then some reason must be given for drawing a line between standard MR and radical MR. But it is not clear that any such explanation could be made without invoking a theory of the nature of mind.10 And if one must invoke a philosophy of mind to justify limiting realizers, then to whatever extent the justification is incompatible with identity theory it is question-begging. A theory-based multiple realizability argument will
9 Minds, Brains, and Multiple Realizability
not have the presumptive force that multiple realizability arguments assume. Needless to say, I dismiss as hogwash anyone’s claim to have metaphysically neutral and prefunctionalist intuitions in favor of radical MR. This is brazen, I know. To see why I can make this claim, it is important to keep in mind how the multiple realizability argument is supposed to work. Multiple realizability is a claim about the generality of accounts of the ontology of minds. Some versions of functionalism, if true, would admit radical MR. But that is not the question at hand. Rather, we want to know whether it is a desiderata on theories of mind that they permit radical MR. Is that degree of multiple realizability a virtue that we expect any such theories to have? Like you, I seem to be able to imagine that things radically different from human beings, with silicon chips rather than neurons, say, could turn out to have conscious mental states. What I am imagining is that some particular substantial account of mental ontology is true, and that on that theory things with silicon chips rather than neurons have minds. Of course I don’t have to articulate the theory; most people most of the time don’t think about that theory when they imagine or enjoy the adventures of affectionate robots or surly talking hats.11 And even so I’m not at all confident that we’re imagining that radical MR is true—that anything could be conscious— rather than standard MR. Radical MR is the provenance of the kind of fantasy that invokes magic to animate trees and teacups, rather than the futuristic robots and aliens whose familiarity in modern storytelling seems to underlie many multiple realizability intuitions. Maybe robots can be conscious. But it is not a prior constraint on any account of conscious experience that it allow radical MR. Yet this is what radical MR asserts when it is used in a multiple realizability argument against identity theory: It maintains that we have prior commitment to ensuring that any theory of mind will be general enough to account for the minds that robots and automata could have. But that cannot be right. If robots can have minds, that is a substantial discovery; it is not something guaranteed merely by a commitment to theoretical generality.12 Some readers will fail to be moved by the powerful argument against standard MR and radical MR just offered, the “Argument
10 Chapter 1
from Nuh-uh.” Let us try another approach, for we identity theorists have many tools at our disposal. The strategy, recall, is divide and conquer—or, more accurately, divide and unify. So far I have been making the case that the stronger forms of multiple realizability should not enjoy our presumptive support. But suppose that, for one reason or another, we are obliged to show that identity theory is compatible with standard MR and radical MR. What resources can an identity theorist offer? We may reply, with Jaegwon Kim: [T]he fact that two brains are physico-chemically different does not entail that the two brains cannot be in the “same physico-chemical state.” Even if we disallow the ad hoc creation of new states by forming arbitrary disjunctions, the remaining possibilities are indeed limitless. . . . If the human brain and the reptilian brain can be in the same “temperature state,” why can they not be in the same “brain state,” where this state is characterized in physicochemical terms? . . . the mere fact that that the physical bases of two nervous systems are different in material composition or physical organization with respect to a certain scheme of classification does not entail that they cannot be in the same physical state with respect to a different scheme. (1972, in Block 1980a: 234–235)
Multiple realizability is typically couched in terms of what sorts of systems are capable of having mental states, and I have followed suit. But the fact—if it is a fact—that many different systems can have the same kinds of mental states does not show that they do not all do so in virtue of having something in common. An easy way to think of this scenario is in terms of systems sharing some properties but not others. Two systems may differ with respect to the properties by which neurological states are categorized, but may nevertheless share other properties that can ground their classification as the same kind of mental state. As Frederick Adams notes, “Just because two [physical-chemical] systems are different kinds of stuff does not mean that they do not share some identical property-kinds” (1979: 158; see also 1985). The identity theorist is free to maintain that creatures that are made of different stuff than human beings, and thereby have different physical states than we do, nevertheless share some properties that we have. Specifically, the usual application of multiple realizability arguments leaves open that these diverse systems may be like us with respect to the properties (whatever they
11 Minds, Brains, and Multiple Realizability
turn out to be) on which conscious experience depends. Multiple realizability intuitions at most suggest that conscious mental states can be realized in various systems, but they do not support the contention that such states are realized by different properties in different systems rather than by properties that the systems share. Call this the Kim–Adams reply. Identity theory contends that mental states are identical to physical states; but that does not require that all mental states of a kind are identical to each other. The relationship between mental states of a kind is merely that they are the same with respect to those properties characteristic of the kind. And it is quite reasonable to think that at least some creatures that are different from us—with different brains, say—could nevertheless have states that have the relevant properties in common with us. If this is right, then weak MR is no threat to identity theory, for it is no problem if some creatures not physically identical to us have conscious states as long as they share some properties with us.13 Lawrence Shapiro makes a similar point when he argues that realizers count as genuinely multiple only if they differ in causally relevant ways, “in properties that make a difference in how they contribute to the capacity under investigation” (2000: 644). Recall again the divide and conquer strategy being pursued. We have seen that a range of claims figures in multiple realizability arguments. The stronger claims, I first argued, are not plausible independently of a commitment to functionalism. Next I argued that the Kim–Adams reply may be employed to show that identity theory can accommodate a broader range of cases than is usually supposed—at least those of weak MR. This reply does not guarantee that all creatures capable of consciousness in fact have some property or properties in common. It demonstrates only that Putnam’s intuition does not by itself provide sufficient reason for abandoning identity theory. Applied to weak MR, the Kim–Adams reply shrugs off multiple realizability concerns. On the other hand, by offering the Kim– Adams reply to standard MR or radical MR, the identity theorist digs in: either the thing shares some properties in common or else it does not have mental states after all. But the Kim–Adams reply has not
12 Chapter 1
persuaded many standard MR loyalists, and we might hope that more can be accomplished than merely reporting a clash of intuitions. Rather than insist on the point, let’s see what more the identity theorist can say. The Kim–Adams reply is one part of the total defense of identity theory from multiple realizability arguments, but not the only part. 3 Empathy and Other Species Consider the form of multiple realizability that I am naming after the Search for Extraterrestrial Intelligence project, SETI MR. Let us imagine that some creature that is quite different from us physiologically, even neurophysiologically, could nevertheless be conscious. How different could such a creature be? Suppose for a moment that the Kim–Adams reply only gets us so far; beyond that point the creatures we imagine are so different from human beings that it is hard to see how their mental lives could be based in properties they share in common with us. Notice how much is being supposed. These creatures would have to be very different indeed if they are such that we can rule out— ahead of time—the possibility that their “brains” could have any properties of the same kinds that our brains have. This is much more than Putnam’s intuition alone suggests; but it does seem to be the idea behind the legion of Martians and other aliens that have populated philosophical discourse for some time now. For the sake of argument let us say that there could be such creatures; if so, they are squarely in the range of SETI MR. Explaining how identity theory accommodates SETI MR requires us to look more carefully at what is being claimed by versions of multiple realizability. We should distinguish two ways of interpreting any multiple realizability claim. As formulated above, SETI MR is the thesis that some creatures that are significantly different from us in their composition could be capable of having some conscious states or other. A stronger claim would be that some creatures that are significantly different from us in their composition could be capable of having exactly the same kinds of conscious states that we have; call this the empathetic interpretation of multiple realizability. According to empathetic
13 Minds, Brains, and Multiple Realizability
versions of multiple realizability, other creatures could have mental states that are of exactly the same kinds as our own. The term “empathetic” is perhaps less than perspicuous. I have in mind that if a creature has, for example, pains that are just like our own, we might be able to imaginatively identify with its experiences. If a creature has states that are not like our own, we might not have a basis for imaginative identification. Of course I might in some sense sympathize or empathize with something like a tree, a building, or a nation, something that does not genuinely have conscious states in the sense at stake herein. I do not have a theory of empathy or sympathy to offer. My metaphor is loose, and I hope not too misleading. The point should be clear enough: the multiple realizability arguments require that other creatures can have exactly the same mental states that we do. Without empathetic multiple realizability, there is no argument against identity theory. For multiple realizability to be a problem for identity theory it is not sufficient that some wildly different creature have some conscious state or other; it must be that different creatures can have exactly the same—empathetic—kinds of mental states. So we should take it that empathetic interpretations of multiple realizability are what most philosophers have in mind. This is strange because it runs counter to ordinary intuitions about sensations. Despite what Bill Clinton may say, none of us believes that he truly felt our pain. We expect that there are differences between individuals, even within individuals over time, that make it unlikely that one sensation is ever exactly similar to another. Of course, we generally allow that other people have some sensations of the same kinds that we do. So empathetic multiplicity is not entirely ruled out among human beings. But we expect limits. We do not, in general, believe that all human beings have all, only, and exactly the same kinds of sensations as each other. We say, for example, that artists and musicians—good ones anyhow—see and hear things that the rest of us do not. Late-night television demonstrates the widespread belief that some people (but only a few people) have an extra sense through which they perceive information about the future; presumably those senses are supposed to be accompanied by distinctive “seeing the future” sensations that I do not have.14 More mundane examples
14 Chapter 1
are humans with reduced capacity for kinds of sensations, for example, persons with cortically based color deficiency (Hurvich 1981). When we consider nonhuman animals, we become even more reticent in our empathetic attributions. Many of us believe that at least some nonhuman animals are conscious. We are pretty sure, for example, that domestic dogs have sensations. But are we committed to the belief that dogs have pains and itches of the same kinds as our own? Perhaps; perhaps not. We needn’t stray far from the human animal before it seems to us that, say, although dolphins experience sensations, dolphin sensations are not exactly the same kinds we have. This, famously, is why Thomas Nagel chose bats as his example in his “What Is It Like to Be a Bat?” Bats are similar to us physiologically, but they seem to have quite disparate ways of being in the world: “Bats, although more closely related to us than those other species, nevertheless present a range of activity and a sensory apparatus so different from ours that the problem I want to pose is exceptionally vivid . . .” (1974: 438). Nagel goes on to explain that bats provide a terrestrial example of a “fundamentally alien form of life” and experience. My selection of dolphins, another echolocating mammal, is similarly motivated.15 When we consider creatures with physiologies very different from ours, our eagerness to attribute empathetic conscious states quickly dwindles. It may have seemed to David Lewis that Martians could have sensations like ours. But it does not seem to me that even avian or piscine sensations are quite like ours, although I suppose I am inclined to believe that birds and fish have sensations of some kinds. This might be wrong. But if it is we will have to do much work to discover that other creatures have sensations empathetically like our own. It is certainly not a platitude of common-sense psychological reasoning that all conscious terrestrial animals have experiences that are exactly, empathetically, of the same kinds. And if they do not, then multiple realizability never gets a foothold. For the claim that pain is multiply realized cannot be any more plausible than the claim that pain is had by various creatures to begin with. “According to functionalists, we have vivid intuitions to the effect that mental states are distributed across an extremely broad spectrum of biological
15 Minds, Brains, and Multiple Realizability
species,” Christopher Hill (1991: 45) writes. Probably many creatures have some conscious mental states or other, but I doubt that empathetically the same mental state kinds are common. I contend that we do not attribute empathetically the same mental state kinds to very many creatures—not even to some terrestrial creatures that probably share many properties with us. Standard MR and radical MR are not at all plausible in their empathetic forms. All this is to say that from the position of pretheoretic intuition— remember, that is the position from which generality arguments from multiple realizability claim to operate—it appears that speciesspecific variation of mental states is to be expected, rather than multiple realizability of empathetically the same states. We expect only very similar creatures to have very similar sensations. To the extent that a creature differs from us, so will its experiences. Pain in human beings may not be exactly similar to pain in dogs, dolphins, or Martians; pigeons and turtles with extra photoreceptors might well have color sensations that we do not.16 Species-specific identities are compatible with our intuitions about what things have minds and compatible with the identity theory.17 The move toward species-specific identities is not new. Paul Churchland has advocated this tactic (P. M. Churchland 1979, 1982), as has Berent Enç (1986).18 And from the start David Lewis (1969) and Jaegwon Kim (1972) each defended identity theories from multiple realizability arguments on the grounds that they need to posit only local or species-specific identities: Let us assume that the brain correlate of pain is species-dependent, so that we have generalizations like “Humans are in pain just in case they are in brain state A,” “Canines are in pain just in case they are in brain state B,” and so on. These species-dependent correlations . . . clearly do warrant—at least they are not inconsistent with—the identification of human pains with human brain state A, canine pains with canine brain state B, and so on. . . . There seems no reason to suppose that species-specific identities are insufficient for the identity theory. (Kim 1972, in Block 1980a: 235)
Putnam, recall, takes the identity theory to claim that every creature capable of a certain kind of mental state must also be capable of exactly the same kind of physical brain state, and he asserts that many creatures do in fact have the same kinds of mental states. Lewis
16 Chapter 1
(1969) argues that a more reasonable view is that mental states are identical to distinct states in various kinds of creatures. Likewise, although Terence Horgan does not himself advocate the view, he writes that “[t]he appropriate form of the identity theory is one that also accommodates species-relative multiple realization—despite the fact that multiple-realization considerations are commonly thought to undermine type-type psychophysical identities altogether” (1997: 165). Of course a good deal more needs to be said about how speciesspecific identities are supposed to work, and we’ll return to that topic shortly. But even if local or species-specific identities can be explained, restricting identities to particular species is not a cure-all for identity theory. Differences between members of a species present a case of multiple realizability that has troubled some philosophers. Horgan paints a particularly apocalyptic picture according to which neurobiological anarchy underlies well-behaved psychological kinds; there is indeed trouble for identity theory if mental states are “radically multiply realizable, at the neurobiological level of description, even in humans; indeed, even in individual human beings; indeed, even in individual humans . . . at a single moment” (1993: 308). And John Bickle notes, “Surely this much domain specificity is inconsistent with the assumed generality of science” (1998: 124). Individual differences and intraindividual differences do not tolerate appeals to species-specificity; we cannot have species of one. Even William Bechtel and Jennifer Mundale, who do not consider intraspecies differences to be a difficulty for identity theory, note that the fan of multiple realizability will find variation to be ubiquitous: “[I]t is well known that even within a species brains differ. Even within an individual over time there are differences (neurons die, connections are lost, etc.). Thus multiple realizability seems to arise within species (including our own) and even within individuals” (Bechtel and Mundale 1999: 177). But identity theorists do not need to invoke yet more localization within species and individuals; we have another tool. Speciesspecificity is one part of the divide and conquer strategy. Some mental states are not multiply realized because they are speciesspecific. But we can also unify. Some states are not multiply realized
17 Minds, Brains, and Multiple Realizability
because all their purportedly various instances in fact share properties in common. That is, we may return to the Kim–Adams reply—it is quite effective for cases of within-species and within-individual differences, for it is quite likely that members of species share properties with one another. Those who are concerned about individual and intraindividual differences seem to have in mind two sorts of examples. On the one hand, there is a myriad of near-miraculous case studies wherein persons with massive neuropathologies nevertheless live normal or near-normal lives. These include the recovery of stroke victims, as well as more amazing cases of individuals who have only a small amount of viable cortex. On the other hand, there is the ordinary variance in neural activity over time, such as the well-known changes to motor cortex depending on recent use.19 Many of us are familiar with diagrams in psychology textbooks that illustrate changes in primate cortex mapped to the fingers caused by exercise or amputation of the fingers (figure 1.1). Of course these are only a tiny fraction of cases. Most subjects who suffer massive neurological traumas do not have uninhibited behavior, and much neural function appears to be reasonably localized and stable over time. Surprising case studies are, well—surprising. But they do not show that exactly the same mental state is multiply realized, nor that similar mental states have wildly different realizations. Rather than supporting multiple realizability, these cases suggest that we do not understand very well how the brain works—how to individuate brain processes, events, states, and properties. The considerations we have been entertaining so far leave plenty of room for identity theory to accommodate Putnam’s intuition. If functionalism is the correct theory of mental states, radical MR and standard MR are not unreasonable; but they are implausible, independent of a commitment to functionalism. So even if it is true that identity theory does not permit that degree of variability in realizers of mental states, it is no defect of the theory. Although I have contended that few creatures have empathetically the same mental states that we have, if they do then the Kim–Adams reply reminds us that even biologically diverse realizing systems can share some properties with one another. Mental state kinds are probably as widespread as
18 Chapter 1
Figure 1.1 Cortical plasticity. A portion of the cortex of an owl monkey is organized as a somatosensory “map” of the hand in which each finger is represented by a distinct area of tissue (A). Within weeks following the amputation of the middle finger the map reorganizes (B) and the area that previously represented the amputated digit is now “fully partitioned by areas representing the two adjacent digits” (Fox 1984: 821). Reprinted with permission from J. Fox, “The brain’s dynamic way of keeping in touch,” Science 225 (1984): 820–821. Copyright 1984 American Association for the Advancement of Science.
weak MR and moderate forms of SETI MR contend. But that is no problem for identity theory, because the Kim–Adams reply explains how identity claims can cover more creatures than is typically supposed. In those cases where identity theory declines to attribute a kind of sensation to members of a species, it does so because the creatures in question are sufficiently different to make it improbable that they share relevant properties with us. But those are just the cases for which our pretheoretic intuitions tell us that it is unlikely they share
19 Minds, Brains, and Multiple Realizability
exactly similar conscious state kinds with us. Species-specific identities are what we would expect if mental states are widespread in the ways that weak MR and SETI MR contend: Alien beings may have conscious states of a sort; and those are identical to some of their physical states, just as our conscious states are identical to some of our physical states. Identity theory thus accommodates multiple realizability claims to just the extent we should want. And this is just what we should have expected all along. Consider everybody’s favorite example, a carburetor. Let us remind ourselves a thing or two about carburetors (see figure 1.2). Carburetors mix air and fuel, and control the air-to-fuel ratio, in combustion engines such as those found in many commercial automobiles. Liquid gasoline must be vaporized before it is introduced into the combustion chamber of an engine. Fuel is drawn into the carburetor where a partial vacuum and the temperature of the intake manifold cause the fuel to vaporize. The continuous partial vacuum, that is, the lowered pressure, in the intake manifold while the engine is running is the key to the operation of a carburetor. One cause of the vacuum is the withdrawal of the piston caused by combustion. In addition: “To make sure that there is always enough vacuum at the carburetor to draw fuel out of the float bowl, the carburetor throat narrows down and then opens up again to form a shape known as venturi. As air passes through a venturi, air speed increases and pressure drops. A vent at the top of the float bowl allows air at atmospheric pressure to push fuel through a passage and into the low pressure area of the carburetor throat at the venturi” (Reader’s Digest 1981). To the best of my knowledge this is more detail about how carburetors work than has ever been explained in a philosophical argument. (So I will skip the nuances of throttle plates, by which the vacuum in the venturi is made to regulate the air-to-fuel ratio.) Are carburetors examples of multiply realized kinds? Of course. There are many manufacturers of carburetors, which are made of various materials and of various designs. Some have more than one venturi, for example. Carburetors allow weak MR. Depending on what you think about the differences between carburetors that have single or multiple venturi, perhaps they allow SETI MR. But there is
20 Chapter 1
Figure 1.2 A simple carburetor. Reprinted with permission from Complete Car Care Manual, © 1981 The Reader’s Digest Association, Inc., Pleasantville, N.Y., www.rd.com. Illustration by Dale Gustafson.
a single structural feature that all carburetors have: they all have venturi. If some device for mixing air and fuel does not work by the venturi principle, if it does not have venturi, then it is not a carburetor. In particular, fuel injectors are not fancy carburetors; they do not have venturi.20
21 Minds, Brains, and Multiple Realizability
4 Granularity and the Evidence for Multiple Realizability Both the Kim–Adams reply and Kim’s and Lewis’s defenses of speciesspecificity have been part of identity theorists’ arsenal for some time. Yet the idea that multiple realizability is a problem for identity theory persists. It would be nice if we had an account that explains the widespread dedication to forms of multiple realizability, while demonstrating that they are no threat to identity theory. Recently William Bechtel, Jennifer Mundale, and Robert McCauley have given just such an account. They suggest that multiple realizability arguments against identity theory depend on misunderstanding the extant neurophysiological evidence. Bechtel and Mundale (1999) are particularly concerned to show that the facts of multiple realizability do not undercut the potential contributions of neuroscience to the understanding of cognition, while Bechtel and McCauley (1999) employ similar considerations to defend a version of identity theory itself. The multiple realizability that Bechtel and his collaborators have in mind is standard MR. In a telling passage, Bechtel and Mundale write: The claim of multiple realizability is the claim that the same psychological state can be realized in different brain states. . . . Taking this a step further, many philosophers became convinced that the same mental activities could be realized in the brains of aliens with radically different composition from ours. The upshot of these speculations about artificial and alien minds is a metaphysical claim that mental processes are operations themselves, and are not identified with whatever biological or other substances realize them. (1999: 176)
Although this passage begins by stating a version of weak MR, it immediately admits that systems of radically different composition may be taken to have minds—which is SETI MR. Then, by attributing to multiple realizability the consequence that minds can be identified independently of “whatever biological or other substances realize them,” Bechtel and Mundale indicate that they are concerned with full-blooded standard MR. A common element in the arguments advanced by Bechtel and his collaborators is that the claims of multiple realizability depend on an equivocation of the “grain” of mental states and brain states. The
22 Chapter 1
grain of a mental state or brain state is the degree of generality with which it is identified. The taste of chocolate is presumably a finergrained sensation than the taste of sweetness, for example.21 Bechtel and Mundale diagnose that multiple realizability arguments against identity theory depend on an equivocation of granularity. If we individuate mental state kinds coarsely it might seem that biologically diverse creatures are capable of sharing psychological states. Putnam seems to have in mind relatively coarsegrained states, for he offers only that psychological laws “of a sufficiently general kind” will be species independent (1967, in Putnam 1975c: 437). But, according to Bechtel and Mundale, in continuing on to claim that the creatures under consideration are biologically diverse, Putnam is illicitly contrasting coarse-grained psychological states with fine-grained brain states: A human’s psychological state and that of an octopus might well be counted as the same insofar as they are associated with some general feature (such as food-seeking behavior, in the case of hunger). But with respect to other considerations, a human psychological state may be considered different from that of an octopus, even if we limit the scope to mere behavior. . . . the assertion that what we broadly call “hunger” is the same psychological state when instanced in humans and octopi has apparently been widely and easily accepted without specifying the context for judging sameness. (Bechtel and Mundale 1999: 203)
If this is right, the appeal of Putnam’s intuition is revealed to be an artifact of our failure to fix our mind–brain comparisons at a common granularity. It may seem that mental state kinds cannot be identical to brain state kinds if we are not considering states of matching granularity. What is the appropriate grain size for sensations and brain processes? The suggestion is that there is no one answer; sensations and brain processes may each be considered with finer or coarser grain. Functionalists and identity theorists can agree that mental states occur at multiple granularities, some at one grain and others at other grains. The problem of determining the appropriate granularity for the brain state associated with a given sensation is part of what George Graham and Terence Horgan call the “grain project”: “Here is a large-scale, long-term, but potentially empirically tractable
23 Minds, Brains, and Multiple Realizability
project for cognitive science: to identify the specific causal role or roles associated with phenomenal states—to identify what, in some sense, those states do. . . . Let us call this scientific program the grain project, since it involves investigating the causal roles associated with phenomenal consciousness at several levels of detail or resolution” (2002: 65).22 Graham and Horgan suggest that a good theory of consciousness (“phenomenal states”) will integrate explanations across the various grains or levels. This is just what William Lycan (1987) argues for functionalism; and there is no reason to suppose that identity theory cannot follow suit. But whereas the functionalist will theorize at multiple functional grains, the identity theorist will operate at multiple biological grains. At this point we are in a position to address the common complaint that, although the Kim–Adams reply and species-specificity could salvage identity theory philosophically, it is nevertheless empirically false. The idea seems to be that even if there are no current cases of multiple realizability, we expect neuroscientists to discover some any minute now. Bechtel and Mundale attribute the stubborn persistence of this belief, also, to confusions about granularity—in effect maintaining as Kim and Adams do that mental state realizers do share some properties. Bechtel and his collaborators point out that the advocates of multiple realizability seem to have ignored the extent to which neuroscientific research already depends on identifying morphologically similar and homologous brain structures across individuals and species. Simply put, the methods of presentday neurosciences depend on our ability to identify common brain structures across reasonably various species of creatures: “Historically . . . neuroscientific practice routinely involved identifying brain areas and processes across a broad range of species as belonging to the same type. These practices continue now. . . . Oddly, when they consider theories of mind–brain relations, philosophers seem to lose sight of the fact that the overwhelming majority of studies have been done on non-human brains” (Bechtel and McCauley 1999: 3–4). When the “same” mental state seems to be identifiable with diverse brain states, neuroscientists typically conclude that they are actually dealing with more than one mental state; they then proceed to look for psychological differences between the mental states.
24 Chapter 1
Bechtel and colleagues survey an array of classic and contemporary studies using comparative neurobiology to identify brain areas by their anatomy or function, concluding that neuroscientific practice and data support mind–brain identity claims, sometimes across species. For example, they observe that Brodmann’s (1909) groundbreaking brain mapping is both cytoarchitectonic and comparative (figure 1.3). Brodmann used cellular staining techniques to distinguish brain areas according to the type and density of neurons. And Brodmann applied his technique not only to humans but to other species as well: “[I]n defending the claim of six different layers in the cortex, he reports preparations made from 55 species ranging over 11 different order of mammals. . . . When he turns to mapping the cortex, Brodmann presents maps for several other species in addition to humans: two lower monkeys (guenon and marmoset), lemur, flying fox, kinkajou, rabbit, ground squirrel, and hedgehog, using the same numbering systems to identify homologous areas in the different species” (Bechtel and Mundale 1999: 180). Bechtel his collaborators’ examples are not decisive, of course, but they suggest that that current and historical neurobiological data and practice do not foreclose the possibility of an identity theory.23 Putnam was too quick to assert that there is neuroscientific support for multiple realizability. Current evidence does not reveal multiple realizability, and there is some reason to think the enterprises of neuroscientific investigation are premised on the hypothesis of brainstate identities. This is exciting empirical work. Does the evidence go so far as to support identity theory? Bechtel and Mundale defer the question: “For the most part, we will have nothing to say about these speculative arguments [multiple realizability arguments against mind–brain identity], nor are we primarily concerned with the metaphysical claim” (1999: 176).24 Although I am concerned with the metaphysical claim, I share Bechtel and Mundale’s hesitation to think that available evidence supports identity theory over functionalism. I doubt that the evidence is even suggestive in this respect, because I doubt that evidence by itself can settle metaphysical questions such as that between functionalism and identity theory. What Bechtel and his collaborators provide is nevertheless quite useful. They give us a
25 Minds, Brains, and Multiple Realizability
Figure 1.3 Brodmann’s brain “maps.” (Brodmann 1909. Used by permission.) (A) Map of human cortex (lateral view). (B) Map of marmoset cortex (lateral view). As Bechtel and Mundale note, “Brodmann adapted the same numbering system as he used on the human brain for each species that he studied. In some species areas identified in the human brain were not identifiable, but each region of the cortex that was identified in a given species was associated with a region of the human cortex” (1999: 181).
26 Chapter 1
way of understanding the lingering appeal of multiple realizability claims, even those as strong as standard MR; and they also give us the tools to deflect those concerns. 5 Species-Specific Identities and the Generality of Psychology We now to return to the varieties of multiple realizability that I claim are compatible with identity theory. We need some assurance that the identities that we can get from such a theory are worthwhile. The Kim–Adams reply to multiple realizability maintains that different mental states share properties after all. Mental states are multiply realizable, but mental properties are general. At the end of the day, this may prove true and thus be the identity theorists’ most powerful defense. However, I have taken a more modest stance, conceding for the moment that the Kim–Adams reply may have its limits; and I have argued that they are just the limits we expect. When a creature is so different that it cannot have states with the properties that our neural states have, then the Kim–Adams reply gives way and our identity claims are localized to each species. How much difference must be involved is not clear at this time. Perhaps, as Putnam thought, octopi are different enough. How, then, does species-specific identity work? Lewis advocates the view that the same kind of mental state can be realized by various kinds of physical states in different creatures: “[Putnam] imagines the brain-state theorist to claim that all organisms in pain—be they mollusks, Martians, machines, or what have you—are in some single common nondisjunctive physical-chemical brain state. Given the diversity of organisms, this claim is incredible. But the brain-state theorist who makes it is a straw man. A reasonable brain-state theorist would anticipate that pain might well be one brain state in the case of men, and some other brain state in the case of mollusks” (1969, in Block 1980a: 233). Lewis thus holds that psychological kinds occur cross-species but that realizers of psychological kinds are speciesspecific. This is the disjunctive view; it identifies each mental kind with a disjunction of physical kinds (figure 1.4). David Lewis and David Armstrong have defended the disjunctive identity view. However, the disjunctive formulation is susceptible to the charge that it equivocates about the granularity at which mental states and
brain state kinds
B1 ...
Mα Bi ... Bn
a.
Mα
mental state concept
Sa ...
Si
... Sn
species-specific kinds of sensations
B1 ...
Bi
... Bn
brain state kinds
b.
Figure 1.4 Two views of multiple realizability. On the traditional view of multiple realizability (a), a mental state kind, Ma, is realized by various brain state kinds, B1, . . . , Bn. On Lewis’s view (b), a single mental state kind concept, say, pain, is realized by different kinds of sensations, S1, . . . , Sn, in different creatures: pain in humans, pain in octopuses, pain in Martians, etc. Each of these sensation kinds falls under the single mental state concept (Ma), and each sensation kind (S1, . . . , Sn) is identical to a species-specific brain state kind (B1, . . . , Bn).
27 Minds, Brains, and Multiple Realizability
mental state kind
28 Chapter 1
physical states are individuated. If the realizers, considered at the appropriate grain, are diverse, then it is a mystery why, say, X in humans and X in dogs should be counted as the same kind of mental state. A better way of construing Lewis’s intuition is to say that X in humans and X in dogs are fined-grained distinctions within a coarsegrained mental state kind X. Thus, as Kim and Adams urge, they share some coarse-grained property or properties after all, despite being diverse fine-grained states (see figure 1.4). An alternative is to hold that the fine-grained sensation X in humans is different from X in dogs or X in Martians, and that each is identical to a distinct type of physical state in the respective kinds of creatures. That is, species-specificity is both psychological and biological. This is the picture suggested by Kim (1972, 1998), Flanagan (1992), and Bechtel and Mundale (1999), and I contend that it is the sort of species-specificity supported by Putnam’s basic multiple realizability intuition. We think that there could be conscious aliens that differ from us not only in their biology, but in the kinds of mental states they have: “If physical realizations of psychological properties are a ‘wildly heterogeneous’ and ‘unsystematic’ lot, psychological theory itself must be realized by an equally heterogeneous and unsystematic lot of physical theories” (Kim 1993: 328–339). The trouble with species-specificity, as I admit from the outset, is that it sacrifices generality. How, then, do we explain what X in humans and X in dogs have in common? According to the view now being considered (that psychological states are themselves speciesspecific), the answer is that they do not have anything in common. X in humans, X in dogs, and X in Martians are not psychologically alike; they are as psychologically and biologically distinct as any three types of sensation. Perhaps we group them because they are typically caused by similar stimuli. Resolving this tension appears to be Lewis’s motivation in “Mad Pain and Martian Pain” (1980). Both Lewis and Kim turn to a kind of functionalism to explain what pain in humans and pain in dogs have in common. For this reason Marian David (1997) dubs Kim’s view “conceptualized functionalism.” If we accept species-specificity of this second sort, then we must admit that psychology is not completely general. There could be minded conscious beings to whom psychology does not apply. But
29 Minds, Brains, and Multiple Realizability
this should not particularly bother anyone: “If Fodor and Putnam were committed to a science of all possible perception, there is no reason to believe they were right” (Block 1997: 129). Likewise, if there are life-forms that are not carbon-based, then perhaps terrestrial biology is not perfectly general. No surprise. It might be thought that denying that psychology is perfectly general opens the door to additional worries, such as those behind Donald Davidson’s (1970) anomolous monism. Davidson denies that mental states are nomologically regular, that is, that there are general psychophysical laws. But with any identity theory there are lawlike psychological correlations and lawlike psychophysical correlations, both of which Davidson denies.25 Moreover, the generalizations of the species-specific identity theory are universal in scope; the mental states that figure in the theory are just not as frequent as some have thought. So Davidson’s concerns are not sparked by species-specific psychological kinds, nor do they assist multiple realizability arguments against identity theory. Notice, in fact, that anomalous monism is no friend to multiple realizability. Philosophers have tended to focus on Davidson’s denial of lawlike psychophysical regularities. That is the aspect that makes his positive suggestion, anomolous monism, tantalizing; and it has seemed to rule out the possibility that psychological kinds could be “reduced to” neurobiological kinds. But Davidson also denies that there are lawlike psychological generalizations. The multiple realizability argument doesn’t even get off the ground unless there are genuine psychological kinds that might have diverse realizers. If there are no lawlike psychological generalizations, then what is the basis for psychological kinds? If there are no psychological kinds, then there are no multiply realized psychological kinds. 6 Multiple Realizability for Identity Theory I have argued that Putnam’s intuition that mental states are multiply realizable does not show that identity theory is known to be false or that it makes improbably strong empirical claims. We need only to distinguish the different intuitions that might pass for intuitions of multiple realizability. Standard MR and radical MR are themselves
30 Chapter 1
implausible. Even so, they do not rule out identity claims because, as Kim and Adams argue, the purportedly various realizing systems may share properties. Within individuals and between individuals of the same or similar species, it is quite likely that realizers have the same or similar properties. Thus cases of weak MR and SETI MR can be accommodated within identity theory. And the evidence does not warrant supporting the stronger claims of standard MR and radical MR. It tends to suggest, rather, that kinds of psychological states are themselves as species-specific as their realizers. When creatures are quite different from us physiologically, we may think that they have some conscious states or other but we do not suppose that they have empathetically the same kinds of conscious states that we have. The view of multiple realizability I am advocating is illustrated in figure 1.5. Particular kinds of sensations, S1, . . . , Sr, are identical to particular kinds of brain states, B1, . . . , Br. Sensation kinds may cluster into coarser, more general species-specific mental state kinds (e.g., Ma or Mb), but insofar as they do, we expect that their members will share physical properties (e.g., Pa).26 Creatures that are similar physically, those that can have Pa or Pb for example, may also have relatively similar mental state kinds, Ma or Mb, say. We should expect human beings and higher primates to have similar conscious mental states because their brains are quite similar to our own. And we should expect the experiences of octopi or aliens to be different from ours to the extent that their brains are quite different from our own. This sort of species-specific identity has sometimes been thought to block the “reduction” of psychological theories to neurobiological theories, traditionally understood as the translation of psychological theories into neurobiological theories. If psychology can be “reduced” to neuroscience then this would show that identity theory is true by showing that the terms that name psychological kinds in fact refer to neuroscientific kinds. But if psychological kinds are multiply realized, the argument goes, then they do not pick out homogeneous neurobiological kinds and theory “reduction” is not possible, leaving us no reason to assert mind–brain identities.27 To this worry a number of responses are available. One might argue, as Paul Churchland (1982) has, that although biological realizers are
qualitative similarity
physiological similarity
ranges of kinds of sensations
Mγ
Sa Sb
...
Si
Sn
...
Sq Sr
Ba Bb
...
Bi
Bn
...
Bq Br
Pα
Pβ
ranges of shared properties
Pγ
Figure 1.5 Multiple realizability for identity theories: an identity theorist’s view of multiple realizability. Similar kinds of sensations are identical to similar kinds of brain states. Very different sensations are identical to very different brain states.
31 Minds, Brains, and Multiple Realizability
Mβ
Mα
32 Chapter 1
diverse, it remains possible that they are alike at subbiological levels. Jaegwon Kim (1989) and Berent Enç (1983) have argued that the local “reductions” that would be licensed by species-specific identities are all that we get in other classic cases of theory “reduction” in the physical sciences, so they are all we should expect for psychology. Another response is to argue, as John Bickle (1998, 2003) has, that the multiple realizability argument depends on an outdated notion of “reduction” that should be replaced by a “new wave” alternative focusing on the relations between theories rather than the relations between ontological levels. But the identity theorist is not forced to deploy any of those strategies. Belief in psycho–neural “reduction” is one reason—admittedly, a common one—to embrace identity theory, but it is not the only reason. Bechtel and McCauley, for example, press for what they call a “heuristic identity theory,” according to which mind–brain identities are posits that neuroscientists make in the course of formulating theories: “Emphasizing the thoroughly hypothetical character of all identity claims in science, [heuristic identity theory] focuses on the way that proposed identifications of psychological and neural processes and structures contribute to the integration and improvement of our neurobiological and psychological knowledge” (1999: 71). On this view, the search for mind–brain identities is a regulative ideal in the neurosciences. But the heuristic aspect of Bechtel and McCauley’s proposal threatens to undermine the support it may give identity theory. A functionalist might be happy to admit that the practices of neuroscientists involve interim hypotheses of structural identity, certainly within similar species. Nevertheless, the functionalist may suppose, heuristic identities are, in the course of theory building, replaced by more sophisticated realization relations. Bechtel and McCauley are correct that the justification of identity claims does not await seeing whether completed psychology “reduces” to completed neuroscience. Likewise, John Perry (2001) adopts an identity theory that does not depend on either the success or promise of explanatory or theoretical “reduction” as part of his “antecedent physicalism.” On my view, identities are not intertheoretical devices, but rather figure in the theories themselves. They do not have to be heuristic or provisional or antecedent; they may figure as inductive conclusions, the result of the usual practices of infer-
33 Minds, Brains, and Multiple Realizability
ence to the best explanation. Lewis (1970, 1972) argues that scientific identities are implied by the theories that posit them. Of course, Lewis believes that theories entail identities by giving the meanings of the theoretical terms and thus fixing the nominal essences of their objects. But Lewis’s is not the only model of theories that implies identities. This is also be the case of theories that pick out their objects by their real essences; in this case, theories work by fixing the reference of their terms. Either way, identities are what we expect from other sciences; and we should expect mind–brain identities as well. Since most scientific identities are species-specific, we should expect mind–brain identities to be localized. If it seems that we will not find the identities we are looking for, then we will face a hard choice. We might give up our identity claims; but we might decide to adjust the way we individuate the kinds involved in order to permit identities. The latter will be quite likely if we are only trying to accommodate a few resistant cases. 7 Eccentric Minds On my picture, some conscious creatures could be different from us to such a degree that we must say that they have kinds of mental states only remotely similar to those that we have, even when mental states are coarsely individuated. Creatures that have very different physiological properties (say, those that have Pg may have sensations [Sq, . . . , Sr] but they will be of wholly different kinds, like Mg; see figure 1.5). The identity theory does not rule out such cases. What this suggests is that at some point the Kim–Adams reply must give way to species-specificity of the sort described by Kim, and by Bechtel and his collaborators. To accept this sort of species-specificity is to accept the possibility of what we might call eccentric mental states that are of physically and qualitatively different kinds than our own.28 The possibility that some creatures have such eccentric conscious states is entirely compatible with identity theory. Their eccentric mental states are identical to their eccentric physical states just as our mental states are identical to states of our brains. Eccentric experiences raise obvious epistemic and methodological problems for how we might detect that these creatures are conscious, or be justified in attributing mental states to them. When creatures
34 Chapter 1
are physiologically like us, we can attribute to them sensations like those that we have. Insofar as a creature is not like us, it does not have those sensations. But now we are considering creatures that are quite unlike us. On what basis do we attribute conscious states to eccentric creatures, and on what basis do we rule out the possibility that rocks and thermostats are not examples of such creatures? The answer is that the identity theorist has available all the tools that any other theory makes use of: functional organization, history, behavior, and the like. Other theories take these characteristics as definitive of mentality and therefore can say with certainty of any creature that it is or is not conscious, or better, to what degree it is conscious. But if we demand certainty, we cannot have it; for an identity theory takes these standard characteristics as only defeasible evidence. The identity theory may have to concede that there could be creatures about which we cannot know with certainty whether they are conscious. But that does not mean that we have nothing at all to say. Conscious states are related to one another; they have structure. The sensation of yellow is both similar to and different from the sensation of green in ways that neither is related to the taste of Château Haut-Brion. Might there be eccentric colors, colors not related by similarity to the colors we experience? What would it mean for there to be a color that was not related to yellow and green? We do not know what to say (Hardin 1988). The view I am advocating is not Colin McGinn’s (1991) view, which Flanagan (1991, 1992) dubs “new mysterianism,” according to which we are not cognitively equipped to understand consciousness. Rather, we are simply not situated to know with certainty whether some creatures are conscious, in much the way that physics holds that we are not situated to know what is going on outside our light cone. There is nothing mysterious about either of these limits. (In fact, there is something suspicious about a theory that has no boundaries.) Some might think it is a conceit to believe that our form of consciousness is even this unique, however trivial that distinction. Far greater the hubris of supposing that we just happen to have a characteristic more cosmically ubiquitous than any other we know of. Even the so-called laws of physics change in the limit, when things are very big, or very small, or very fast. It should not be troubling
35 Minds, Brains, and Multiple Realizability
that mental categories—those of human psychology—break down when we try to apply them to creatures wildly different from ourselves. 8 Why Identity Theory? I’ve been making the case that identity theories are consistent with the available neuroscientific evidence. But some will argue that there is no motivation for holding an identity theory. Daniel Dennett, for example, has raised the concern that there is no motivation for adopting such a theory (2000: 382). If that were correct, it could quash any gains I might have made in arguing that identity theory can handle the generality arguments based on multiple realizability. Identities, of course, are not deductively entailed by the evidence. Empirical experiments will never record more than correlations; and correlation is in some sense a weaker claim than identity. All identities are correlations, but not all correlations are identities. If correlations do all the predictive and explanatory work and are all that we can directly observe, then the additional posit of identity will seem not only explanatorily superfluous but bought on credit of metaphysical faith that can never be repaid. To distinguish identity from a weaker realization claim, on this view, is to insist on a difference that makes no difference. The most obvious problem with this objection is that identities are not explanatorily empty. Identity dissolves, with brutal simplicity, a number of problems that are otherwise quite recalcitrant. One, mentioned earlier, is the problem of mental causation, which never arises for an identity theory. If mental states are identical to physical states, then mental states ipso facto have the causal powers of physical states. And we shall see that puzzles such as the explanatory gap and zombies readily yield to identity theory. This should not be a controversial claim; it is widely believed that identity theory is a powerful theory that, lamentably, happens to be false. So how it is that mind–brain identities came to be explanatorily suspect is itself a bit obscure. We typically arrive at identities by the practice of inference to the best explanation. From the point of view of explanatory practice, to
36 Chapter 1
refuse to accept identities on the grounds that there could be some possible creature for whom the correlation fails amounts to no more than a form of skepticism. We can meet it as we would meet the skeptic, demanding that we be given some explanation for why we should doubt this or that particular identity. My point is not that we should ignore the modal obligations of necessity claims; I take those quite seriously. Rather, when a theory that is supported by the evidence implies identities, we should not suddenly develop an aversion to the metaphysical claims and try to hedge by embracing only correlations. The naturalized metaphysician takes seriously the metaphysical commitments of our explanatory practices. There is an even tougher response open to the identity theorist: Perhaps we can make all the same predictions on the basis of correlation alone. But prediction and explanation are different matters. So we might say that a correlation-based psychology would explain nothing at all; rather, it would itself stand in need of explanation. There may be good reason to admit psychological kinds that are identified with functional kinds.29 But one cannot claim that identity theory is a more ontologically extravagant theory than every nonidentity theory. Identity theory posits one kind of process, event, state, or property; other accounts require multiplication of at least some of those—of processes, events, states, or properties.30 What, then, of the metaphysical credit with which identity theory is purchased? Yes, identity is a stronger relation than correlation. But identity is not a new or unusual kind of relation, and by invoking identity we are not admitting any new relations into our ontology. Moreover, by insisting on mere correlations we must add rogue psychological kinds to the “furniture of the universe” along with an puzzling new relation, functional realization. This has the effect of generating a new problem, the problem of explaining why the correlations hold, a problem that identity theory avoids. Functionalists will argue that the mind–brain correlations hold because brains realize minds. Understanding the realization relation is one of the goals of this book, and I will take up the problem directly in chapters 4 and 5. Whatever the realization relation turns out to be, we will want to know why it holds. This is the metaphysical “open question” for the mind–body problem. Adopting the terms that Kim now
37 Minds, Brains, and Multiple Realizability
uses to discuss supervenience, we should demand an explanation of the realization relation that is “metaphysically deep” (Kim 1998). But this question makes no sense when the relation is identity. Identity holds of necessity; no further explanation is required or available. So the identity theory circumvents a burden that every other account must shoulder, whether naturalistic or nonnaturalistic. This claim is widely but not universally accepted. David Chalmers (1995, 1996a) complains that identity is brute, and therefore that identity theories fail to explain the mind–body relation. Similarly, Chalmers and Jackson (2001), responding to Block and Stalnaker (1999), argue that “identities are ontologically primitive, but they are not epistemically primitive” (Chalmers and Jackson 2001: 354). As I understand them, Chalmers and Jackson are calling for a conceptual explanation (they say “reductive”) of the identity, namely, an analysis that would allow, for example, claims about conscious states to be deduced a priori from claims about physical states alone without use of any additional principles such as identity claims. If this cannot be done then the identity claim is on no surer footing than a property dualist’s psychophysical laws, for a posteriori identity claims are epistemologically no different than very strong correlation claims: “this view may preserve the ontological structure of materialism. But the explanatory structure of this materialist view is just like the explanatory structure of property dualism” (ibid.). Thus Chalmers and Jackson urge that one must either adopt analytic materialism (Jackson unpublished) or property dualism (Chalmers 1996a). This line of argument leads us into a territory outside the scope of this book. Since I have limited myself to advocating identity theory in favor of other naturalistic theories, I will leave for another day the questions posed by property dualism for naturalism generally. Chalmers’s view, it seems, is that a nonanalytic identity theory is no better off than other varieties of materialism; even if that is correct, I will be satisfied for now to show that such a theory is at least not any worse off. Nevertheless, identity theory as I construe it resembles a theory that Chalmers and Jackson mention briefly. According to this possible account, mental states are ontologically “reduced” to (i.e., identified with, I assume) physical states, but those identities do not
38 Chapter 1
support “transparent reductive explanation” of conscious-state facts/claims by physical facts/claims (i.e., deduction of the former from the latter without bridge principles). Although there is more to be said, the view that I advocate has those characteristics. (I will come to the second aspect in chapter 6.) About such a theory Chalmers and Jackson offer: “With this sort of theory, as with a property dualist theory, the explanatory gap between physical and phenomenal might be bridged, but it would not be closed” (2001: 356).31 Yet I believe that the identity theory fares better against the explanatory gap (Levine 1983) than Chalmers and Jackson suppose. This is because the identity theory can avoid the metaphysical contingencies that trigger the gap reasoning. It is this issue that we now consider. I have argued that the identity theory can answer multiple realizability concerns. This is remarkable because multiple realizability is the most widely mobilized argument against identity theories and is often thought to be decisive. If I am right, then identity theory can overcome one of the hurdles that was supposed to reveal it to be inadequate. Now we can turn our attention to the second traditional obstacle to identity theories: Saul Kripke’s modal argument against materialism.
2 The Illusion of Contingent Identity
In the previous chapter I argued that the identity theory can accommodate an appropriate degree of multiple realizability. To do so I employed several tactics, individually and in combination. These were more or less familiar arguments; there has always been resistance to the claim that multiple realizability is incompatible with identity theory. If I am right, I’ve explained how the identity theory can overcome one hurdle. But only one. Multiple realizability arguments against the identity theory, while perhaps the most widely discussed, are not the most challenging. The reason is straightforward: multiple realizability arguments purport to establish only that minds are not (in fact) identical to brains.1 But other arguments are more ambitious; they attempt to conclude not only that minds are (in fact) not identical to brains, but moreover that minds could not possibly be identical to brains. Saul Kripke’s essentialist argument is one such. Kripke’s argument is aimed at materialism in general, not just at identity theories. But it is most effective, and most worrisome, for identity theorists. The standard way for materialists or naturalists to avoid Kripke’s concerns is to deny that his argument applies to the theory in question. In particular, Kripke assumes that the target theories use their key terms to refer “rigidly,” that they refer to the same objects or kinds in all possible worlds (1971, 1972/1980). For reasons that will become clear, the functionalist is able to resist this assumption while the identity theorist is not. Identity theory makes just the sorts of claims that Kripke’s argument is directed against. Thus the standard functionalist maneuver for escaping Kripke’s argument is
40 Chapter 2
not available (at least not readily so) to identity theorists. So although Kripke’s argument is aimed at materialism in general, it presents a special challenge to identity theorists. The present chapter makes room for an identity theory of conscious mental state types that specifies identities using rigid designators and does so without dismissing Kripke’s analytic tools. 1 Kripke’s Antimaterialist Argument One result of Kripke’s overall arguments in Naming and Necessity and elsewhere was the understanding that proper names, natural kind terms, and the like are used (typically) to designate their referents rigidly. That is to say, they pick out the same thing, or things, or kind of things in all possible worlds. Identities between rigidly designated things or kinds are necessary. The reasoning is fairly straightforward: Since any thing is necessarily identical to itself, if two terms pick out the same things rigidly (i.e., in all possible worlds) then if they are identical in any possible world they are identical in all possible worlds. Thus, identity between things rigidly designated is necessary. A hackneyed example is of the identity claim, “water is H2O.” In such a scientific identity claim, Kripke argues, the two terms “water” and “H2O” designate rigidly. So the identity claim is, if true, necessarily true. Water is itself in every possible world; water by any other name is itself in any possible world. Of course we didn’t always know that water is H2O. A revolutionary consequence of Kripke’s view is that we can discover necessary identities by empirical means, that there can be necessary a posteriori truths. Before it was discovered that water is H2O, it did not seem to anyone that water is necessarily the molecular kind H2O. Many ancient philosophers believed that water was an element—Thales, famously, thought it was the element. And even among molecular substances, it might have seemed at one time that water could have been some other molecule—perhaps Putnam’s mysterious XYZ. If the relation between two rigidly designated things or kinds is contingent, then that relation cannot be identity. (This is merely modus tollens on the argument concluding that identity between rigidly designated objects is necessary.) So to maintain that scientific identities are
41 The Illusion of Contingent Identity
necessary, Kripke needs a way to explain away the apparent contingency. The explanation, he argues, is that we can be in an epistemic position vis-à-vis a thing or kind of thing such that we mistakenly believe that it could be other than it is. We thus mislocate the apparent contingency. In the case of water and H2O, we might have thought that water seems only contingently to be H2O. But the apparent contingency of the identity is to be explained by the actual contingency that H2O is the stuff with which we are familiar. We are to understand our belief as something like, “the clear, cool, thirst-quenching liquid found in the rivers and lakes, that rains from the sky, that we call ‘water’ . . . might not have turned out to be H2O.” But this, Kripke points out, is not a case where water fails to be identical to the chemical kind H2O. Rather, it is a case in which the clear, cool, thirst-quenching liquid found in the rivers and lakes, that rains from the sky, that we call “water” . . . is not water. The apparent contingency is a result of the fact that something other than water could have the superficial characteristics—the appearance—of water. Still, water is H2O necessarily. This is Kripke’s epistemic model of explaining away the apparent contingency of scientific identities. The epistemic model also works with Kripke’s other favorite example, borrowed from Frege, of Hesperus and Phosphorus. It seems plausible that ancient astronomers believed that they observed a different object last in the sky in the morning (Phosphorus) and first in the sky at night (Hesperus). It has since been discovered that Hesperus is Phosphorus—and it is not a star at all, but rather the planet Venus. Venus is, of course, necessarily itself. Hesperus is necessarily Phosphorus. If it seems that Hesperus is only “contingently identical” to Phosphorus, Kripke’s explanation runs, that is because it really is contingent that the last bright object seen in the sky in the morning is the same item as the first bright object seen in the sky in the evening. There might have been two such objects, and they might even have been stars rather than a planet. But, again, that is not a case where Hesperus fails to be Phosphorus. It is merely a case where some other things, perhaps other things called “Hesperus” and “Phosphorus,” for that matter, appear as Venus in fact appears.
42 Chapter 2
If the appearance of contingency can be revealed as illusory, then identity can be maintained. Water is H2O. Hesperus is Phosphorus is Venus. Heat (in a gas) is kinetic molecular energy. Gold is the element with atomic number 79. And so forth. If the apparent contingency cannot be explained, then we are obliged to apply modus tollens and conclude that the case is not one of identity, but of some other relation instead. This is the problem that Kripke raises for identity theory. Suppose that we want to maintain some mind–brain identity claim: “pain is C-fiber activation,” “sensing red-afterimagely is activation in V5,” or your favorite example. It certainly seems that my pain now could have been other than, say, activation of my C-fiber #237 now. Why couldn’t it have been C-fiber #236? Or why couldn’t C-fiber #237 pain be just a little bit more piercing? Mind–brain identity claims have the appearance of contingency. A consequence of Kripke’s view is that, if we are going to maintain such identity claims, we need to explain away the appearance of that contingency. But, Kripke argues, the epistemic model fails when it comes to mind–brain identity claims. Sensations do not admit of the appearance–reality distinction that allows us to explain away the apparent contingencies in other cases. Pain is painful. Anything superficially like pain is pain. That’s just what pain is: “To be in the same epistemic situation that would obtain if one had a pain is to have a pain; to be in the same epistemic situation that would obtain in the absence of pain is not to have a pain. The apparent contingency of the connection between the mental state and the corresponding brain state thus cannot be explained by some qualitative analogue as in the case of heat” (Kripke 1972/1980: 152). Because the apparent contingency of the relation between pain and C-fiber activation cannot be explained away as in the other cases, Kripke concludes that pain cannot be identical to C-fiber activation. There are many objections to Kripke’s argument. William Lycan suggests a reading on which Kripke’s analytic tools “drop out” of the argument, leaving him with only a question-begging “Cartesian” essentialist claim (1987, ch. 2, passim).2 Moreover, Lycan and other functionalists deny that “pain” is used rigidly in, for example, “pain is C-fiber activation.” If “pain” is not used rigidly then Kripke’s argu-
43 The Illusion of Contingent Identity
ment is eluded. This would be the case, for example, if psychological terms are functional terms that refer to whatever occupies certain functional roles. Since the same functional role could be occupied by different kinds of states in different possible worlds, it seems that terms that refer to functional kinds are not rigid designators.3 These objections are of no use to me because I wish to grant (at least provisionally) Kripke’s modal treatment of identity, his essentialist intuitions, and that mind–brain identity claims use rigid designators. Another class of objection resists Kripke’s modus tollens directly by arguing that we have in fact discovered that sensations are brain processes, thereby showing that the intuition behind the apparent contingency is false, even if we cannot explain it away. This option is entertained in detail (but not endorsed) by George Bealer (1994). It is an interesting strategy, and perhaps I could be forced to accept it as a fall-back position with some amendment. But the hard-line scientific essentialist response fails to explain why we do not recognize the necessity of the identities; it would be preferable to provide an account that can explain the appearance of contingency. I am trying to make room for an identity theory about conscious mental states. This means that I have some explaining to do—more precisely, some explaining away. Kripke himself puts it thus: “[The materialist] has to hold that we are under some illusion in thinking that we can imagine that there could have been pains without brain states. . . . So the materialist is up against a very stiff challenge. He has to show that these things we think we can see to be possible are in fact not possible. He has to show that these things we think we can imagine are not in fact things we can imagine” (1971: 163). I accept that identity theory will make claims such as that a red sensation of such-and-such sort is identical to this-and-that activation, event, property, or some such in the brain. And I accept that the terms of the identity theory’s claims are used rigidly. Moreover, I accept Kripke’s claim of essentiality: pain is essentially painful, this sensation of red is essentially like this. I therefore accept that mind–brain identities would have to be, as scientifically discovered identities, necessary a posteriori—just like “water is H2O” and “Hesperus is Phosphorus.”
44 Chapter 2
To accept all of this is to take a hard line on the identity theory, a much less compromising stance than identity theorists have traditionally taken. Identity theorists such as C. D. Broad, U. T. Place, J. J. C. Smart, David Armstrong, and David Lewis insisted that the identities they had in mind were contingent. Kripke put a stop to that; now the identity theorist must maintain that the identities are necessary, or admit that they are not identities at all. William Lycan remarks that “[t]he upshot of all this is that the Identity Theorist has (and must make) a few choices that no one noticed prior to 1971; and the choices have rather serious repercussions” (1987: 21). It seems that most former mind–brain identity theorists would take the second route and back off from the strong identity claim. Lycan considers what the members of the above mentioned “Identity Theorists’ Hall of Fame” would claim about mind–brain identities post-Kripke. He conjectures that only Smart would maintain an identity theory with necessary identities (ibid.: 20–21, 133–134 nn. 16–20). Among others, Berent Enç (1986) has been perhaps the most prominent advocate of necessary mind–brain identities.4 For those of us who insist on continuing to defend identity theory, our task is set: Explain away the contingency. Kripke says that the epistemic model for explaining away the apparent contingency of certain identity claims (“water is H2O,” “Hesperus is Phosphorus”) is the only model he can think of, but he leaves it open that there could be others: “[T]he only model I can think of for what the illusion might be . . . simply does not work in this case. . . . [An alternative model] requires some very different philosophical argument from the sort that has been given in the case of heat and molecular motion. And it would have to be a deeper and subtler argument than I can fathom and subtler than has ever appeared in any materialist literature that I have read” (1971: 163). I am proposing just such an alternative explanation.5 Following Kripke’s lead, I will not try to cure us of the intuition of contingency we feel about mind–brain identity claims. Rather, I will reveal the sense of contingency to be other than it seems. The apparent contingency of mind–brain identities is not genuine metaphysical contingency; it is an illusion that can be explained by appeal to our epistemic situation.
45 The Illusion of Contingent Identity
2 Identity Conditions Kripke maintains that some identities can be empirically discovered. How could we go about discovering the necessary a posteriori scientific identities that Kripke allows? What is involved? Alan Sidelle (1992) explores the question of what distinguishes those cases in which we entertain the possibility of scientific identity (e.g., sensations and brain processes) from those we do not even consider (e.g., sensations and weather activity). Sidelle argues that such a discovery requires that we already know something about the things in question. He writes: While water turned out to be H2O, it could have turned out to be H2SO4, or XYZ, and if it had, we would presumably be as willing to accept these as H2O as identifications with water. On the other hand, it seems unlikely that we would consider “liquid enjoyed by George Washington” as a candidate for identity with water, even if it turned out that all and only water (the items we call “water”) has this feature. What this suggests is that there are “candidacy” criteria which anything must meet if it is to be a candidate for identification with our subject of inquiry. And I suggest that in order to be such a candidate, something must have identity conditions which are compatible with those for our subject. Identity conditions are compatible if, and only if, when spelled out precisely, they might be the same. (1992: 273)
Sidelle explains that identity conditions are: the sorts of things that are represented by statements saying, for any possible object, what features it must have in order to be, or those which suffice for it to be (identical to) some particular thing (or, for kinds or properties, for something to be a member of that kind, or possess that property). . . . a specification of identity conditions need not state with full precision—need not mention—what the relevant features are; for certain purposes, “this chemical microstructure” or “this thing’s origin” will do as well as “H2O” or “sperm S and egg O.” (Ibid.: 291)
The conditions of identity for a thing or kind of thing are the boundary conditions for that thing (token-identity) or that kind of thing (type-identity). They are the conditions for being a particular thing or a kind of thing. Identity conditions may be specified with more or less precision. But the specification of identity conditions need
46 Chapter 2
not exhaust them—the identity conditions for a thing may outstrip any actual specification. Water has always been H2O, even before that specification was available. Completely specified identity conditions give the precise principium individuationis for a thing or kind of thing. Consider water. Something is a sample of water only if it has a certain “chemical microstructure.” H2O and XYZ are candidates for identification with water, but not the liquid enjoyed by George Washington, because H2O and XYZ (but not liquids enjoyed by George Washington) are also things identified by their chemical microstructure. If Sidelle is right, before we can even consider whether a putative case of identity such as “water is H2O” or “temperature of a gas is mean kinetic molecular energy of a gas” is a genuine case of identity, we must already know that the things being identified have compatible identity conditions, at least when specified with some generality.6 It was discovered that gold is not a compound (much to alchemists’ despair) and that it is not the element with atomic number 14; but we do not even consider that gold is a Bordeaux wine. Common material stuffs have identity conditions (specified generally) that involve their contemporaneous microstructure, whereas French wines have identity conditions (specified generally) that involve their appellation and vintage. We do not even consider the identification of gold with Bordeaux, for the identity conditions (specified generally) for gold are incompatible with those for Bordeaux. Something is gold whether or not it comes from a certain place, at a certain time; not so for Château Haut-Brion. As it turns out (again, much to alchemists’ despair) the identity conditions of gold—specified precisely—are also incompatible with those of lead. But that empirical discovery is prefaced on knowing that the general identity conditions for each were microstructural, and then discovering that gold and lead differ in precise microstructure. However, precision in the identity conditions, the full principium individuationis, is not what is relevant to candidacy: Compatibility of conditions is required here, rather than sameness, because at the time an identification is being considered, it will be unclear what the precise identity conditions are—if, for example, the identity conditions for elements are to be given by their atomic structure, all atomic structures are
47 The Illusion of Contingent Identity
candidates for identification with some element, but before we make the identification we cannot specify, say, “atomic number 79.” If we had to know that the fully specified conditions were the same, we would already know whether the identity held. (Sidelle 1992: 273)
To know that a thing is a thing of a certain kind, one must know not only some information about the thing, but also which information is relevant to deciding whether the item is an item of a particular kind. One must know what sorts of properties are relevant to being a thing of such-and-such kind. Sidelle introduces the example of a schoolchild’s art project: Many of us, from some experience in elementary school, are familiar with aborted pottery projects. We know of urns that are made, looked over, smashed back into unformed clay. Does this not show that an urn is not identical to the piece of clay from which it is constructed—and perhaps even yield enough knowledge of the identity conditions involved to let us know when an urn has come into existence? Plainly not. Of course, on one reading we do know that an urn comes into existence when something of a certain shape, perhaps designed with certain intentions, is formed. But this does not tell us when an entity which is an urn comes into existence. . . . [W]ithout independently knowing that urns exist for only so long as they are urnish— and that this is false of lumps of clay—we cannot know that the smashed lump of clay is not identical to the urn. (Ibid.: 278)
If I want to know whether a certain bit of clay is an urn, I need to know that, say, macroshape is among the relevant criteria for being an urn. It’s not enough to know the actual shape of the thing. I need to know that shape rather than microstructure or atomic number is a condition for being an urn. To know that water is H2O, one needs to know that molecular structure rather than, say, density at sea level, is the condition of the identity of basic substance types. It looks as though “identity conditions (generally specified)” for a kind are at work in most philosophical discussions of individuation and identity through time. This is so of the literature on personal identity at least since Locke’s discussion. Locke (Essay, II.xxvii) holds that it is a matter for courts or psychiatrists, or God, to determine if this is the same person as that (i.e., the precise principium individuationis for this person). But it is a matter for philosophers to inquire after the general conditions for the identity and survival of substances, trees,
48 Chapter 2
brutes, persons, and so on. That is, the conditions for when something comes to be or ceases to be the thing or kind of thing that it is. Knowing that things have the same identity conditions is necessary for knowing that they are, in fact, identical. (Perhaps it is sufficient, but that is another matter.) If Sidelle is right, it is reasonable to suppose that believing that things have compatible identity conditions is required for considering whether they are identical. If we are certain that two things, gold and Bordeaux, say, have incompatible identity conditions (specified at whatever generality) then they are not even candidates for identity. We can rule out identity on that basis alone. What happens when one does not know the identity conditions for a thing? Perhaps you are a proverbial American analytic philosopher who comes across a cricket match. Suppose you know that it is a cricket match on superficial grounds (you have seen cricket bats sold in a store, someone reliable has told you that it is a cricket match). Will you thereby know the identity conditions for cricket matches? Plainly not. You might be entirely unaware of the identity conditions for the match; you might not know that the match can continue even when there is nothing “cricketish” occurring on this field, unlike the urn, which must remain urnish to be an urn. Or you may have some notion of the identity conditions for the match, but not enough to know for certain, in borderline cases, whether an event is part of the match. In this case, one knows some of the properties of a thing, but not enough to know the identity conditions for things of that kind. Another case wherein identity conditions may be unknown is when one is entirely unsure of what kind of thing one is faced with. Consider the other side of Sidelle’s tale about gradeschool pottery projects: One day your child, sibling, niece, friend, or whoever arrives with a painted lump of clay. “Look what I made!” Is it an urn, you wonder? A bowl? The notorious ashtray? If the identity conditions involve shape it might be one thing; if they involve function, another; and so on. It seems that one must know the identity conditions (generally specified) for this kind of thing in order to begin to guess what it is. But to discover those conditions, it seems that one must already
49 The Illusion of Contingent Identity
have some idea of what (generally) the thing is! You must know, for example, that craft projects brought home from school—despite overwhelming prima facie evidence—are not simply lumps of clay that have been painted. Suppose we are considering the type-identity of two kinds of things, widgets and thingamajigs. If we know that their identity conditions are incompatible, then we know that they cannot be identical. If we know that they have the same identity conditions, then we have (at least) strong reason to think that they are identical. And if their identity conditions are compatible as far as we know, then—as far as we know—they are candidates for identity. Likewise, if we do not know how to individuate either widgets or thingamajigs, or both (if we do not know their identity conditions) then as far as we know they are candidates for identification—for as far as we know, their identity conditions are compatible.7 But at the same time, as far as we know their identity conditions are incompatible. If we are uninformed about the identity conditions for some widgets and thingamajigs, then even if they are identical it might seem that they could fail to be, that it could be otherwise. Thus arises the appearance of contingency. Let us return now to the cases of scientific identity that concern Kripke and identity theorists. Before the advent of molecular chemistry, the identity conditions for water (as far as they were known) were compatible with its being an element or a molecule. If water could be an element, then it could fail to be the molecule H2O. Even on the view of molecular chemistry, the identity conditions (generally specified) for water might be compatible with its being some molecule other than H2O. Only when the identity conditions for water are more precisely specified, perhaps fully specified, do we conclude that water could not be other than H2O. The identity conditions for water are identical to the identity conditions for H2O, and they are incompatible with those of any other stuff. Having discovered that the identity conditions for water and H2O are identical, the apparent contingency of “water is H2O” is eliminated.8 I am suggesting that some identity claims may appear contingent, even though they are not, because we do not know the relevant criteria of identity to apply. This will often be the case if two things are
50 Chapter 2
candidates for identity, but are not known to be identical or nonidentical. Candidacy—compatibility of identity conditions (generally specified)—thus provides a second model for explaining away the apparent contingency of putative identities; call it the candidacy model. It might seem that water does not have to be H2O if you do not know how to identify typical material mass kinds, and if you do not know that molecular structure is the way we individuate substance kinds.9 It might seem that this thing does not have to be a chair if you do not know how to identify chairs; but once you know how chairs are identified, you know that this thing cannot fail to be a chair. You might still, of course, fail to recognize that this is a chair—you might be unsure, you might think that it is not a good example of a chair, or that it is only vaguely a chair. But all of these possibilities that you might consider depend on your already having beliefs about the conditions of identity (generally specified) for chairs.10 The candidacy model works because failing to know something’s (precise) identity conditions is a way of being in an epistemic position toward it that creates the appearance of contingency. But unlike Kripke’s model, the candidacy model does not depend on an appearance–reality gap.11 Take the case of chemical elements. It might not have seemed, at one time, that gold must be the element with atomic number 79. It did not seem that way, for example, before the atomic theory of chemistry was developed. The atomic theory tells of how to individuate things of certain kinds. After the atomic theory, we do not think that gold could have atomic number 80, or that something with atomic number 79 could fail to be gold. I am claiming that, when it comes to sensations and brain processes, we are in a position analogous to chemistry before the atomic theory.12 This is precisely where human beings stood vis-à-vis water and H2O in 1640; and it is precisely where we still stand with respect to the identification of sensations and brain processes today. We do not know how to individuate brain states. We do not know the identity conditions of brain states. It is sufficient for my case that we do not know how to identify brain states; but I also contend that we do not know, with much specificity, the identity conditions for sensations. The latter is a more controversial claim. As I say, it is sufficient that
51 The Illusion of Contingent Identity
we are under- or uninformed about the identity conditions of one or the other; if it is both about which we are in the dark, so much greater will be the illusory sense of contingency. The candidacy model of explaining away the apparent contingency of identity claims is clear: If we do not know the complete identity conditions for things that are candidates for identification, then the appearance of contingency may be accounted for. This is what I am claiming about sensations and brain processes. In the next section I briefly defend my contention that we do not know the identity conditions for brain states or sensations. 3 Concerning Our Ignorance of Identity Conditions If true, identity claims such as “pain is C-fibers firing” are necessary a posteriori.13 According to the candidacy model, we might believe that pains are only contingently related to C-fiber firings because we do not know that pains and C-fibers have the same principium individuationis; we might therefore (wrongly) believe that they could have different identity conditions. This would be the case if we did not know the precise identity conditions for brain states, or if we did not know the precise identity conditions for sensations. I am suggesting that we do not know the precise identity conditions for either—neither for sensations nor for brain processes. I will begin my explanation with brain processes. We do not know how to individuate brain states, properties, processes, events, and so forth. Not only do we not know how to individuate these things, we don’t really even have a clue what such things are. “Brain states” are place holders for things that we hope someday to understand, the soon to be discovered gears of neuromechanics. This is why, for better or worse, the clumsy example of C-fiber firings has had such a long life in philosophical writings: It is merely a placeholder. From the point of view of the recent history of philosophy of mind, it’s almost an accident that C-fibers have anything at all to do with pain sensation in humans, and it’s more or less irrelevant that they are not the whole story. Bechtel and Mundale put the point sharply: “Actually, the notion of a brain state is a philosopher’s fiction; a notion closer to what neuroscientists would
52 Chapter 2
use is activity in the same brain part or conglomerate of parts” (1999: 177). Of course, brains are things that have states, properties, processes, and in which events may occur. Brains, that is to say, are physical objects. In whatever sense we understand what it is for any generic physical object to have states, we have a notion of a brain state. Brains are objects that have properties at times. But that notion of a state is too thin to do much work in understanding brains as minds. Similarly, I don’t expect the thin concept to be the only or primary notion of “state” invoked by chemists or automobile mechanics to explain chemical reactions or why my car won’t start. For one thing, the thin notion makes no distinction between states of molecules, states of automobiles, and states of brains. But beyond the thin notion, philosophers have nothing more to say about brain states, brain properties, brain processes, brain events. The philosophical notion of a brain state is that of a mechanism that will play an explanatory role in a science of brains. When philosophers talk about brain states they are gesturing toward a thicker notion of a brain state—one that calls for a substantial neuroscientific theory. And neuroscientists don’t—yet—have much to say. I do not know how to thoroughly defend the negative existential claim I am making: that as a matter of empirical fact at this time, we do not know the identity conditions for brain states, properties, processes, events, and such. How does one defend the claim that some putative bit of knowledge is not now had? The best I can do is to consider the only candidate I know of for knowledge about brain states: brain-imaging studies. You’ve probably seen nice brain-imaging pictures in National Geographic or on the Discovery Channel. Brain-imaging studies reveal such interesting results as that there is localized brain activity distinctive of the visual processing of faces (see, e.g., McCarthy et al. 1997). The images are clear and distinct, the stuff of science fiction. And the spatial and temporal resolution of imaging techniques is improving almost daily. Aren’t those functional magnetic resonance imaging (fMRI) pictures, for example, images of brain states? No. We must be careful about what we take those wonderful images to show.
53 The Illusion of Contingent Identity
Grossly put, an MRI machine turns the subject’s body into a radio transmitter. The signals transmitted by different parts of the body depend on the characteristics of the machine itself, and of the tissue and other substances of the body. Hemoglobin, a molecule found in red blood cells, has different magnetic properties depending on whether or not it is oxygenated. Exploiting this property of hemoglobin, the fMRI technique produces a representation of where the metabolic demands of neurons require additional blood flow. Given certain plausible assumptions about circulation and cellular oxygen consumption, fMRI tells us where and when there is activity in the brain. There are other techniques that directly or indirectly measure activity in the brain, for example by measuring electrical fields on the scalp, the electrical potentiation of intracranial cells, or the reflective qualities of tissue on an exposed cortical surface (see figure 2.1). Which technique yields pictures of singular brain states? We do not know. Probably none. That is the first reason that brain imaging is not the answer to what brain states are and how to individuate them. Each picture or kind of picture is at best an incomplete representation of what is going on in the brain. None of them gives pictures of “brain states” that could play a general explanatory role in a science of the brain. Of course the value of brain images to physicians and researchers is that we can observe, with differing degrees of resolution, that various kinds of ability or disability are correlated with activity levels or pathologies in various brain regions. But one imaging technique may not show any neural difference between prosopagnosiac and normal patients, while another reveals some correlation. (Prosopagnosiacs cannot recognize people by seeing their faces.) Different imaging techniques are useful for studying different aspects of brain and mental life. One might think that this tells us, in a piecemeal way, about some kinds of brain states and processes. Not so fast. Direct measuring techniques, wherein electrodes are put into brain tissue, are local measures—they give information about a small number of neurons, even just one. But they don’t tell us what is going on elsewhere in the brain. One or a few isolated neurons or
54 Chapter 2
3 MEG+ERP
Brain 2
Lesions PET
Size (log mm)
Map
fMRI
1
Microlesions
Column 0 Optical Dyes
Layer
2-Deoxyglucose
-1
Neuron Single Unit
-2 Dendrite
Patch Clamp
-3 Synapse -4
Light Microscopy
-3 Millisecond
-2
-1
0 Second
1
2 Minute
3
4 Hour
5
6
7
Day
Time (log seconds)
Figure 2.1 Experimental techniques of neuroscience. This schematic diagram depicts the range of spatial and temporal resolutions at which various experimental techniques are typically applied. Temporal resolution is represented on the horizontal axis, and spatial resolution on the vertical axis; in each case resolution decreases (the unit size is larger, so the technique is more coarse grained) as it extends from the origin. Note that some techniques are applied over a wide range of resolutions, and others are quite restricted. Used with permission from P. S. Churchland and T. Sejnowski, “Perspectives on cognitive neuroscience,” Science 242 (1988): 741–745. Copyright 1988 American Association for the Advancement of Science. Adapted following Cohen and Bookheimer (1994) and Deutsch et al. (2000).
55 The Illusion of Contingent Identity
ensembles do not a brain state make. Wide-view techniques, such as fMRI, give us the bigger picture. But that picture is too big. It is already highly interpreted. This is the second problem in thinking of brain images as pictures of brain states. Remember, fMRI uses every hemoglobin molecule as a transmitter. All those transmitters are sending out signals. The genius of fMRI is that, through careful experimental design and sophisticated signalprocessing algorithms, we can get a picture indicating which regions of the brain are, statistically, differentially active during specific kinds of mental tasks. The resulting images show those regions for which the activation is statistically correlated with the task. Analyzing the fMRI data into those pretty pictures we see takes some serious mathematics. But then we face a tough question: What is the “object” that an fMRI image depicts? Figure 2.2 includes three fMRI-produced images of my brain activity during a passive viewing experiment. On the left is an image indicating the average activity in my brain at each time a visual stimulus was presented. In the middle is an image colored to represent, for each brain area over many trials, the degree to which activation in the area was accounted for by the experimenters’ model for the data. On the right is the same image colored to indicate just those areas that were highly correlated with the experimental model, that is, that are well explained by the hypothesis. For most readers, and for me, images like that on the right are what we usually associate with functional neuroimaging results. But only a small part of the total activity in the brain is represented by the clear and distinct image on the right in figure 2.2. The right-hand image is not in any obvious sense a picture of a single brain state. The experiment I participated in was looking for brain activity that was highly correlated with certain events during the course of a task. Moreover, the experimenters were interested only in activity in brain areas already known to be involved with visual processing. So the images in figure 2.2 (center and right) ignore highly active areas that, for example, remained active throughout the experimental trials, such as activity that may be associated with visual processing generally but is not specific to the experimental task. Finally, these images were adjusted to highlight the areas of signal
56 Chapter 2
Figure 2.2. fMRI images of Polger’s brain. Courtesy of Scott Huettel, Brain Imaging and Analysis Center, Duke University Medical Center. Used by permission. Three fMRI images of the author’s brain during a visual processing experiment. On the left is the “raw” image, showing brain activation at the time that the stimulus was presented, averaged across all trials. Each block (called a “voxel”) composing the image represents the activity of a 3.125 mm ¥ 3.125 mm ¥ 7 mm volume of the brain, averaged across 80 trials. The variance in the signal from any voxel over the course of a trial is less than 3 percent. In the middle image, every voxel is colored to indicate the probability that the activity in that voxel during the 18 seconds surrounding the task trial (from 5 seconds before the presentation of the stimulus, until 13 seconds after) is due to the task, as determined by statistical analysis comparing the actual activity to a hypothesized model. On this depiction, highly correlated areas appear lighter, areas of low correlation are dark. The right side image is the result of setting a threshold for significance on the middle image data—only highly correlated voxels are colored. The threshold is selected so as to highlight only those brain areas thought to be of relevance to the experiment, in this case the region around area V1, the primary visual cortex. (For spatial context, the colored voxels are overlaid onto the “raw” image, from left.) The study in which I was a subject is reported in Huettel and McCarthy (2000). See .
activity that are maximally correlated with the experimental model. Much signal—including some signal that was statistically correlated with the task to a high degree—has been ignored to render this clear picture. It would be rash to think that any of these three fMRI images is a complete picture of denumerable brain states, or that the fMRI showed us how to individuate those states. Iceberg metaphors are appropriate, but matters are even worse. The analyzed fMRI images show only those portions of the brain differentially active during a certain task; but a great deal of other activity that is common to many tasks has been, as it were, subtracted out. To think that an fMRI shows how to individuate brain states would be like thinking that the identity conditions for cricket matches are to pick out only those features that, statistically, differentially occur during all the cricket games of the past year. Obviously, this will leave out certain aspects of at
57 The Illusion of Contingent Identity
least some cricket games that were unique or statistically anomalous (unusual plays, injuries). In addition, it could include some aspects (number of fans, snack-purchasing behavior, average temperature in London) that are not part of the identity conditions for cricket matches at all. Note that, in figure 2.2, some signal appears to be coming from a location outside my brain—and it was reasonably well correlated with the task model! It is sufficient for my purposes that we do not know how to individuate brain states. If I am right, that alone can explain the apparent contingency of mind–brain identity claims. But I also suggested that we do not know how to individuate sensations. This is more controversial, and as I do not need the claim I will spend little time defending it. But it is a reasonable claim. Pains, I have agreed with Kripke, are essentially painful. Whatever is qualitatively like a pain is a pain. We know, from a first-person perspective, when we are experiencing pain. But that doesn’t mean that we know how to individuate pain events of different sorts from each other, or from every nonpain event. The sorts of examples I have in mind include the conundrums about affect that philosophers from Berkeley (1998 [1713]) to Dennett (1988) have exploited. But I also point to the tremendous body of phenomenological research, from introspectionist psychology to present-day psychophysics, that is aimed at better describing and understanding our sensations.14 Phenomenology, it should come as no surprise, has to be studied. Consider such common examples as a person learning to appreciate music or wine. In learning to appreciate music, one learns to make finer discriminations than one previously made.15 Likewise, when one is learning to appreciate wines, one comes to make finer discriminations in the tastes of wines. To an untrained person, all red wines taste alike; to a connoisseur, this red wine is fruity whereas that red wine is peppery. If someone is teaching you to appreciate wine she may say such things as, “notice the flavor of lychee,” in order to call your attention to other experiences distinct from but qualitatively like the present one. (Expressions like “the flavor of lychee” seem to be used rigidly in such cases.) Less pedestrian examples are to be found in the work of introspectionist psychologists who trained their subjects to make finer discriminations for themselves, or contemporary psychophysicists who
58 Chapter 2
follow their psychological forebears in attempting to quantify the quality of experiences.16 Both groups assume that people can be mistaken about fine-grained individuation of sensations, so that they must either be trained or controlled, or both, to make an accurate assessment. An example that has recently excited renewed attention from philosophers is the study of color vision. Almost all of the parties to debates over color now agree that the structure of color space is not widely known or understood, and that it is relevant to scientific and philosophical concerns about color. Almost no one still maintains that color sensations are simple and unstructured. The study of color vision is precisely a case where philosophers and color scientists have been seeking physiological features that have the same structure as the phenomenal features of color experience. One way to view this project is as attempting to discover what in the brain or world has the same identity conditions—so far as we know them—as sensations of color.17 The project I have been describing is a difficult one. Discovering the identity conditions of sensations and brain processes is no simple matter, and I am not here making any specific proposals about how to proceed. But I believe it is a tractable project, and indeed, one that is being actively pursued by philosophers and scientists. It would be foolish the declare even small bits settled today, for controversy swirls; but there is some progress. Larry Hardin (1988) reignited interest in color and color vision that has been, to my mind, quite productive.18 Recent work on pain and dreams seems equally promising.19 It should be clear that I am not pushing the mysterian line of reasoning, according to which knowing the identity conditions for sensations and brain processes, and thus knowing the mind– brain relation, is forever beyond our grasp. On the contrary, Owen Flanagan and I have argued that understanding conscious minds is a tractable project once it is understood in a naturalistic framework (Polger and Flanagan 1999). So the ignorance that, according to the candidacy model, explains why we think that mind–brain relation is contingent is not permanent. Of course, it may be that the average person who is not an expert neuroscientist will typically not know the identity conditions in enough detail to entirely eliminate the appear-
59 The Illusion of Contingent Identity
ance of contingency. But this is just as it is now with water and H2O or any other scientific identity. The candidacy model predicts and explains the widespread persistence of such beliefs, even while explaining how they can be incorrect. 4 Competing Explanations of Apparent Contingency How does my candidacy model compare with Kripke’s epistemic model? Kripke claimed that there could be no epistemic failure in the case of sensations in the way that there could be with temperature or water. The reason, recall, is that to be in an epistemic situation like that of experiencing pain is to experience pain.20 This is the part of Kripke’s argument that he and others sometimes call essentialist. At least for the present purposes, I have not disputed that claim. But the candidacy model does not maintain, as Kripke’s epistemic model does, that we have a true belief about a contingent relation that we mistake for a belief in contingent identity. Rather, it proposes that we have a false belief about the possibility that a certain sensation could fail to be the (in our case) brain state, property, process, or event that it is. We may make this mistake because we do not know how to properly identify either sensations or brain states. The candidacy model does not require us to pretend—what Kripke thinks the materialist must absurdly claim—that there is a state qualitatively like sensation (something toward which we could be in the “same epistemic situation”) that is not a sensation. We need only observe that we are not very good at individuating sensations or brain states, and so we can mistakenly think that this (insert inner ostention here) sensation could be other than this (insert nextmillennium neurobiology here) brain state. Likewise, we need never claim that there is something qualitatively like a brain state (something toward which we could be in the “same epistemic situation”) that is not a brain state. Unlike Kripke’s epistemic model, the candidacy model does not depend on a contingent relation between appearance and reality. Instead, the candidacy model gains its purchase because our knowledge of the essential properties of sensations and brain states is at best incomplete.
60 Chapter 2
Kripke is right: To be qualitatively like a sensation is to be a sensation. But there can be something qualitatively like this sensation that is not this sensation; for example, a slightly different sensation. I may mistake cold for pain, C sharp for F sharp, the sight of red dye no. 9 for the sight of red dye no. 7, or the taste of 1981 Bordeaux for the taste of 1983 Bordeaux. This is not a problem—not a metaphysical problem, at any rate. I might think that the stuff in this glass could fail to be H2O. But if I am not good at discriminating water from other substances that are not water (gin, vodka, XYZ, whatever) then it is easy to see why I think that the relation between water (i.e., the stuff in this glass) and H2O is contingent.21 In this case the stuff I call “water” is not just water, but water and (sometimes) gin, vodka, XYZ. This is not because I correctly use “water” nonrigidly. Rather, it is because I am using the term rigidly but sometimes incorrectly. I am making mistakes and I do not recognize it. In this case, I think there is just one kind of stuff, but I am bad at distinguishing it from similar stuff. Likewise, if I am not good at telling the difference between pains #124 through #133, then it will be no wonder that I think that pain #129 could fail to be identical to brain state #129. Moreover, I can agree with Kripke that we individuate both sensations and brain processes (insofar as we try to pick them out at all) by essential properties. But that is not to say that we pick them out by all of their essential properties.22 In practice—in life—we pick out sensations by essential properties. But many of the essential properties of sensation S1 are also essential properties of sensation S2. Some things (S2, S3, S4 . . . Sn) can be like S1 with respect to those essential properties by which we normally pick out sensations, but still not be S1. A sensation could be mistaken for S1 because it differs with respect to (essential) properties to which we are not ordinarily sensitive when differentiating sensations. We can be mistaken, or just sloppy, about the fine-grained individuation of sensations. In the case of color, the conditions of identity for precise individuation have been more carefully studied than other sense modalities. What about pains? Neuroses? Moods? Has anyone never been wrong about when they are experiencing burning versus freezing, lust versus love, or some such? Of course Kripke maintains not only that we believe that S1 could fail to be identical to B1, but moreover that we conceive of sensations
61 The Illusion of Contingent Identity
such that they could fail to be identical to any brain state at all. But this is precisely what we would expect in the condition of being under- or uninformed about the identity conditions for sensations and brain states. If I am uninformed about the identity conditions for widgets and thingamajigs, then it may seem to me that this widget could fail to be identical to any thingamajig. And I could be wrong. But what I will need to learn to resolve the issue is not merely more descriptive facts about widgets and thingamajigs; I need also to know about the identity conditions of widgets and thingamajigs. This is how we can understand why more neuroscientific information by itself will not ameliorate Kripkean concerns. We need more than knowledge of how neurons relate to one another, how sensations relate to one another, and the correlations between them; we need an account of the identity conditions of sensations and brain processes. The candidacy model explains the apparent contingency of mind–brain identities without denying what Kripke holds dear. The mistake is that Kripke settles for an extremely coarse individuation of sensations and brain processes and thus for overly general conditions of identity. If we consider pains only generally, rather than some specific sort of pain, then we should not be surprised that it seems that pain could fail to be identical to some general type of brain state (C-fiber activation, rather than some specific neural state). This is just like the mistake about granularity, discussed in the previous chapter, that Bechtel and Mundale (1999) criticize in Putnam with respect to multiple realizability. Richard Boyd comes close to considering my response to Kripke when he suggests: What seems to have been overlooked is that the successful employment of [the epistemic] strategy does not require that it be the less “technical” or “scientific” term in the identity that is replaced by a nonrigid designator. . . . Thus the employment of a corresponding contingent qualitative statement to explain away the apparent contingency of “Pain = C-fiber firings” does not depend on the existence of a possible world in which pain does not feel like (actual world) pain or in which some nonpain feels like an actual world pain. (1980: 84)
This response (an application of what Boyd calls the “[new] standard rebuttal” to essentialist criticisms of materialism) depends on taking
62 Chapter 2
one term of the identity claim as nonrigid. In Kripke’s example that term is “the watery stuff”; in Boyd’s it is the mental term, like “pain.” This works for Boyd because he is mainly concerned to argue that the materialist need not be committed to true identity statements using rigid designators. But that will not do for the identity theorist. Whereas the “(new) standard rebuttal” can explain away the intuition that my sensation now might not be a C-fiber firing, it cannot similarly explain away the converse intuition that my C-fiber firing might not be a pain.23 The reason is that the functionalist view that Boyd advocates rejects that intuition: If C-fiber firing is one of many realizations of pain, it could not fail to be a pain. Instead of explaining away the intuition, Boyd must explain why we can be mistaken about our intuition. He does this by arguing that when pain is taken nonrigidly (e.g., as a functional state) we can be mistaken about the contingency of its instantiation by a firing C-fiber because, “[o]ne cannot tell by just looking (at least not easily) just what function a particular circuit will compute. . . . It is difficult to deduce the function computed by a computing device just from a specification of its internal structure” (1980: 98). Because the computing device is complicated and its function is not readily apparent from its structure, we simply have “mistaken modal intuitions” about whether or not that system must compute a certain function. Of course if someone (an electrical engineer, say) knows enough about the system and about computing devices then she might not share the mistaken intuitions that Boyd and I have—they are a symptom, it seems, of our ignorance about electrical engineering. This part of Boyd’s explanation, like my candidacy model, rests on a simple failure on the part of persons to recognize the necessity of the relation.24 The candidacy model supposes that we can be wrong about the intuition of contingency in both directions—we could not have this brain state and fail to have this sensation, nor vice versa. And my model explains why we might have these false beliefs. I agree that it is a merit of Kripke’s epistemic model that it explains our mistaken sense of contingency by attributing to us a true belief, rather than by attributing to us ignorance as my model does. All else being equal, we might prefer a theory that does not indict us of an error. But that
63 The Illusion of Contingent Identity
advantage does not outweigh the explanatory power of the candidacy model. All else is not equal. 5 Levine’s Explanatory Gap An influential reinterpretation of Kripke’s modal argument is Joseph Levine’s “explanatory gap” (1983, 1997, 2001; see also Chalmers 1996a). Levine endorses a roughly functionalist strategy against Kripke’s metaphysical conclusion: He denies that the relata of the mind–brain identity are rigidly designated. Even so, he contends that an epistemic variation of Kripke’s essentialist argument shows that materialism is explanatorily inadequate, at least with respect to qualitative states. Levine argues that although Kripke’s argument fails to secure a metaphysical judgment against materialism, it nevertheless shows that materialism leaves an “explanatory gap.” Levine constructs an epistemological variation on Kripke’s reasoning, arguing that the apparent contingency of mind–brain identity claims reveals that these claims fail to yield adequate explanations of mental phenomena. Claims such as “pain is C-fiber firing” (whether interpreted as using the object terms rigidly or nonrigidly) fail to make the relationship intelligible: “What is left unexplained by the discovery of C-fiber firing is why pain should feel the way it does! For there seems to be nothing about C-fiber firing which makes it naturally fit the phenomenal properties of pain any more than it would fit some other set of phenomenal properties” (Levine 1983: 357). The explanatory gap comes from the same apparent contingency of mind–brain identity claims that yields Kripke’s metaphysical conclusion. On some familiar conceptions of explanation, for example, Hempel and Oppenheim’s (1948) deductive-nomological model, explanations require necessity, say, in the form of causal laws. Likewise, Chalmers and Jackson (2001) suppose that genuine understanding comes only from deductive entailment. On these views, explanations explain—make intelligible—precisely because they show that one phenomenon is necessitated by another. If this is right, the contingency of the mind–brain relation seems forever to foreclose that it could be the basis for an explanation of the mind in terms of the brain.
64 Chapter 2
Levine believes that identity, in contrast, leaves no explanatory gap. But he is persuaded by multiple realizability arguments against the identity theory, so he does not consider that view as a contender. If I’m right, multiple realizability concerns do not eliminate identity theories. But there is reason to think that Levine would not welcome my friendly proposal to revive identity as an answer to the explanatory gap (Antony and Levine 1997). Instead he looks to functional explanations to provide the basis for token identity.25 According to Levine, only “reductive” explanation can close the gap. Such explanation has two stages: first we get our concepts in order by “identifying the causal role for which we are seeking the underlying mechanisms,” and then we do the empirical work of looking for those mechanisms (Levine 1997: 551). The trouble is that our ordinary psychological concepts, in particular our concepts of qualitative conscious states, are not relational concepts; they cannot be analyzed in terms of their functional roles. The “reductive” strategy therefore falls short. What seems to be responsible for the explanatory gap, then, is that our concepts of qualitative character do not represent, at least in terms of psychological contents, causal roles. Reduction is explanatory when by reducing an object or property we reveal the mechanisms by which the causal role constitutive of that object or property is realized. Thus, to the extent that there is an element in our concept of qualitative character that is not captured by features of its causal role, to that extent it will escape the explanatory net of physicalist reduction. (Ibid.: 553)
Thus, the explanatory gap. In the past I have been inclined to a combination of three responses to the explanatory gap. One response is to admit that there is a feeling of “gappiness” but deny that that intuitive gap is either a metaphysical or explanatory problem. We might, following Owen Flanagan, suggest that Levine has set the bar of intelligibility too high. Theories of mind do not have to be intuitively satisfying, just explanatorily satisfactory (Flanagan 1992: 93–97, 119). A second response is to point out that Levine’s gap may be ubiquitous, that there are such explanatory gaps in all identities. I have already confessed my own persistent intuition that water could be other than H2O. Both of these responses dig in their heels and accept that there
65 The Illusion of Contingent Identity
are some brute facts about mind–brain identities that we must accept whether they are intelligible to us or not. Why should this be a special problem for identity theory? As Flanagan writes, “it is hard to see why the accounts of the nature of water in terms of its molecular structure, of life in terms of DNA, and so on, aren’t also brute” (ibid.: 119).26 These seem reasonably satisfactory responses, though admittedly not very satisfying. Even accepting that the explanatory gap is no more a threat to our materialist explanations than Kripke’s argument is to materialist metaphysics, we would like to be able to explain why it seems (at least to some people) that the mind–brain case is different from, for example, the water–H2O case. Levine fairly requires, “we need a good materialist explanation of the explanatory gap” (1991: 39). A third response that attempts to provide such an explanation is what Flanagan and I have called the structure of experience reply (Polger and Flanagan 1999). Earlier I pointed out that sensations and brain processes have a structure that is not self-evident, the discovery of which requires study and a research program. The more that the structures on the sensation side match the structures on the brain side, the more intelligible identity claims (whether rigid or nonrigid) will be: “The more we can articulate structure within the phenomenal realm, the greater the chances for physical explanation; without structure we have no place to attach our explanatory ‘hooks.’ There is indeed residue that continues to escape explanation, but the more we can explain relationally about the phenomenal realm, the more the leftover residue shrinks toward zero. Though I admit we are as yet a long way from that” (Van Gulick 1993, in Block, Flanagan, and Güzeldere 1997: 565). If we follow Van Gulick we can see that the residue, the that-which-is-not-explained by our explanations, generates the intuition that our explanations are incomplete—that there is an explanatory gap. If the residue is reduced, so is the gap. This strategy is employed by, for example, Hardin (1988, 1991) and Flanagan (1992), who argue that the explanatory gap reveals that we have been dealing too coarsely or sloppily with our mind–brain identities (pains and c-fiber firings). To make the gap disappear, on these views, we need to give our explanations at the correct grain of analysis. In this sense, the explanatory gap, like standard views of
66 Chapter 2
multiple realizability, is a symptom of the incompleteness of the grain project described by Graham and Horgan (2002). While I continue to believe in the structure of experience strategy—indeed, it is the only methodological program that shows any promise whatsoever—I have come to think that even if it can narrow the explanatory gap, the structure of experience approach can never by itself close the gap entirely. The reason, you will have anticipated, is that discovering the structure of experience—like making and smashing clay urns in gradeschool—does not in itself tell us what the identity conditions for sensations and brain processes are. Unless we know the identity conditions for sensations and brain processes, all the structural knowledge will not yield the explanation we seek. On the present view, when Levine (1983) laments that “there seems to be nothing about C-fiber firing which makes it naturally ‘fit’ the phenomenal properties of pain,” he is reporting the illusory contingency that appears because we do not know how to individuate sensations or brain processes.27 If we knew what the identity conditions for C-fiber firing are and we knew that they were the same as the identity conditions for pain, then perhaps we could see that C-fiber firing “fits” the properties of pain. The structure of experience reply said some of this, but it stopped short. It is not enough that we discover some structure of sensations that matches some structure in brains. If we want to claim identity, we must also discover the criteria for identifying sensations and brain states. Then we will be able to see that when there comes into existence something that is a C-fiber activation there necessarily comes into existence something that is a sensation of pain, and the converse. Unless we know that the identity conditions for sensations and brain processes are incompatible, the possibility remains open that they are identical. So until we know the precise principium individuationis for sensations and brain processes, it will appear that they could fail to be identical. If they are identical, that appearance is illusory; and it is an appearance that, pace Kripke and Levine, can be explained without abandoning identity theory. The candidacy model shows promise in closing the explanatory gap—when we know the full identity conditions for sensations and brain processes, the sense of contingency may subside.28
67 The Illusion of Contingent Identity
If Sidelle is right, then coming to know the identity conditions for a kind of thing is a necessary part of any process of discovering necessary a posteriori identities. If so, it is likely that the candidacy model that I have proposed can be applied broadly to explain away apparent contingency in past and future cases of scientific identification. Levine (1991) does not think the structure of experience reply closes the explanatory gap; as noted above, he argues that closing the explanatory gap could require some conceptual work as well as empirical discovery (1997).29 For example, Levine writes, “When we know enough chemistry, we see why water is liquid at room temperature, colorless, etc., because we see that it could not—given the facts of chemistry—be anything else” (1991: 38). On the present proposal, however, that is only going to do the job if knowing “enough” chemistry includes knowing the identity conditions for chemical kinds. If the identity conditions of a thing are part of the complete concept of it, then getting our concepts right would in fact give us the conditions of identity. Whether identity conditions are part of the very concepts of the kinds in question, I do not know. Prima facie, the identity conditions of a thing do not have to be part of the concept of it. The example of the American philosopher at a cricket match seems to suggest that one could have a concept of a kind of thing without knowing its (precise) conditions of identity. At the very least, it would take an argument to show otherwise. According to the candidacy model, there is little or no explanatory gap in the identity of water and H2O because we know with great precision the identity conditions for water and H2O; there is an explanatory gap in the case of sensations and brain processes because we do not know the identity conditions for sensations or brain processes well at all. The candidacy model can also explain why functionalism has seemed to be closer to closing the explanatory gap than the identity theory: The identity theorist is not yet in a position to give the identity conditions for brain states, whereas the functionalist’s computational paradigm proposes (with varying precision) a model for the identity conditions of functional states. And it is those states with which sensations are to be identified. Even though functionalists present themselves as having a better-specified candidate for identification with sensations than the identity theory, the explanatory gap persists. The candidacy model explains this:
68 Chapter 2
Functionalists still don’t know the precise identity conditions for sensations; even if functional kinds are candidates for identification, as far as we know the functionalists” proposed identity conditions for the physical bases of sensations (viz., functional states in the brain) are wrong. So the candidacy model explains the apparent contingency of even functionalist claims better than functionalist accounts on their own. 6 Reconsidering Identity Theory A final word about the structure of my argument is required. I reject Kripke’s conclusion that the mind–brain relation is known to be contingent. But I do not assert the contrary claim that the mind–brain relation is known to be necessary. Nor do I argue that the identity conditions of sensations and brain processes are known, with specificity, to be compatible. For this reason my argument is not simply begging the question against Kripke, for I do not assert what he denies. The point, rather, is that Kripke is not entitled to his conclusion. Like Kripke’s epistemic model, the candidacy model aims to explain how a relationship can appear contingent even if it is in fact necessary. I argue that, in our current ignorance of identity conditions of both sensations and brain processes, as far as we know sensations and brain processes may be identical. This is enough, on the candidacy model, to explain why we might think that the relation between sensations and brain processes is contingent even if it is not. My explanation does not assume that mental states are in fact identical to brain states. (This, again, is what protects my strategy from the charge of question-begging.) And my argument does not conclude that minds must be identical to brains. But remember Kripke’s challenge to the naturalist: if we cannot explain the appearance of contingency between the objects of a putative identity, then we must deny the identity; but if we can explain the appearance of contingency, then it is possible that the items are identical. By showing how the appearance of mind–brain contingency could be explained in a way compatible with the truth of identity theory, the candidacy model opens the door for such a theory. I have not argued in this chapter that an identity theory is correct, much less that we know it
69 The Illusion of Contingent Identity
to be so. My burden has merely been to defuse Kripke’s objection. An added benefit is that we also see how to close the explanatory gap. Likewise, in the previous chapter I argued that multiple realizability is not a problem because identity theory can accommodate the varieties of multiple realizability worth wanting. Where does this leave us? As I admitted from the start, there is not much room for modification to the core metaphysical doctrine of identity theory: Mental states, processes, events, or properties are identical to brain states, processes, events, or properties. Attempts to tinker with the identity relation invariably compromise the strengths of the theory. This focus on the internal mechanics of the account has made it seem that identity theory has few options for responding to assaults from Putnam, Kripke, and others. But just because the theory appeals to a bare relation it does not follow that the rest of the account must be equally plain. I have held the line on mind–brain identity, offering no qualifications to the relation itself. Instead, I proceeded by denying that an identity theorist is constrained to offer accounts within the traditional bounds. Let’s review. Identity theory identifies mental kinds with biological kinds, say, states or properties, especially those having something to do with brains and neurons. I have argued that we do not know how to individuate brain or mental states. It is an open question what the correct granularity of explanation may be for any kind of mental state. Biological states and properties in general occur at many levels or grains; so we should expect the same of brain and mental states and properties. The identity theory is not committed to the existence of a single level or grain at which mind–brain identities occur. Identity theory does have to restrict itself to species-specific identities, but, I have argued, that is what we should expect. Identity theory nevertheless allows a measure of multiple realizability. To be conscious, a creature need not be physically or physiologically just like us, or anything like us at all. In fact, there may be creatures whose eccentric conscious states have qualities entirely unlike the qualities of our own psychological states. Of course, those creatures may not be capable of having states empathetically like ours, just as we are unable to have sensations empathetically like their eccentric sensations.
70 Chapter 2
Finally, identity theory accepts that the relation between the mental and the physical—identity—is necessary. This is an especially important feature of the theory because it secures instances of mental state kinds the same causal powers as the biological kinds with which they are identified. It also allows some explanatory maneuvers unavailable to functionalists—for example, the dissolution of explanatory gap concerns. The purpose of this first section of the book is only to argue that identity theories are not stuck at the gate, as has often been supposed. This is a modest goal, yet one that set me against two important arguments whose conclusions are often taken for granted in philosophy of mind. All this is to argue that the theory is not already proven false, and not even to argue that it is in fact correct! But I am willing to accept even small successes. If I am right, I have demonstrated that identity theory is at least a viable option. Identity theory is not dead. Even so, it might be that some other theory should be preferred. The main alternatives are versions of functionalism, introduced briefly in the previous chapter. We now turn to them.
3 Varieties of Functionalism
Functionalism is a mess. Whether you are an advocate or an opponent of the view, there can be little doubt about that. Varieties of functionalism are as varied as fingerprints but not nearly so constant. We may say of functionalism what Jerry Fodor says of mental representation: “Every conceivable position seems to have been occupied, along with some whose conceivability it is permissible to doubt” (1985, in 1990: 3). Indicative of this theoretical vertigo is David Lewis’s frank admission, “In view of how the term is contested, I do not know whether I am a ‘functionalist’ ” (1994; see also 1983: xi, n. 5). Despite widespread disagreement about exactly what functionalist theories claim, they have ensconced themselves as the received views in philosophy of mind. I aim to clear things up a bit. In the previous chapters I argued that identity theory is a live option; it has not been shown to be untenable as it is usually supposed to have been. Now we need to see how it compares with the competition. Since functionalism is the main naturalistic alternative to an identity theory, we have to understand what sort of alternative it is. My strategy is first to provide a framework for understanding the many varieties of functionalism and how they are related to one another. How many distinct formulations of functionalism are there? My taxonomy is able to distinguish over one hundred variations; probably there are more. Fortunately, we will need to examine only a small number of them, just enough to get a general feel for the landscape of functionalist views. Only some of these variations compete with identity theory. The purpose
72 Chapter 3
of this chapter is to distinguish the many varieties of functionalism and to identify those whose metaphysical claims are incompatible with identity theory. Specifically, I will argue that there is a distinctive family of functionalist views that is a genuine metaphysical alternative to and is structurally distinguishable from the identity theory. Having isolated metaphysical versions of functionalism, we will be in the position to carefully examine and evaluate the variations. 1 Anatomy of Functionalism To see where we are headed, consider a version of functionalism that William Lycan calls machine functionalism: “To be in a mental state M is to realize or instantiate machine program P and be in functional state S relative to P” (1987: 8). This is a nice version of functionalism to begin with because it is simple and it is clearly statable. Perhaps it is too simple. Hilary Putnam says, “I do not know of anyone who ever actually held the one-computational-state version of functionalism to be right” (1988: 84). Let us not worry about whether machine functionalism is a good theory, nor about how much historical revisionism is at work in remarks like Putnam’s. Let us rather concentrate on the anatomy of machine functionalism (figure 3.1). Machine functionalism is a metaphysical theory that applies to all mental states and that depends on some sort of computational notion of function. (1) Machine functionalism is a metaphysical theory. It is a theory about what it is to have a mental state—what it is for a state of a 1a
2
1b
(MF) To be [in] mental state M is to realize or instantiate machine program P and be [in] functional state S relative to P.
3a
3b
Figure 3.1 The anatomy of machine functionalism. Machine functionalism is a metaphysical view (la and 1b) that applies to all mental states (2) and appeals to a computational/machine notion of function (3a and 3b).
73 Varieties of Functionalism
system to be a mental state. Machine functionalism says that “to be in a mental state . . . is to realize or instantiate” a system of a certain sort, and to be in a system-relative state of a certain sort. To be a mental state is to be a state of such a system. (2) Second, machine functionalism is general in that it applies to all mental states. Machine functionalism is a theory of what it is to be a mental state of any sort at all. Mental states might include at least intentional states and sensations, and (if they are different still) other sorts, for example, moods and psychoses. This is only a catalog; whether any one sort is more basic than the others is a separate matter.1 (3) Third, machine functionalism says that to be in a mental state is to be a system that is specified in terms of the states of a machine program, such as that of a computing device like a Turing machine. So machine functionalism relies on a yet to be specified concept of computational function. In each of these respects other versions of functionalism may differ. Some philosophers who consider themselves functionalists do not make metaphysical claims. Some think, for example, that to explain mental states is to explain the functional role they occupy in the system that has them—but that such a theory need not say anything about the nature of mental states themselves. Other functionalist theories apply only to some kinds of mental states—just to intentional states, or just to sensations. Finally, there is great variation in the sort of function deployed. Versions of functionalism depend on notions of function drawn from mathematics, engineering, or biology. From these differences come the many varieties of functionalism. Together the three aspects determine each variety’s distinctive features. We can codify the different aspects of functionalism in three diagnostic questions that are useful as taxonomic principles: (i) What kind of functionalism is it? What is the theory supposed to be a theory of ? I will describe six possible answers to this first question: Functionalism may be metaphysical, intentional, semantic, theoretical, explanatory, or methodological. (ii) To which mental phenomena does it apply? I am interested in consciousness. But it hardly needs to be mentioned that not all
74 Chapter 3
functionalist philosophers of mind are offering theories of consciousness—or even think that there ought to be a theory of consciousness. (iii) What notion or notions of function does it use? There is a cornucopia of notions of function in the offing these days. Which one is operating in each functionalist account? A function may be understood as abstract or mathematical, in any of several causal variations, teleologically, and so forth. No doubt we could continue to draw more distinctions between versions of functionalism, but these will suffice. I do not have much to say here about the second taxonomic question. I indicated earlier that I take consciousness to be the distinctive mental phenomenon. Most any variety of functionalism could be applied to consciousness, even if it has not been endorsed as such in the past. Although I am mainly concerned with those versions of functionalism that apply to (or can be applied to) at least experiential or qualitative states, on my view that category covers all mental states. This taxonomic question is therefore of less interest to me than the other two, so I focus my attention on them. In this chapter, I try to get the horses into the correct stalls, using the first taxonomic principle as one way to distinguish versions of functionalism. In subsequent chapters, I will pursue the remaining question of which notions of function a functionalist should use, flushing some serious metaphysics through the stables to see what remains. Thus I proceed with the task of cleaning up functionalism. 2 Varieties of Functionalism Not all doctrines about minds compete (or have versions that compete) to answer every question or problem about minds. For example, it is common fare that behaviorism is not one monolithic theory but rather a family of related doctrines: methodological behaviorism, psychological behaviorism, philosophical (logical or analytical) behaviorism, and metaphysical behaviorism. That behaviorism comes in these flavors, and perhaps others, is not controversial.2 Nor is it controversial to point out that the flavors of behaviorism answer different questions and are solutions to differ-
75 Varieties of Functionalism
ent problems; understanding the relations between those problems is a substantial pursuit. John Watson’s (1913) and B. F. Skinner’s (1953) behaviorism began as a proposal for a methodology for the science of psychology, in particular, to respond to positivist concerns about the status of psychology as a science that arise as a result of the privacy of its purported subject matter. Analytic behaviorists in philosophy maintained that mentalistic terms refer to syndromes of behavior.3 “Pain,” accordingly, means the disposition to respond in certain typical ways (yelling, wincing) to certain typical stimuli (pricking, prodding). Moreover, there is no reason to think that all questions about minds are exhausted by just those that versions of behaviorism address; most philosophers would say that they are not. (Whether any of the versions of behaviorism should be accepted is another matter.) So it is with functionalism: “Functionalism should be regarded not so much as a particular view as a general strategy for analyzing mental phenomena,” Georges Rey writes (1997: 165). One way to distinguish versions of functionalism is according to the projects that they participate in, the problems for which they are proposed as solutions. Paul Churchland (1988) distinguishes four projects in philosophy of mind: ontological (I will say metaphysical), methodological, semantical, and epistemological. To those I add two further distinctions. Within Churchland’s “semantical” project I distinguish two kinds of questions, semantic and intentional. And among Churchland’s “epistemological” questions I distinguish theoretical and explanatory projects. Altogether there are six flavors of functionalism distinguished by my first principle, each marked by the distinctive kinds of questions it answers: metaphysical, intentional, semantic, theoretical, explanatory, and methodological. Figure 3.2 is a template for the logical space of functionalisms to be explored in the next several chapters. The aim of this chapter is to be clear about what various functionalists claim, how those claims are related, and which of them is supposed to be a problem for identity theory. Robert Van Gulick comments: Among the theories of mind which have been proposed in recent years, a great many might be termed in one sense or another functionalist. The
ABSTRACT
CAUSAL ROLE Computational
A Priori Psychology
DIRECT- TELEOLOGICAL Empirical Psychology
Etiological
Propensity
HYBRID Hybrid
Semantic Intentionality Theoretical Methodological
Explanatory
FUNCTIONALIST PROJECTS
Ontological
Turing Machine
Figure 3.2 The logical space of functionalisms. Rows are labeled according to different functionalist projects, columns according to varying kinds of functions that can be engaged.
76 Chapter 3
CONCEPTS OF FUNCTION
77 Varieties of Functionalism
many proponents of functionalism have differed greatly both in their formulations of the thesis and about its implications. Functionalism has been put forward as an argument for the psychophysical identity thesis and as a refutation of that view. It has been taken by some to show the reducibility of psychology to physics, and by others as a basis for antireductionism. Still others have found in functionalism support for a disappearance theory of mental terms. (1982: 185)
If I am going to argue that identity theory is a viable—even preferable—alternative to functionalism, I am going to have to say more about which functionalism I have in mind. To that end, each of the subsequent six sections describes a philosophical question that one style of functionalism is proposed to answer, indicates what the significance of the question is taken to be, and outlines the schema for each proposed functionalist solution. I must emphasize that the formulations I offer are only schemas or skeletons of theories; they must be filled out by answering other questions—about functions, functional roles, and instantiation— before any would specify a particular theory. For each I will give some examples of views that fit the schema and address each question. Together the schemas and examples anchor the taxonomy. I hope they will serve to characterize, albeit sometimes only in a rough way, the similarities and differences between various styles of functionalism. But I do not claim to have developed the canonical formulation of each and every variety of functionalism. My purpose is to provide some landmarks in the logical space of functionalisms; but I am not attempting to provide a comprehensive catalog of all functionalist accounts that have been advocated or entertained.4 Similarly, some of the philosophers whom I locate as a functionalists would not themselves adopt that label for their theories.5 And—this is especially true of my discussion of functions in the next chapters—I am sometimes extending views beyond the scope that their inventors explicitly endorse. I hope I will not thereby seem to be tilting at windmills. My goal is to exhibit the full breadth of functionalist theories while illuminating their common elements. At the end of the day it will be metaphysical functionalism that is of interest to me. It is the one that makes claims about the nature of minds, and thus it is the one that competes with identity
78 Chapter 3
theories. Lets us therefore begin our investigation with metaphysical functionalism. Metaphysical Functionalism Churchland says that the metaphysical (“ontological”) question in philosophy of mind is the mind–body problem. But it is common to note that “the mind–body problem” does not pick out a single timeless question (see, e.g., Kim 1998). Gilbert Ryle (1949), for example, thought he was dissolving the mind–body problem, yet it would be a mistake to call his view ontological or metaphysical. If there is a metaphysical mind–body problem, then it must be distinguished from other mind–body problems—for example, problems about the compatibility of mental and physical concepts that Herbert Feigl (1958), J. J. C. Smart (1959), and David Armstrong (1968) spent much ink on.6 The central metaphysical questions in philosophy of mind are those such as: What sort of thing is a mind? What sorts of things are mental states, processes, properties, or events? That is, they are questions that ask for an explanation of what it is to be a mental state, property, process, event, or entity. Jaegwon Kim has come to use the notion of supervenience to state the mind–body problem in terms of mental and physical properties: “Mental properties supervene on physical properties, in that necessarily, for any mental property M, if anything has M at time t, there exists a physical base (or subvenient) property P such that it has P at t, and necessarily anything that has P at a time has M at that time” (1998: 9). The metaphysical mind–body problem is the problem of explaining why—in virtue of what—the mind–body supervenience relation holds. Reversing his earlier (e.g., 1993) claim that supervenience is the key to solving the mind–body problem, Kim argues, “The mere claim of mind–body supervenience leaves unaddressed the question what grounds or accounts for it—that is, the question why the supervenience relation should hold for the mental and the physical. . . . If this is right mind–body supervenience states the mind–body problem—it is not a solution to it” (1998: 13–14). An answer to the metaphysical mind–body problem—to the metaphysical question—demands an
79 Varieties of Functionalism
explanation of what grounds or accounts for the supervenience of mind on body. Supervenience, although it puts some constraints on theories of mind, “is not a metaphysically ‘deep’ relation” (ibid.: 14). The metaphysical mind–body problem demands an answer that begins, “To be a mental state is. . . .” Cartesian dualism is one such answer. It says that to be a mental state is to be a state of an immaterial, nonextended substance, res cogitans. Identity theory is another proposed answer to the metaphysical question. It says that to be a kind of sensation is to be identical to a kind of brain process: “Sensations are nothing over and above brain processes” (Smart 1959: 103). Other answers to the metaphysical question must be competitors with Cartesian dualism and identity theory. Hilary Putnam first proposed functionalism in just this way, as a superior alternative to identity theory: “I shall, in short, argue that pain is not a brain state, in the sense of a physical-chemical state of the brain (or even the whole nervous system), but another kind of state entirely. I propose the hypothesis that pain, or the state of being in pain, is a functional state of a whole organism” (1967, reprinted in Putnam 1975c: 433). The functionalist claims that a mental state M is entirely constituted by its relations to inputs, outputs, and other mental states. But if those other states are not all themselves functional or physical states, then although M is a functional state in a sense, it is not the sort of state that metaphysical functionalism is concerned with (Shoemaker 1981b). M would be a functional state only in a “weak” sense; for in that case only some mental states would be functional states, and others would be nonfunctional. The point can be made in terms of identity conditions of the sort discussed in chapter 2: “The stronger functionalist claim is that type identity conditions for mental states can be specified purely in terms of their mutual interconnections and relation to stimulus conditions and behavior. The specification of a psychological state’s role within a theory of behavior is held not merely relevant to its explication, but sufficient as well” (Van Gulick 1982: 186; see also Shoemaker 1981b). A state is a functional state in the strong sense only if all the states to which it is related are physical states or are themselves functional states in the strong sense. Metaphysical functionalism claims that mental states are functional states in the strong sense. On the
80 Chapter 3
present analysis, the “weak” versions of functionalism are not in some way less strict versions of metaphysical functionalism; only “strong” functionalist doctrines are theories of metaphysical functionalism at all. Putnam’s work is the undisputed locus classicus for metaphysical functionalism.7 We’ve already seen a basic sketch of Putnam’s proposal: “Encouraged by the fruitfulness of comparing humans and other sentient organisms to computing machines . . . Putnam envisioned a theory of mind whose explications of individual mental state-types would take the form ‘To be in a mental state M is to realize or instantiate machine program P and be in functional state S relative to P’ ” (Lycan 1987: 8). The form of functionalism that Putnam proposed is machine functionalism, our stepping stone above. Functionalist answers to the metaphysical question are going to share their general form with machine functionalism. We need only add the qualification that the metaphysical claim requires that mental states be functional states in the strong sense, while leaving it open that the system itself could be specified by some means other than a machine program. This yields a general schema for metaphysical functionalism: (Fm) To be a mental state M of system S is to realize functional state (in the strong sense) F relative to S, or to S and its environment.8 Anyone who assents to a view of this form is espousing a metaphysical claim. If you do not wish to be making a metaphysical claim, you should not agree to views fitting schema (Fm). One reason that someone might believe in a version of metaphysical functionalism is, as Putnam himself suggests, that it is simply more plausible than the alternatives: “my strategy will be to argue that pain is not a brain state, not on a priori grounds, but on the grounds that another hypothesis is more plausible” (1967, in Putnam 1975c: 433). Another view holds that metaphysical functionalism is the only live option among proposed candidates. Fodor recognizes this argument in his earlier writings: “Functionalism in the philosophy of mind teaches that believing that such-and-such is a functional state; for, it argues, since dualism and type physicalism are false, func-
81 Varieties of Functionalism
tionalism is all that’s left, and whatever is all that’s left must be true (see, for example, Fodor 1968, where this line of argument is pursued interminably)” (1986, in Lycan 1990: 430). A third reason for holding a version of metaphysical functionalism might be that the analogy of the mind with a computing machine is independently attractive. Metaphysical functionalism is the heart of the traditional project of artificial intelligence (AI). In particular it is part of the view that John Searle targets with his famous Chinese Room thought experiment, a view he calls “Strong AI”: “[A]ccording to Strong AI the computer is not merely a tool in the study of the mind; rather the appropriately programmed computer really is a mind in the sense that computers given the right programs can be literally said to understand and have other cognitive states” (Searle 1980: 417). Strong AI, as characterized by Searle, aims to replicate minds by showing how to build things that genuinely have mental states. It is therefore a version of metaphysical functionalism and in fact seems to be a version of machine functionalism. Of course, not all metaphysical functionalists formulate their theories in terms of computing machines. Even so, machine functionalism has a central place in the history of functionalism, and it remains the model for metaphysical functionalism. Many philosophers would gladly agree that (Fm) at least roughly sketches the theory of the nature of mental states that they endorse. But it is difficult to find philosophers who expressly defend formulations whose metaphysical commitments are as transparent as those of (Fm). Insofar as formulations like (Fm) are endorsed, they are usually offered as analyses of what we might call the standard functionalist formulation: (SF) Mental states are functional states. Putnam makes this sort of statement (“pain, or the state of being in pain, is a functional state of a whole organism”), and examples similar to the standard formulation are quite common: The Lewis–Stalnaker approach to belief involves the idea that believing is a functional state. (Field 1978, in Block 1980b: 88) [T]he starting point of my paper . . . is the truism that the mind is a functional device. (Sober 1985: 165)
82 Chapter 3
A third strategy is to embrace functionalism with respect to phenomenal states. . . . Phenomenal states, like mental states generally, have a functional character. (Tye 1995: 63)
These sorts of claims are familiar. If the endorsement of the standard formulation counts as endorsement of metaphysical functionalism, then metaphysical functionalism is a widely held view indeed. So the question we must ask is: Is the standard formulation equivalent to (Fm)—to metaphysical functionalism? The answer is that it is not. The standard formulation is ambiguous between metaphysical functionalism and other versions that make no distinctive metaphysical claims at all. Metaphysical functionalism claims more than just that mental states are functional states. Metaphysical functionalism claims that mental states are essentially functional states, and that anything with such-and-such distinctive functional states thereby has mental states. The standard formulation is compatible with those claims of metaphysical functionalism, but it does not itself include the stronger qualifications. We can explain the difference between the two ways of understanding the standard formulation as follows: Mental state terms may be analyzed by their functional relations, or the functional relations may simply fix the reference of mental state terms.9 In the former case, claims like “pain is functional state Fp” are treated as definitional and are necessary de dicto. The functional specification, then, gives the nominal essence of pain and thereby makes a metaphysical claim about the property of being a pain. It tells us the identity conditions for being a pain: The property of being a pain is identical to the property of having relations Fp. On the other hand, Fp may not give an analysis of pain but merely a reference fixing description. If Fp is a heuristic description that leads us to some particular brain state, then “pain is functional state Fp” asserts the de re necessary identity of pain with that state which in fact satisfies Fp. But in this latter case no distinctive claim is asserted concerning the nature of the state that is picked out by Fp. In particular, pain may only contingently satisfy Fp. The de re functionalist specification is therefore entirely compatible with identity theory. Indeed, the identity theorist would expect that, as with other scientific identities, the identification of conscious mental states with particular brain states will
83 Varieties of Functionalism
in practice proceed by means of exactly such reference-fixing descriptions. The standard formulation masks the distinction between metaphysical functionalism and its nonmetaphysical cousins because it is ambiguous between reference fixing (de re) and analytic/ definitional (de dicto) readings. The standard formulation may be understood (de re) as claiming that mental states are identical to the particular states that happen to realize thus-and-such functional states; or it may be understood (de dicto) as claiming that mental states are identical to whatever states realize thus-and-such functional states. Whether we have a reference-fixing or definitional account seems to depend on the functional specification (Fp) and how it is used. This means that functionalists need to assert more than just the standard formulation mantra for us to be able to determine whether they are metaphysical functionalists. Consider some familiar kinds of examples that don’t involve conscious mental states. Cal Ripken, Jr., is the record holder for most consecutive games played in Major League Baseball. This might mean that Cal Ripken, Jr., is identical to the person who, in fact, holds the record for most consecutive games played—which is true. But it also might mean that whoever holds the record for most consecutive games played is Cal Ripken, Jr.—which is not true; someone else could hold the record. It used to be Lou Gehrig who held the record, and someday someone might break Cal’s record. The description “record holder for most consecutive games played in Major League Baseball” only fixes the reference of the name “Cal Ripken, Jr.” Cal Ripken, Jr., is, de re, the record holder for most consecutive games played in major league baseball. But being Cal Ripken, Jr., is not identical to being the person who holds the record for most consecutive games played. They are not identical states of being; the property being Cal Ripken, Jr., is not identical to the property being record holder for most consecutive games played in Major League Baseball. Put another way: The identity conditions for Cal Ripken, Jr., and record holder for most consecutive games played in Major League Baseball are not the same. Likewise, nine is, de re, the number of planets in our solar system. Nine is not identical to whatever the number of planets happens to
84 Chapter 3
be; it is the number of planets that there are, in fact. Recently astronomers considered demoting Pluto to “minor planet” status. Although this caused some public outcry, it did not especially threaten mathematicians. If Pluto were not a planet, nothing would be different about the number nine.10 Being nine is not identical to being the number of planets. Now consider a different sort of example, suggested by Lycan (1987: 19). The president of the United States of America is the person who gets the most votes in the electoral college. This is true of every particular president since the Constitution was ratified, and it will be true of every president unless or until the Constitution is changed. There is not just one person who is the president of the United States; whoever gets the most votes in the electoral college is the president. Being the president of the United States is identical to being the person who gets the most votes in the electoral college. The president of the United States is, de dicto, the person who gets the most votes in the electoral college. The property being president of the United States of America is identical to the property being the person who gets the most votes in the electoral college.11 Now consider the standard formulation of functionalism: mental states are functional states. We can see that it is ambiguous—it may be interpreted in either way, de re or de dicto (reference fixing or definitional). So the standard functionalist formulation is not equivalent to my sketch of metaphysical functionalism. This is because (Fm) is not ambiguous between de re and de dicto interpretations; (Fm) says that being a mental state (“to be a mental state”) is identical to realizing (“to realize”) a functional state. (Fm) is the de dicto interpretation of the standard formulation. Metaphysical functionalism says that mental states are like presidents. It says that mental states are—de dicto—instantiations of functional states. (Fm) is an answer to the metaphysical question; it tells us what it is to be a mental state. Whatever realizes the appropriate kind of functional state is a mental state; mental state kinds are functional state kinds. In contrast, the de re reading of the standard formulation does not make any distinctive metaphysical claim. It merely says that mental states are, as a matter of fact, functional states. This claim is metaphysical insofar as it is committed to the existence of mental
85 Varieties of Functionalism
states; but it does not assert any claim about the nature of mental states, about what it is to be a mental state. The de re reading is compatible with Cartesian dualism: mental states, as a matter of fact, realize functional states; but what makes some states mental states is that they are nonphysical states of an immaterial substance. This would be one form—a peculiar form, to be sure—of what Georges Rey (1997) calls “anchored” functionalism. In short, the de re reading of the standard formulation is compatible with mental states being functional states merely in Van Gulick’s and Shoemaker’s weak sense. The mere claim of de re functionalism leaves unanswered the question of why mental states occupy functional roles. That is the question that metaphysical functionalism claims to answer. But de re functionalism is not itself a view about the nature of mental kinds. It is not metaphysical functionalism because it does not ensure that mental states are functional states in the strong sense. If I am right, although metaphysical functionalism can be stated using the standard formulation, the standard formulation does not by itself entail metaphysical functionalism. One reading of the standard formulation is metaphysical functionalism, namely, the de dicto reading. The other, de re, reading is not a version of metaphysical functionalism at all. This nonmetaphysical reading is nevertheless important; it expresses claims that are best understood as intentional, explanatory, theoretical, or methodological. Intentional Functionalism Churchland (1988) says that what he calls the “semantical” problem in philosophy of mind has to do with the meanings of mentalistic terms. This problem, that of the meaning or reference of putatively mentalistic terms (“belief,” “desire,” “feel,” “pain”), should be distinguished from the problems of the specific content of some mental states or the contentfulness of mental states generally. In one way, questions about the meaning of mentalistic terms are run-of-the-mill semantic questions, just like questions about the meaning or reference of any linguistic tokens. And if thoughts are words or other representational items (e.g., of a “language of thought”) then there will be semantic questions about their content,
86 Chapter 3
just as there are about the content of public linguistic vehicles. Insofar as we are asking these questions about what is sometimes called “psychosemantics,” we’re asking what I call semantic questions; these I discuss in the next section. But mentalistic language is also distinctive. After all, it purports to refer to the sorts of things whose metaphysical status is open to question. This presents a difficulty different than other potentially nonreferring expressions (like “unicorn”), in part because it is unclear to what degree mentalistic terms are unlike other bits of language in being embedded in some theory of psychology, whether “scientific” psychology or so-called folk psychology. (It is unclear to what degree mentalistic terms are embedded thus; and it is also unclear whether or to what degree everyday linguistic expressions are likewise theory-embedded.) If this sort of problem is what one is concerned with regarding mentalistic terms, then it is more useful to think of the question as part of one of the epistemic projects, namely, the theoretical project. It is one of a family of questions about the content of theories and the referents of theoretical terms, to be discussed below. Churchland groups both of these sorts of questions about mentalistic terms as “semantical” problems because his classification does not distinguish semantic from intentional problems, and his framework does not have an alternative place for them as epistemic questions. He construes epistemic questions about mind narrowly, as problems of other minds and of introspection; but those are theoretical or explanatory questions in my framework.12 Churchland also groups questions about the particular contents or meanings of mentalistic terms with questions about the contentfulness or intentionality of mental states generally. But we can distinguish the question of what gives a word or a mental state a particular semantic content (the semantic question) from what endows it with the capacity, in general, to bear any content at all (the intentional question). Of course the questions may be linked; or it may be that we should prefer a theory that jointly answers them. It is the content-bearing mental states that concern us presently. Contentful mental states are sometimes called “propositional attitudes,” for example by Jerry Fodor (1985). The problem of the con-
87 Varieties of Functionalism
tentfulness of mental states is indeed a central—according to some, the central—problem in philosophy of mind. It is more commonly known as the problem of intentionality, owing to Brentano. The intentional question is: How can a state, property, process, event, etc. have the capacity to be about something? Answers to this question should explain what it is about a state, process, event, etc. that accounts for its aboutness or directedness, its intentionality.13 Intentional functionalism is a doctrine about how mental states can have any intentional content. The intentional functionalist answers this question by offering an explanation that fits the schema: (Fi) For a mental state M of system S to have (or be capable of having) intentional or representational content C is for M to occupy functional roles {R1, . . . , Rn}. Fodor seems to have intentional functionalism in mind when he writes that to be a functionalist is “to hold that to believe it’s raining is to have a token of ‘it’s raining’ play a certain functional role in the causation of your behavior and of your (other) mental states, said role eventually to be specified in the course of the detailed working out of empirical psychology . . . , etc., etc. This is, perhaps, not much of story, but it’s fashionable, I know of nothing better, and it does have the virtue of explaining why propositional attitudes are opaque” (1978, in Block 1980b: 53–54). The schema (Fi) may be filled by narrow (“internalist”) or wide (“externalist”) accounts, depending on different notions of function and of functional role. On a narrow account, a mental state has the capacity for bearing content if it occupies a certain role in the agent’s (system’s) cognitive economy. On a wide version, a mental state has the capacity for bearing content if it occupies a certain role in the agent’s interaction with its environment, even distal portions of its world. A particularly wide version of intentional functionalism that has a growing following depends not only on current relations to the world, but also on historical relations: “the semantic category of a thought is determined relative to its biological functions, which depend in turn on its history, upon its place relative to certain prior events” (Millikan 1993: 186). (The relative merits of wide and narrow functionalist conceptions are often contended with respect to semantic functionalism.) Intentional
88 Chapter 3
functionalism is frequently held in conjunction with semantic functionalism, and we’ll see several examples of this combination as we proceed.14 A black sheep case of intentional functionalism needs mention: Daniel Dennett’s “intentional stance” approach. On Dennett’s view, a system has intentional states just in case it is useful in predicting the system’s behavior to treat it as having intentional states (1971, 1987). Often this view is treated as being antirealist or instrumentalist about intentional states (see, e.g., Fodor 1985). But Dennett himself is not so clear, complaining that the question of whether he is an instrumentalist is “poorly conceived” (1991: 460).15 The form of the prediction and explanation that Dennett has in mind when he describes his three stances is broadly functional, and so in a way the intentional stance amounts to a deflationary variety of functionalism concerning intentional states. Systems that have a certain kind of functional organization are usefully treated as having intentional states. This reading of Dennett is controversial, especially since Dennett makes a point of rejecting the “deeper” question of whether all systems that we treat as intentional “really” have intentionality— precisely the question that intentional functionalism means to answer. But I have argued elsewhere (2000, 2001), and will argue again in chapter 7, that we should interpret Dennett’s apparently instrumentalist remarks not as analyses of, for example, intentionality, but as stating empirical criteria for intentional systems. Of course, if the “really” question about intentionality is thought of as metaphysical, Dennett will resist. But it is by no means clear that intentional functionalism, by itself, has any metaphysical consequences. Recent debates over functionalism have centered around intentional functionalism. If one accepts Brentano’s thesis that “intentionality is the mark of the mental” then one will hold that to understand how a state can have intentional content is ipso facto to understand how it can be a mental state. On this view, any answer to the question of intentionality will thereby be or yield an answer to the metaphysical question. Consider some representative statements of this approach: [T]he claim that “intentionality” won’t be reduced and won’t go away . . . has sometimes been called “Brentano’s thesis,” after a philosopher who is
89 Varieties of Functionalism
(not completely accurately) credited with defending it in the latter part of the nineteenth century. Sometimes the view is stated as a positive claim: the claim that intentionality is a primitive phenomenon, in fact the phenomenon that relates thought and thing, minds and the external world. (Putnam 1988: 1–2) The position I defend is an intentionalist one: phenomenally conscious states are essentially representational states of a certain sort. (Tye 1995: 66) I am concerned to maintain a weak version of Brentano’s doctrine that the mental and the intentional are one and the same. (Lycan 1996: 11) [M]y larger project is to secure the hegemony of representation, the doctrine that the mental and the functional/intentional are one and the same and that the mind has no distinctive properties that outrun its functional and intentional properties. (Ibid.: 69)
Brentano’s doctrine has provided a common way of approaching the metaphysical question for some time now. But the intentional and metaphysical questions need not be so related. Intentional functionalism by itself does not entail metaphysical functionalism. That result depends on the additional, substantial claim that intentional states are ipso facto mental states, a version of Brentano’s thesis. In the introduction I stated my opposition to Brentano’s thesis; but that argument is beyond the scope of this book. My view on this matter is no doubt relevant in assessing the overall theory that I am advocating; but it will not be important in evaluating my critique of metaphysical functionalism. The argument against metaphysical functionalism that I advance in the next chapters is entirely independent of any reasons for holding metaphysical functionalism, so it does not depend on my rejection of Brentano’s thesis. And none of what I will have to say takes a stand on the adequacy of intentional functionalism in its own right. Semantic Functionalism Intentional functionalism is a doctrine about what makes some states or items capable of bearing meaning or being directed toward some (intentional) objects. Semantic functionalism is a doctrine about what fixes the specific meaning, content, reference, or objects of words, or of mental states or tokens, such as symbols in internal computations or words in a language of thought. Semantic
90 Chapter 3
functionalism provides one possible semantic theory for words, or for mental states (in which case we can say that it provides one possible psychosemantics). Whether the question of semantics can be settled independently of the question of intentionality is controversial. Jerry Fodor thinks that a state’s status as a representation and its representational content are determined separately. But this seems to allow the possibility that something could be a representational state yet lack a particular content. Ruth Millikan says—incredulously—that on Fodor’s view, “[t]here is such a thing, it seems, as behaving like a representation without behaving like a representation of anything in particular” (1993: 84). Millikan (1993) and Fred Dretske (1988, 1995) hold that what makes a state a representation is precisely that its semantic content is fixed in a certain way.16 In contrast, Lycan (1987, 1996) is drawn more in the direction of Fodor. What makes something an intentional state is its functional role, but what fixes its semantic content is a separate matter. Lycan entertains several different psychosemantics. He views the question of semantics as secondary—he’ll accept whatever semantics he needs to make his views on intentionality and consciousness work (see especially Lycan 1987, 1996). Millikan and Dretske, but not Fodor and Lycan, are examples of philosophers who look to functions to answer the semantic question. The semantic functionalist answers questions about the content or reference of mental states and items as follows: (Fs) The content or reference O of a mental state M is determined by the functional roles {R1, . . . , Rn} occupied by M in a system S, either with respect to (i) the internal cognitive economy of S or (ii) the interaction of S with its environment. (Fs) is the doctrine that could be called, speaking rather loosely, functional-role semantics.17 Grossly put, my “tree” thoughts are about trees because those thoughts play the role of mediating my actions and thoughts in response to actual and possible trees in the world. Some would make the qualification that my thoughts about trees normally mediate my tree-related behaviors.18 Or, for example: “What makes a pattern of activity occurring in my visual cortex the per-
91 Varieties of Functionalism
ception of a small red cylinder is that under normal conditions it is produced only by objects having those properties, and it can guide my behavior relative to such an object in an appropriate way should I need to grasp it, roll it, fill it with water, or pick it out from among other objects differing from it in colour, shape, or size” (Van Gulick 1988b: 90). A merit of the normalized way of formulating semantic functionalism is that it allows that one could have red cylinder behaviors without the presence of a red cylinder and that, say, some tree thoughts could fail to mediate tree behaviors. The question of narrow and wide functionalist accounts, mentioned with respect to intentional functionalism, arises in full force for semantic functionalism. According to the narrow view, the particular semantic content of a mental state is determined by its functional role in the system of which it is a part. On this view, “preserving isomorphism between causal and semantic networks is all that there is to the assignment of contents to mental states” (Fodor 1985, in 1990: 15). On the wide account, the semantics of a mental state “is not just its functional role but also its causal connections to objects ‘in the world.’ . . . It is these mind-to-world causal relations that are supposed to determine the denotational semantics” of an intentional state (ibid.).19 On my way of talking, the “width” may also be temporal; the content of a state may depend on its past or future (Millikan 1984; Dretske 1988). The connection between intentional and semantic functionalism is a matter of great dispute.20 Although he rejects the view, Fodor presents functional-role semantics as part of standard realism about intentional states (1985, in 1990: 15). Likewise, Van Gulick writes, “Considerations of functional role are held to differentiate not only among general psychological state types such as beliefs, desires, and intentions, but also among subtypes such as believing that p and believing that q. A state has whatever content it does on the basis of its functional role” (1980, in Lycan 1990: 108). In contrast, Shoemaker finds it obvious that no one would hold such a view: “Of course, no functionalist would maintain that each different belief and each different want must be defined separately; in the case of belief, for example, the functionalist will want a definition of ‘S believes that P’ which holds for all values of ‘P’ ” (1981b, in 1984: 262, fn. 3).
92 Chapter 3
Shoemaker has in mind that psychological theories will not formulate explanations of belief that P, belief that Q, and so forth independently (with respect to theoretical functionalism, discussed in the next section). But the fact remains that at least some philosophers seem to be committed to the view that the content of each belief and want would have to be fixed separately in the system’s causal or historical network. Perhaps the most elegant approach is to jointly solve the problems of intentionality and semantics. Fred Dretske proposes: “d’s being G meansf that w is F = d’s function is to indicate the condition of w, and the way it performs this function is, in part, by indicating that w is F by its (d’s) being G” (1986 in Lycan 1990: 133). On this view, what gives a state, d, the capacity to be intentional or contentful at all (in a functional sense, “meaningfulf”) is that it has the function of bearing some particular information, G.21 Just as there is controversy about the relation between intentional and semantic functionalism, so too is there controversy over the connection between semantic and metaphysical functionalism. Van Gulick contends that there is a connection: “States, including contentful states, are held to be of a given psychological type in virtue of their functional role within such a network. Thus, the functionalist is obliged to explain how content is to be unpacked by reference to functional role” (1980, in Lycan 1990: 107). But Fodor disagrees. Although he qualifies as a metaphysical functionalist and an intentional functionalist, Fodor rejects functionalist semantics for mental states: “The (usually tacit) assumption that Functionalism in philosophy of mind somehow comforts—or even implies—functional role semantics is responsible for no end of confusion. In fact, the best you can get from Functionalism is that what makes something a belief is its functional role. Functionalism does not certify that functional role makes a belief the belief that P” (1986, in Lycan 1990: 430).22 Notice that Fodor seems to concede that metaphysical functionalism may at least entail intentional functionalism (it is “the best you can get”), if not semantic functionalism. Likewise, Robert Cummins takes the connection between metaphysical and semantic functionalism to be part of the standard functionalist picture, but he too questions its validity:23
93 Varieties of Functionalism
The conceptual-role and functional-role approach to representation is the natural by-product of a familiar “quick and dirty” argument for functionalism about mental content. One begins with an ontological question: What is it in a person’s head—my head, say—in virtue of which my states are contentful? On what does content supervene? The answer provided by the [computational theory of cognition] is relatively straightforward: I am a thinking system in virtue of the computational structure of my brain. Thoughts, and contentful states generally, must be what they are in virtue of their place in that structure. But a thought is what it is in virtue of its content. So having a content . . . must be a matter of occupying the right role in a computational structure! Right? (1989: 123)
The trouble, according to both Fodor and Cummins, is that functional-role semantics “provide us with no hint as to why being a node in a network of computational or causal relations should make something a representation or endow it with particular content” (ibid.: 122–123).24 In reply, Dretske and Millikan seem to be arguing that their accounts provide more than a hint; in fact they provide complete explanations of exactly that relation. But to settle that issue would take us too far from our path. Theoretical Functionalism Theoretical functionalism is a special case of semantic functionalism. It is a doctrine about the structure of theories and the terms of those theories, particularly of psychological theories. Theoretical functionalism holds that the correct theory of psychology is one in which all the significant terms of the theory are jointly specified by their functional roles in the theory. In short: (Ft) The content of a given mental term or predicate P of theory T is determined by the functional roles {R1, . . . , Rn} that P occupies in T. The full import of any functionalist theory will depend on the functional relations that determine the roles—whether they are wide or narrow, whether they are functional in the strong or weak senses, and so forth. The canonical formulation of theoretical functionalism comes from David Lewis (1970, 1972), who also provides this kind of
94 Chapter 3
functionalism with its characteristic methodological device: the Ramsey–Lewis method, or “Ramsification.” As Shoemaker summarizes the technique:25 One starts off with a theory that incorporates propositions stating all of the causal facts about mental states—about their relations to inputs, outputs, and one another—in terms of which one proposes to define them. One then constructs the Ramsey-sentence of this theory, which says that there is a set of states satisfying the open sentence which results from the replacement of the psychological terms in the original theory by variables. . . . From the Ramsey-sentence . . . one can then extract noncircular definitions of each of the mental terms that figured in the original theory. (1981b, in 1984: 261)
Ramsification accomplishes several objectives. First, it demonstrates that a theory of psychology can be stated without the use of any distinctly mentalistic terms. This is supposed to show that the theory is at least “topic-neutral” (Smart 1959) in its metaphysical commitments, that it is not committed to any particular ontological kinds. Second, by interdefining all of the terms of the theory together, we cut off concerns that the interconnecting functional definitions could turn out to be viciously circular. Finally, it recovers some of the specificity and formality that is sacrificed in the move away from Turing machine formulations of functionalism, without giving up the generality gained by departing from the computing model. (This point will be considered in detail in later chapters.) A machine table can be thought of as a special kind of functional theory that applies to certain devices. Despite the rigor of theoretical functionalism and its widespread use as a device for explaining the notions of functionalism in general, it is a doctrine with some notable baggage. Theoretical functionalism, as described by Lewis and others, appeals to only one, specific notion of function: The relations between states on which Ramsification relies to implicitly define theoretical terms—the mentalistic terms in the “old” psychological theory—are almost always causal relations. Perhaps one could apply the Ramsey–Lewis method to a theory that was functional in some other sense; Lewis mentions in passing the possibility that the relations be other than causal relations (1972, in Block 1980a: 210). But it’s none too clear how Ramsification of heterogeneous or noncausal functional expressions
95 Varieties of Functionalism
would proceed, and at any rate there has been no serious discussion of that option in the literature. Ramsified theories, at least those that have actually been entertained, appeal to a causal notion of function. Of course, if the causal role notion of function works, then such worries are academic. It is true that many who discuss functionalism assume that the causal-role notion of function is what matters. For example, Kim writes, “[a]ccording to functionalism, a mental kind is a functional kind, or a causal-functional kind, since the ‘function’ involved is to fill a certain causal role” (1996: 76). But the causal-role notion of function is not the only notion of function available, or even the only one mobilized by functionalists. So we must not presuppose it is the only or correct account of function, and we ought to understand functionalism in a way that at least allows other accounts of function to be candidates. Moreover, theoretical functionalism does not by itself guarantee that the resulting theory will be a naturalist or physicalist theory. One of the goals of Ramsification is to express a theory in terms that are neutral regarding the underlying ontology, so-called topic-neutral terms. This is sometimes said to be a merit of such theories; but if we are after a naturalistic theory, then functional theories do not secure all that we want. If a functional theory of psychology involves a ghostly res cogitans, then the Ramsified theory will involve it as well. Ramsification, recall, begins with a psychological theory expressed as a conjunction of sentences; one then proceeds by making the permissible substitution of existentially quantified terms in place of the psychological terms. We replace, “p causes the Id to produce effect q” with “there is an x such that p causes x to produce effect q.” But that does not rid the theory of its ontology: “if the theory is only existentially quantified, then it says nothing about the metaphysics of the states quantified over” (Kim 1996: 106). We can treat a functional theory as telling us about basic ontology, as Lewis does; but such claims come not from the theory alone but from the combination of the functional theory and an additional doctrine about the meanings of theoretical terms. Finally, the holism that allows theoretical functionalism to avoid problems of definitional circularity also introduces concerns about the theory’s applicability. For a system to be attributed even one
96 Chapter 3
mental state described by the theory, the entire theory must be true of it. If the entire theory does not apply to the system or creature in question (perhaps because creatures of that type do not experience anxiety, say) then we cannot attribute to that system any of the states defined in terms of the theory. Kim has the right reaction to this consequence of holistic functionalism of this sort: “This sounds absurd: It does not seem to require that for two persons to share a mental state, say, the belief that snow is white, the total set of logical regularities governing their behavior be exactly identical” (1996: 92–93). The flip side of this concern is probably more familiar and is often raised with respect to the question of which theory ought to be Ramsified. We want to Ramsify the right (correct? true? best? ideal?) psychological theory, for if the Ramsified theory contains any inaccuracies then all the resulting definitions will be wrong: “If our T is to yield our psychological concepts all at once, it is going to be a long conjunction of myriad psychological generalizations, and even a single false component will render the whole conjunction false” (ibid.: 108–109; see also Shoemaker 1981b; Block 1980c).26 The question of which psychological theory should be Ramsified is a persistent source of controversy surrounding theoretical functionalism.27 Should it be scientific psychology (Putnam 1967)? Or “folk psychology”—ordinary persons’ commonsense “platitudes” about minds (Lewis 1972)? Perhaps just the analytic and conceptual truths about minds (Shoemaker 1981b)? One option is to Ramsify a theory of scientific psychology, ideally the best or completed theory of scientific psychology. This is the view that Block (1978) dubs “psychofunctionalism.” Putnam (1967) is usually counted as a psychofunctionalist, as are Lycan (1987, 1996) and Rey (1997), who explicitly applies the label to himself. Shoemaker explains, “The Psychofunctionalist . . . holds that just as chemistry can tell us the real essence of gold or water, psychology (once it is developed further) can tell us the real essence of pain, anger, belief, desire and thought” (1981b, in 1984: 279). Most functionalists, whether or not they make use of the Ramsey–Lewis method, are psychofunctionalists.28 Alternatively, one may choose a commonsense psychological theory, offering the Ramsified theory as analysis of the meanings of
97 Varieties of Functionalism
the mentalistic terms in our ordinary discourse. This is what Block (1978) calls capital-“F” “Functionalism”; it is sometimes also called “analytical functionalism.”29 One might then choose to treat all of the “platitudes” of the so-called folk psychology as the theory to be Ramsified (Lewis 1972), or one might restrict the theory to the analytic and conceptual truths of commonsense psychology (Shoemaker 1981b). J. J. C. Smart’s topic-neutral analyses (1959) seem to be the locus classicus for analytical functionalism, and Armstrong (1968) is usually interpreted as having held this view. But the decisive statement of it is surely that of Lewis (1972). Whatever theory one begins with, by depending on the Ramsey–Lewis method, theoretical functionalism appears to purchase a degree of rigor and clarity—it’s quantified formal logic, after all. Nevertheless, theoretical functionalism has been the locus of a great deal of confusion and controversy over the years. A striking example is the debate that raged for some time over whether or not theoretical functionalism is compatible with identity theory (see, e.g., Block 1978). The sources of the confusion are repeated in recent discussions of functionalism, including Block (1978, 1980c), Kim (1996), and Rey (1997). Each of those authors discusses Ramsification and Ramsified functional psychology in the context of metaphysical functionalism. But if I am right, there is nothing inherent in functionalist theories in general or Ramsified theories in particular that gives them metaphysical weight. Some Ramsified theories make claims about the natures of mental kinds, but when they do it isn’t because of their Ramsified form. As noted above, a Ramsified theory says nothing distinctive about the nature of the items whose names are replaced with existentially quantified variables. The resulting theory is neither more nor less metaphysically functionalist than the theory that is Ramsified; a Ramsified theory “will have the same form as the psychological theory from which our Ramsey-sentence was derived” (Shoemaker 1981b, in 1984: 264). It is easy to see why this is the case. The Ramsey–Lewis method begins with a causally specified theory of psychology. We then replace all the mentalistic terms in the theory with existentially bound variables. But those mentalistic terms will be predicates, so the predicates of the theory are now existentially quantified. But since Ramsification changes only
98 Chapter 3
the theory’s predicates, the objects—states, say—of the theory retain whatever quantification they had in the original theory prior to Ramsification. In particular, the Ramsified theory does not dictate whether or not those states are essentially functional. To get metaphysical functionalism, we need to have a theory that universally quantifies over the objects of the theory. Earlier I argued that to be a version of metaphysical functionalism a theory must treat mental states as essentially functional; it must take the functional specification as de dicto functional, as an analysis of the relevant mental property. This is a matter of how we apply the psychological theory. Mere Ramsification will not make a reference-fixing (de re) functional theory into an analytical (de dicto) theory.30 If the original, pre-Ramsification psychological theory is a de dicto theory (i.e., if it is committed to metaphysical functionalism) then— regardless of whether it is a theory composed of platitudes, analytic truths, or scientific jargon—the resulting Ramsified theory will also be a de dicto theory and make metaphysical claims. This is the situation that Shoemaker assumes when he says that functionalism and psychofunctionalism “do not differ in the sorts of ontological claims they make” (1981b, in 1984: 272). The best that can be said for this approach is that it is by no means clear that our commonsense or analytic “folk psychology” theories are metaphysically functionalist. Scientific psychology is a better candidate for a theory that would give the real essences of mental states. Even so, it would take an argument to show that scientific psychology makes such claims.31 Lewis gives such an argument. He maintains that “theoretical identifications in general are implied by the theories that make them possible— not posited independently. This follows from a general hypothesis about the meanings of theoretical terms: that they are definable functionally, by reference to causal roles” (1972, in Block 1980a: 207). Clearly it is the theory of meaning, and not the Ramsification, that is doing the metaphysical work.32 If the original, pre-Ramsification, psychological theory is a reference-fixing theory, then the Ramsified theory merely provides heuristic descriptions of the psychological states, or perhaps operational definitions. Such a theory would make no distinctive metaphysical claims about the nature of the mental states themselves. On
99 Varieties of Functionalism
the other hand, if the original theory is composed of commonsense platitudes or analytical and conceptual truths, then the Ramsified theory might be an analysis of the meanings, in causal terms, of the mentalistic terms of the folk theory. In that case—if the analysis exhausts the meanings of our mentalistic terms—then it could be plausible to regard the Ramsified theory as giving the “nominal essence” of the mentalistic terms. Indeed, this seems to be what Armstrong (1968), Lewis (1972), and Shoemaker (1981b) had in mind. Whether such a nominal essence claim constitutes a form of metaphysical functionalism is less clear; it might be better thought of as a Rylean alternative to metaphysical functionalism. That is, it seems to give an analysis of how we use mentalistic language rather than a theory of mental states, properties, or events. So perhaps it is not a version of metaphysical functionalism at all. Is functionalism compatible with identity theory? Some varieties are; we saw that metaphysical functionalism is not. Is theoretical functionalism—a Ramsified psychology, say—compatible with identity theory? That depends on the psychological theory with which you start and what you do with it. It doesn’t depend one bit on the devices or techniques of theoretical functionalism. The bottom line is that whether the functionalist theory is a version of metaphysical functionalism has nothing to do with its Ramsification and everything to do with its substance and the assumptions with which you begin. This depends, in turn, on the form of the explanations that comprise the target theory, for these theories are thought of as conjunctions of individual explanatory generalizations. Explanatory Functionalism Georges Rey sees the commitment to theories specified by Ramsey–Lewis method as the common core of all varieties of functionalism: “someone who is a functionalist about the mind is someone who looks to ramsification to provide the constitutive analysis of mental states by its ability to specify those states in terms of the causal relations among them” (1997: 175). On my view Rey is not quite right, but he is close. To see why, we must again consider the form of Ramsified theories. Rey seems to think that Ramsified theories are
100 Chapter 3
metaphysical theories, that they specify versions of metaphysical functionalism. Although metaphysical functionalism is the most recognizable variation, it is not the most basic form of functionalism. One of my goals in this chapter is to illustrate that only some functionalist theories are metaphysical, and indeed that Ramsified theories may or may not be versions of metaphysical functionalism. But Rey is certainly right that Ramsification can be an important conceptual tool for functionalists of all sorts. Ramsified theories are constructed by binding together (simply conjoining, in fact) functionalist explanations. This is why theoretical functionalisms themselves have no distinctive metaphysical content. But theoretical functionalism is close to the common core of functionalism after all. More basic are the functional explanations from which functionalist theories are composed. Explanatory functionalism is the view that something or other about a thing can or must be explained by reference to the functions of that thing, or in terms of its functional parts. Every functionalist theory is composed of functionalist explanations. But variations of explanatory functionalism differ about just what is to be explained. As with all versions of functionalism, the details matter a great deal. Versions of explanatory functionalism stand out from other flavors in that explanatory functionalists have done an admirable job of paying attention to the details, perhaps because explanation has been more fashionable than metaphysics for the past thirty-odd years. I am honor-bound to mention that Aristotle advocated some kinds of explanatory functionalism. Aristotle thought that everything had telos, an end toward which its motions are directed (Physics II.8). Telos is therefore referred to both in explanations of how a thing has come to be the way it is, and also in explanation of its current motions. Contemporary explanatory functionalists divide Aristotle’s two explanatory tasks between different notions of function, usually formulated as distinct views that may be held individually or jointly. But the Aristotelian heritage of functional explanation is marked by the general use of the term teleological explanation.33 After some years of ill-repute, explanations of a thing’s presence in terms of its function have found a comfortable home in evolu-
101 Varieties of Functionalism
tionary biology. Evolutionary biologists sometimes make use of adaptationist explanations, by which the presence (origin) or maintenance (persistence) of a trait in a kind of organism is explained by appeal to the function of the trait. The notion of function usually at work in adaptationist explanation these days is what we will call etiological function. This idea is usually traced back to Larry Wright’s paper, “Functions,” but the best-known formulations are due to Ruth Millikan (1989) and Karen Neander (1991). In the next chapter I will be concerned with the details of different notions of function; but for the moment I will gloss the idea by saying that the etiological function of a trait is an effect that it has had in the past, for which it was selected by natural selection, an effect that contributed to the inclusive genetic fitness of creatures bearing the trait. A trait has an etiological function if it has been selected for by natural selection. Adaptationist explanation explains the presence or maintenance of a trait in an organism by telling what that trait has done that contributes to the total fitness of the organism that possesses the trait— that is, by stating its etiological function.34 There has recently been much controversy over exactly how to formulate the etiological notion in detail, and whether evolutionary biology uses or needs to use any notion of function other than the etiological notion. Most participants in the debate agree that evolutionary biologists at least use an etiological notion of function. The question now is whether evolutionary biologists also use another notion of function. There is a growing consensus—even including some who previously held that the etiological notion was the only one that evolutionary biology needed—who answer in the affirmative: evolutionary biologists use an etiological notion and another notion of function, such as that described by Cummins, discussed below (see Amundson and Lauder 1994; Godfrey-Smith 1996). There are two important special cases of adaptationist explanation of mental states and processes: biopsychology and evolutionary psychology. “Biopsychology” is Ruth Millikan’s term for a science she envisions that treats psychological categories as biological categories, traits in the sense employed in biology. According to Millikan (1993), if the categories of psychology are biological categories, then the science of psychology should be the study of the biological functions
102 Chapter 3
of psychological kinds in just the way that evolutionary biology studies the functions of other biological kinds. Biopsychology would be a subfield of evolutionary biology that is interested in psychological traits: The object of biopsychological study is not a chunk of matter, warm or cold, lying on the lab table waiting for its structure to be examined, for its inputoutput dispositions to be tabulated, or waiting to see what causes applied will produce what effects, what “special science” laws may hold for it. Nor is the point of biopsychology to examine or speculate on details of the complex structures inside the black box, to check on the dispositions of the components, nor to examine how the little dispositions inside add up to the complex dispositions of the whole. Biopsychology is not, then, all of what has traditionally been labeled “psychology.” (Ibid.: 177)
Millikan thinks of traditional psychology as attempting to provide causal or nomic explanations. But that is not the goal of biopsychology. Rather, biopsychology works “by subsumption of behaviors under biological norms rather than laws and/or by noting departures from these norms and perhaps causes of these departures” (ibid.: 187). Millikan calls this sort of biopsychological explanation “normalizing explanation.”35 Since psychology as we know it is at least—perhaps solely—interested in the “details of the complex structures inside the black box,” it is clear that biopsychology does not replace psychology. Biopsychology, like evolutionary biology, is concerned with the biological functions of things, with explaining the presence and maintenance of traits, rather than with explaining the current operation of traits. Millikan’s baptism of “biopsychology” hasn’t caught on, but her interest in the teleological function of minds is shared by the field of evolutionary psychology: The main claim of evolutionary psychology is that the human mind is a set of cognitive mechanisms that are adaptations to the environment of the Pleistocene. This claim emerges from two convictions. First, evolutionary psychologists maintain that the mind is largely a set of cognitive mechanisms or “organs” that are devoted to solving problems. The second conviction is a commitment to adaptationism, the view that natural selection is the only important cause of most phenotypic traits. The combined force of these claims leads evolutionary psychologists to maintain that natural selection is
103 Varieties of Functionalism
the only important cause of most mental organs. (Grantham and Nichols 1999: 47)36
Evolutionary psychology attempts to explain current human mental capacities by referring to what would have been adaptive for our hunter-gatherer ancestors. There are two main differences between biopsychology and evolutionary psychology. The first is their scope. Biopsychology does not replace psychology-as-practiced, and Millikan leaves it open that some states with which psychology deals will not be objects of biopsychological study. In contrast, evolutionary psychology, as I understand it, is committed to the claim that most psychological state types (as conceived by scientific psychology) are adaptations. Evolutionary psychology is the descendant of the sociobiology of the late 1970s and 1980s (Wilson 1975). Like sociobiology, evolutionary psychology is quite controversial, and settling the controversy is not part of my interest.37 I will note that it is, at least, Panglossian (Gould and Lewontin 1978) in that it takes most mental states to have evolutionary functions, and that they currently have the effects for which they were selected. That is a strong claim. The second main difference between Millikan’s biopsychology and evolutionary psychology is that evolutionary psychology endeavors to explain our current cognitive capacities by appealing to their evolutionary history. Unlike biopsychology, evolutionary psychology also concerns itself with what Millikan refers to as “the dispositions of the components” and “how the little dispositions inside add up to the complex dispositions of the whole.” Millikan, in contrast, is adamant that the notion of function she employs does not license claims about either the current frequencies or current propensities of traits. This is not an accident. It is quite important to Millikan’s account that adapted traits can fail to have the effects that normally make them adaptive—that they can fail to perform the functions that are explanatory. It is not part of my project to assess either biopsychology or evolutionary psychology.38 But one noteworthy feature of Millikan’s modest biopsychology is that it leaves room for psychology as we know it. Familiar scientific psychology does not appeal to etiological notions of function; most of psychology does not offer explanations
104 Chapter 3
of the biopsychological or evolutionary biological varieties. But it may nevertheless offer functionalist explanations of a sort. The most vocal advocate of the view that psychological explanations are species of functional explanation is Robert Cummins (1975, 1983, 1989). Cummins agrees with Millikan that psychological explanation should not be a variety of subsumption under causal laws,39 but he is nevertheless concerned with the dispositions of psychological states and the complexities of what goes on inside the black box: “Cummins’ concept of function is not used in explanations of why the functionally characterized thing exists, but explanations of how some larger system, which the functionally characterized thing is part of, manages to exhibit some more complex capacity or disposition” (Godfrey-Smith 1996: 15). Cummins offers a causal style of explanation, but a kind that is very localized. He emphasizes that although some explanations aim to account for causal changes or for a trait’s coming to be, other explanations aim to account for properties. According to Cummins, this is what psychology is in search of, property explanations—explanations of how mental properties are realized. He writes, “We know a lot about what causes pain, but there is no very good theory of how pain is instantiated. . . . I think we are close to knowing how life and intelligence are instantiated, though we are a long way from understanding how consciousness or intentionality are instantiated” (1983: 15). Property explanations answer questions such as, “What is it for S to instantiate P ?” or, “In virtue of what does S have P ?” According to Cummins, “The natural strategy for answering such a question is to construct an analysis of S that explains S’s possession of P by appeal to the properties of S’s components and their mode of organization. The process often has as a preliminary stage an analysis of P itself into properties of S or S’s components” (1983: 15). When property analysis is applied to dispositional properties, Cummins calls the strategy functional analysis, and the resulting explanations are a variety of functional explanation. “When a capacity of a containing system is appropriately explained via analysis, the analyzing capacities emerge as functions” (ibid.: 28). Cummins’s causal-role notion of function results from the analytical strategy: A thing’s causal-role function is the effect it has that contributes to the explanation of the
105 Varieties of Functionalism
system in which it is a component. A special case of causal-role functional explanation is the project that John Searle (1980) calls “Weak AI”. We might think of this, more generally, as the informationprocessing view of cognitive psychology. Weak AI, recall, is contrasted with Strong AI (a version of metaphysical functionalism). According to Weak AI, computers are useful tools in the project of understanding and explaining cognitive capacities of organisms and systems, but computers are not thereby taken to have or to realize the cognitive properties or capacities explained (Searle 1980). So there are two basic sorts of functionalist explanation, each of which admits of several variations: explanations of the presence and maintenance of traits, and explanations of the current powers and dispositions of traits. We can give a general (if disjunctive) schema for these different sorts of functionalist explanations. According to explanatory functionalism, the presence, maintenance, or behavior of a thing, property, state, process, event, etc. is explained when we can say what functional role it occupies in some system; likewise, the capacities of a system are explained when it is exhaustively decomposed into functional units: (Fe) An explanation of the presence, maintenance, or behavior of a mental state M in system S refers to (i) the functional roles {R1, . . . , Rn} that M occupies in system S, or (ii) how some properties {P1, . . . , Pn} contribute to M’s instantiating functional states {F1, . . . , Fn} in S. Metaphysical functionalism licenses a functionalist explanation of metaphysics of mind, intentional functionalism licenses a functionalist explanation of intentionality, and so forth. It is their reliance on functional explanation that unites the otherwise diverse varieties of functionalism. When one thinks of functionalism in philosophy of mind, metaphysical functionalism first comes to attention. But not every version of functionalism is metaphysical, whereas all forms of functionalism have in common that they admit to some version of functional explanation. Each explains something by reference to its function, or to the functions of its parts. A functionalist explanation of intentionality need not entail intentional functionalism. And—more
106 Chapter 3
important—functional theories and explanations do not have to entail metaphysical functionalism. Even if psychology gives functional explanations—as Armstrong, Putnam, Lewis, Cummins, Dennett, and many others claim—a question remains: Does functionalist psychological explanation thereby give the essence (real or nominal) of the states to which it refers? It is commonly supposed that it does, that explanatory functionalism entails metaphysical functionalism. Explanatory functionalism is the most central sort. Admittedly, the varieties of functionalist explanations are heterogeneous, and it may be objected that I have gerrymandered explanatory functionalism in order to construe it as the unifier of functionalist theories. Certainly not all functionalists are self-consciously adhering to some variety of explanatory functionalism. It seems that belief in the standard formulation—“mental states are functional states”—has some claim to stating what functionalists have in common. But we have seen that the standard formulation is ambiguous, which makes it a poor candidate to be the common core of any doctrine. I am suggesting that assent to the standard formulation, such as it is, typically indicates a commitment to one or another version of explanatory functionalism. Methodological Functionalism One last strand of functionalism requires our attention: methodological functionalism. This is a bit anticlimactic, but we are nearly through with our initial catalog of functionalisms. In discussing theoretical and explanatory functionalism, I have already mentioned several sorts of methodological functionalism. Most obviously, there is the process of Ramsification, the Ramsey–Lewis method. But the Ramsey–Lewis method begins with a functionally specified theory of psychology, so there must also be some method or methods for producing such a theory. Indeed, two methods were mentioned: one might perform an analysis (in the traditional philosophical sense) of ordinary discourse in order to discover the tenants of folk psychology, or the analytic truths of it (per Lewis 1970 and Shoemaker 1981b); or one might follow the methods of scientific psychology to
107 Varieties of Functionalism
produce an empirical psychological theory (per Fodor 1968 and Cummins 1975, 1983). Both of these options would qualify as what Kim (1998) calls “functionalization,” and both could be thought of as versions of—usurping Cummins terminology—functional analysis. Methodological functionalism, functional analysis broadly construed, includes any process or technique that seeks to understand a system in terms of function—either the functions of its components, the functional role that constitutes its kind, or its function within some other system. (Fa) The method for understanding a mental state M of system S is to determine the functional roles {R1, . . . , Rn} that M occupies in S, or in some supersystem S* (in which S can be determined to have a functional role); or to determine the functional units {P1, . . . , Pn} that compose M. Versions of methodological functionalism make up a ragtag bunch, but they are pervasive. There are a few accounts, such as Cummins’s original (1975) suggestion, wherein a functional method is the basis for a notion of function and a kind of functional explanation. But more often methodological functionalism is derivative on other versions of functionalism. For each notion of function, or functional explanation, and so on, there will be some method for discovering the function, constructing the explanation, and so on. Perhaps that is a sign that we should rein in our notions of function. I will settle for mentioning a few obvious applications of functionalist methodology. I have already noted that the Ramsey–Lewis method should count as a version of methodological functionalism if anything does. Evolutionary biologists and biological anthropologists and anatomists each perform functional analyses of a sort. As discussed above, evolutionary biologists are largely concerned with adaptations and etiological explanation (Wright 1973; Millikan 1989; Neander 1991). Along with other biologists, they are also concerned with the current function (along the lines of Cummins’s causal-role function) of biological systems.40 If Cummins is right, then scientific cognitive psychology follows the methods of functional analysis and interpretive analysis and so
108 Chapter 3
engages in methodological functionalism. Cognitive neuroscience, neuropsychology, systems neurobiology, and perhaps even molecular neurobiology all use “functional” methodologies, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET). These methods identify brain areas according to the cognitive capacities that are correlated with localized neural activity. All of these can be augmented with, but do not require, the methodology of computational modeling and simulation that Searle calls “Weak AI.” Finally, we can say that Dennett’s intentional stance is a functionalist method with respect to mental states: Dennett (1971), for example, says that we should treat certain things as though they were intentional systems in order to explain their behavior. And intentional systems are understood, at least operationally, along the lines of intentional functionalism. An interesting and important question is whether any of these methodologies produces or aims to produce the de dicto explanations that are needed if psychofunctionalism is to entail metaphysical functionalism. 3 Concerning Metaphysical Functionalism Peter Godfrey-Smith reports, “[m]ost recent philosophy of mind has been ‘functionalist’ in some sense or other” (1996: 13). Yet the endurance of the “somewhat amorphous functionalism that philosophers have come to favor” (Van Gulick 1988a: 150) is something of a mystery. If many philosophers believe that one or another version of functionalism will work, most everyone agrees that each of the extant versions of functionalism does not work. Somehow this situation is made out to support the contention that there must be some version of functionalism that will satisfy everyone. Moreover, this ideal functionalist theory, according to the received view, will be a version of metaphysical functionalism. My conjecture is that the apparent broad acceptance of metaphysical functionalism is based on a mistake. It is the result of conflating versions of functionalism with one another or, as I will argue in chapter 6, with altogether distinct theses concerning scientific explanation.
109 Varieties of Functionalism
The geography of functionalism is territory that we needed to explore. I have taken the time to distinguish metaphysical functionalism from other versions so that I can tack down exactly that one with which I am concerned. Most of the others we can now leave behind, though we will later return to some—in particular to functionalist explanation. But in the next few chapters I will focus on metaphysical functionalism. We can for the moment set aside questions of why one might endorse metaphysical functionalism, for my argument is entirely internal to the view. Perhaps unfairly, the diversity of functionalist theories has protected the doctrine. Against the backdrop of the varieties of functionalism discussed above, standard arguments against functionalism are revealed as haphazard attacks on miscellaneous parts of disparate functionalist theses. Now we can see that most of those parts do not belong to metaphysical functionalism. Although they are parts of theses that may or may not entail metaphysical functionalism, few are essential to it. If I am going to show that identity theory is a contender then I will have to deal with metaphysical functionalism directly, not mistaking it for its cousins. If the standard antifunctionalist arguments are indiscriminate, then I’ll need a new sort of argument.
4 Realization and Generic Functionalism
Metaphysical functionalism is the doctrine that to have a mental state is to have a state that realizes or instantiates a functional state, that is, that occupies a functional role. It is a metaphysical thesis because it says that to be a mental state is to realize or instantiate a functional state. If we are to understand the metaphysical claim that is being made by functionalists, then we shall have to understand what it is to realize a functional state. We need to understand the realization relation. You might be surprised to find that in all the considerable literature about functionalism there has been little sustained discussion of the realization relation itself.1 This is not to say that there has been no attention to the matter; many short mentions are peppered throughout the literature. But few of these even take the realization relation itself as their topic. More often realization is simply invoked in the service of an explanation of functionalism. Some philosophers acknowledge that there is more to realization than meets the eye. Jaegwon Kim, for example, remarks in a footnote, “The term ‘realize’ used in [the multiple realizability principle] has not been explained. As we make progress . . . its meaning should become clearer; in the meantime, you will not go far astray if you read ‘P realizes M’ as ‘P is a neural substrate, or correlate, of M’ ” (1996: 102, fn. 4). Kim is right to notice that the realization relation stands in need of explanation, particularly as he has now made it the centerpiece of his philosophy of mind (e.g., 1998). But Kim underestimates the hazards of leaving the matter untended. My
112 Chapter 4
purpose in this chapter is to argue that we will indeed go very far astray if we are not careful about the realization relation. On my account, functional realization is much more tightly intertwined with the functionalist theories of mind that invoke it than is usually appreciated. Developing this observation, in the next chapter I’ll provide accounts of functional realization for available versions of functionalism. I argue that no current version of functionalism supports the functionalists’ metaphysical claims, and that there is little reason to suppose that any could. 1 The Disunity of Realization Here is a quick sketch of the problem. The metaphysical burden of any version of functionalism is borne by two parts of the theory: the realization relation and the notion of function. To be a mental state M is to realize functional state S. According to the prevailing orthodoxy, these parts are independent. The notion of function determines the sorts of relations to other states that constitute each functional state. And realization gives functionalism its distinctive metaphysical relation, the relation between physical states or properties and functional states or properties. It is generally taken for granted that the metaphysics of realization either does not present any special difficulty or that it is basically understood, for example, on the model of the relation between hardware and software, when it is sometimes called “implementation.” Attention is therefore focused on functions and functional ascription, on discovering and formulating the functional roles of mental states. The metaphysical mind–body problem can be considered solved; the mind–brain relation is realization. What remains is just the task of working out all the details. Working out the details is, of course, a difficult task; but it harbors no deep metaphysical puzzle, so the philosophers’ work is done. The rest can be left for the cognitive and neural sciences. The orthodox picture, then, is one according to which theorists can each plug their favorite variety of function into a more or less standard realization relation. The orthodox picture is badly mistaken. Consider the obvious point: To play a functional role is to have a function. If physical state
113 Realization and Generic Functionalism
P realizes functional state M then it has function F. Whatever the notion of function, the realization of a functional state entails having a function. What could it be to realize (occupy, instantiate, implement, etc.) a functional state if not to have a function? Occupying the functional role of a chair, for example, perhaps involves having a function of supporting a seated person or some such. If you do not believe that realizing a functional state involves having a function, then perhaps you are not a functionalist after all. (This suggestion is not flippant; it is one that I will take quite seriously in chapter 6.) Because realizing a functional state is a matter of having a function, it stands to reason that the realization relation varies with the notion of function. The variety of relations that a state must have to be a functional state is dictated by the notion of function at work. We’ll soon explore this claim in detail, but for now a few examples will be helpful. One notion of function, roughly that described by Cummins (1975), says that the function of an item is determined by the causal role it plays in the capacities of a system. Another notion of function, roughly that of Millikan (1989), says that the function of a thing is the effect its ancestors had that contributed to the evolutionary fitness of the systems in which they occurred. Think about what is involved in having a function of each of these sorts. One sort involves currently having certain causal effects; the other involves having a certain history regardless of current causal proclivities. It follows that realizing a causal role function (à la Cummins) may be quite different from realizing an evolutionary function (à la Millikan). This shows, moreover, that the realization relation cannot be understood independently from the notion of function; realization cannot be assumed to be the same relation for every notion of function. In short, there is no single realization relation. Of course we could still say that all cases of realization have something in common—they are all a matter of having a function, of occupying a functional role, or satisfying a functional specification. There’s nothing wrong with thinking of realization in this way for some purposes; but it will be misleading to think in this way if the project is that of understanding metaphysical functionalism.
114 Chapter 4
We are not helped by the illusion of unity when it obscures that the relations doing the work are as many and distinct as the notions of function. For this reason I say that there is no single realization relation.2 Caution is called for. When we talk of “realization” it may be tempting to suppose that there is a new sort of entity that exists when realization obtains, a realization. But the realization relation is now revealed to be in a way more modest. A functional state (in the strong sense) is entirely constituted by its relations to other states, that is, by its functional role. Functional roles specify functional kinds. To occupy a functional role is to be a member of a kind—to be a token or an instance of the kind of functional state constituted by the role. The idea that kinds have tokens or instances is entirely familiar, and it is not distinctly functionalist. When self-help gurus urge us to “actualize” or “realize” our potentials, they do not recommend some specific activity (“actualization” or “realization”) that is distinct from the other activities of life. Nor do they hope that we will become some new kinds of objects (“actualizations” or “realizations”) other than people who have certain specific but ordinary characteristics. So too for functional realization. Realizing a functional state is a matter of having a function, no more and no less. To cast this observation in the terms Kim (1998) uses to critique his own former views about supervenience: Realization, as it has been treated in the past, is not a metaphysically deep relation. The mere claim that functional realization obtains leaves unaddressed the question of what grounds or accounts for it—that is, the question of why the realization relation holds. This is plain to see in Kim’s formulation, quoted above, which appeals to a correlation that is itself left unexplained: “P is a neural substrate, or correlate, of M.”3 Another example is Sydney Shoemaker’s suggestion that we “understand ‘realization’ in such a way that if a property P can be construed as the disjunction of properties Q and R, then Q and R are realizations of P” (1981b, in 1984: 276–277). The trouble is that the question about realization is precisely the question of under what circumstances a property P can be construed as a disjunction of other properties. What is it about Q and R that makes them—but not S,
115 Realization and Generic Functionalism
say—realizations of P? Shoemaker’s formulation gives us no guidance on that question, neither in general nor for the particular case of the mental and the physical. We need an account that tells us which correlations or correspondences are realizations and which are not. The answer, of course, may be that every correlation or correspondence is a case of realization; but that could only be the outcome of some theory of realization. It does not fall out of mere appeal to such correlations. Jerry Fodor has long been aware of the issue: “Suppose . . . that, in the internal organization of this machine, relays will play the role of neurons: e.g., that the closing of a machine relay should correspond to the firing of an organic neuron, and so on. The difficulty is to cash this equivocal ‘e.g.’ for a staunch ‘i.e.’ ” (1968: xviii, italics removed). Until an account of realization is given, to say that a physical state realizes or instantiates a functional state is only to name a relation that stands in need of explanation. Shoemaker’s formulation makes it look as though realization is a matter of being one of some number of disjuncts. But this is misleading; all the work is being done by whatever relationships fix the members of that set. Being a realization is a matter of actually having those relations, whatever they may be. And those relations will vary according to the particular sort of function invoked. How does realization work? We now see that the answer will depend on what sort of functional state is being realized. That there is no single realization relation is not by itself any objection to functionalism. I daresay that it reveals functionalism to be a more sophisticated and subtle doctrine, and it exposes resources not previously recognized. After all, only one version of functionalism needs to be correct; now functionalists have a fresh and unexplored variable to tinker with. Needless to say I aim to dash any new hope that I might temporarily lend functionalism. In the next chapter I will work out the realization relations for a representative sample of functional kinds, and I’ll argue that the resulting theories fall short of functionalists’ goals. But first we need to at least consider some alternative accounts of realization.
116 Chapter 4
2 What Functional Realization Is Not According to my approach, giving an account of functional realization is a matter of explaining how it is that states can occupy functional roles, and I claim that what that involves depends on the kind of function the state is supposed to have. If I am right then there is no one realization relation that operates in every one of the versions of functionalism and their various notions of function. Each realization relation is essentially tied to a specific notion of function. These features distinguish my account of functional realization from every other. Before I say more about my own view, I ought to consider whether it is I who have missed the point. After all, I seem to be claiming not only that all the other approaches to explaining functional realization are wrong, but moreover that they have entirely overlooked a basic insight. There is a maxim among poker players: If you look around the table and you can’t figure out who the dupe is, then it’s you. Could it be that I am the one who is off track? I began this chapter by lamenting the scant discussion of the realization relation itself. But there have been a few discussions, and recently some philosophers have noticed and begun to fill the lacuna. John Heil (1992) discusses realization in the context of supervenience and causation, Jeffrey Poland (1994) in formulating physicalism, and Terence Horgan and John Tienson (1996) in comparing classical and dynamical information-processing systems.4 More recently there has been a flurry of work whose main topic is the realization relation. Jaegwon Kim (1998) recognizes the need for an account of realization and now rests his general account of mind on the realization relation. Unfortunately Kim does not himself explain the realization relation, perhaps assuming the orthodox view. However, Sydney Shoemaker (2000, 2001, forthcoming) offers an account, as do Robert Wilson (2001) and Carl Gillett (2002, unpublished).5 Horgan was early to note that realization itself requires our attention: “The notion of physical realization, which has been widely employed in philosophy of mind for some time, obviously deserves philosophical attention in its own right. To my knowledge, as yet this
117 Realization and Generic Functionalism
project remains to be undertaken in a systematic way” (1993: 573, fn. 20). Horgan and Tienson (1996) come close to addressing the relation of functional realization. They don’t offer an analysis of the relation, but they recognize the need for such an account in light of the extensive use they get out of the realization relation in the context of comparing classical and connectionist computing models in psychology and cognitive science. In this latter respect, their interests are more narrowly focused than my own. Horgan and Tienson are mainly concerned with how algorithms and computational processes can be realized or implemented. They take as their starting point David Marr’s (1982) influential three-level model of informationprocessing systems—an analysis that has been quite influential in the cognitive sciences generally. According to Marr, cognitive and other information-processing systems should be thought of in terms of computational, representational/algorithmic, and physical implementation levels (1982: 23–27). Horgan and Tienson (1996: 23) offer two constraints on realization relations that they identify in Marr’s model: realization is a transitive relation, and realization is a many-one relation. On Marr’s view computations may be implemented or realized by any number of algorithms using different representational schemes, which are in turn realizable by various physical systems or hardware; many physical systems can realize a given algorithm or computation. In short, both computations and algorithms are multiply realizable. And when a physical system implements an algorithm that in turn implements a computation, the physical system thereby implements or realizes the computation. This is the transitivity of realization. Horgan and Tienson depend on realization more explicitly than other theorists, more so even than most who discuss computational models. And their explicit dependence on Marr’s three-level model probably exposes a widely influential way of thinking about realization. So, on the one hand, Horgan and Tienson present an important strategy for understanding the realization of computational functions, which we must keep in mind when we return to that topic specifically; but they also caution that the lessons learned from Marr do not by themselves answer the deep questions about the metaphysics of realization. They leave the realization relation itself
118 Chapter 4
unexplored: “How best to characterize the realization relation is a metaphysical question beyond our concerns in this book” (Horgan and Tienson 1996: 23). But that metaphysical question is exactly what concerns me. In the next chapter I discuss realization relations for computational accounts of function, as well as other notions of function.6 From my perspective, Horgan and Tienson focus too narrowly. Computational or algorithmic functions are only one or two of the sorts to which functionalists might appeal. Other philosophers who discuss realization cast their nets more widely. They implicitly assume or explicitly argue that realization is a basic and general metaphysical relation, and not specific to functionalism. Wilson, for example, wants a relation that covers not only mental states but also cases of realization in “the banking system, the criminal justice system, or the electoral system” (2001: 14). Likewise, Poland’s account potentially includes “social and cultural objects” (1994: 67). Heil (1992) mentions the realization of mental states, but also the realization of a desk by its parts. And Gillett’s (2002) primary example of realization is not that of mental states but rather the hardness of a diamond. This broadness differentiates their approaches from my own. I suggest that we think of these authors as addressing a general and metaphysically basic relation called “realization,” which may not be directly related to functional realization or to functionalism; in contrast, I am concerned with functional realization, specifically. I won’t take a stand on whether there is such a general relation of realization that covers the realization of liquidity, mental states, and social states. But if there is a general metaphysical relation of realization, then we can say a few things about its relation to functionalism. For one, functional realization presumably should be compatible with general realization. In particular, if realization is a basic constructive relation in naturalistic metaphysics and functionalism is to be a naturalistic theory, then functional realization ought to be a special case of the more general relation. (This is so only on the assumption that the two sorts of “realization” are supposed to be related at all. If they are related only by an accidental and confusing homonym, then obviously functional realization might be quite unaffected by whatever becomes of the apparently more general
119 Realization and Generic Functionalism
relation.) One might try to argue that some or another functional realization relation fails to satisfy the requirements of the general relation and thus fails to be naturalistically acceptable. This would be a bad result for functionalism. Worse yet, one might try to argue that no notion of functional realization is capable of meeting the requirements for realization. These are interesting strategies, but I will not pursue either of them in this book. Instead I will be concerned with whether any version functional realization can satisfy the requirements of functionalism, leaving for another day the question of whether functionalism can satisfy the requirements of naturalism or of any other metaphysical framework. There is no reason to suppose that the more general notion answers all our questions about functional realization. In particular, we cannot assume that functional realization is realization unembellished. If there is a general and basic notion of realization, we will still expect some explanation of how realization works in particular cases. Poland is keenly aware of this limitation of the general formulation: [T]here is no reason to suppose that there is just one sort of realizing relation that all entities bear to the physical entities that realize them (for example, composition of causal powers). The world is a rich, hierarchically organized, and multifaceted place containing a wide range of different objects and attributes, varying in complexity and constituted in diverse ways. The realization of this world by a purely physical basis may well be the result of many different sorts of modes of constitution. . . . Thus the sorts of realizing relations consistent with a physicalist outlook ought to be the subject of ongoing inquiry. (1994: 19)
Poland introduces the notion of a realization theory. A realization theory gives an account of how some kinds of properties or objects can realize some other kind of property or object, and explains how this constitution comes about in specific cases. In so doing, a realization theory specifies “something like the essence of an attribute” (ibid.: 211). According to Poland, functionalism in philosophy of mind is an example of a realization theory in the more specific sense; it offers a realization theory for mental states (ibid.: 210–211). But Poland, like Horgan and Tienson above, considers only one version of functionalism; he writes, “[a]ccording to the functionalist, the
120 Chapter 4
essence of a mental state, M, is a certain pattern of causal relations” (ibid.: 211). This is a version of functionalism that appeals to causal functions rather than computational functions. I will consider causalrole functionalism in detail in the next chapter; for the moment I simply want to note some important features of Poland’s view. First, Poland agrees that there need not be just one realization relation. His treatment suggests the view, endorsed earlier, that there is a general way of thinking about realization that at most clears the way for specific theories of various kinds of realization—functional realization among them. He offers a general schema for realization relations, “To say that an attribute, N, is realized by a class, R, of physical attributes is to say that N is instantiated in virtue of the instantiation of the members of R” (1994: 16). Since “in virtue of ” is not itself a basic metaphysical relation, we must provide an explanation of how it is that one attribute is realized in virtue of another; and we cannot presuppose that a single general relation will be available. Poland demands realization theories for each kind of attribute or object, thus recognizing the diversity of realization relations.7 Given such commitments, we can reckon that if Poland would acknowledge the different versions of functionalism that I distinguish then he would agree that they could have different realization relations. And he would agree that those different realization relations should be “the subject of ongoing inquiry.” So Poland shares the commitment to seek a metaphysically deep explanation of the mind–body relation, namely, a realization theory of mental states. One way to think about my endeavors later in this chapter and in the next is that I am describing functionalists’ constraints on realization theories of mental states, and then exploring whether any specific realization theories for different kinds of functions can meet those criteria. In this way my project complements Poland’s. Finally, Poland himself does not offer specific realization theories for mental states as functional states, at least not beyond the generic version (“the essence of a mental state, M, is a certain pattern of causal relations”). It is perhaps better to think of him as enumerating general constraints on realization theories rather than formulating specific candidates. These sorts of constraints are what we find
121 Realization and Generic Functionalism
also in Heil, Wilson, Shoemaker, and Gillett. Heil, for example, offers the following schema: “x’s being a at t is realised by x’s being b, just in case x’s being a is constituted, at t, by x’s being b” (1992: 137). In addition to the example of a desk and an arrangement of its parts, Heil suggests that realization is the relationship between liquidity of a bowl of soup and a structure of molecules, and he offers that the intention to wave at Wayne is realized in Clara by a sequence of neuron firings. There is no problem with Heil’s schema as such; but we should not mistake it for a realization theory because it cannot do that work. (Similarly for Poland’s “in virtue of” formulation, above.) If we understand “constitution” loosely then the proposal is plausible but does not have the explanatory depth we’re looking for. Yet if we let the examples guide or understanding of constitution then it tends to be difficult to see how the relation between minds and brains could be the same as the one between desks and their parts. It’s better to regard Heil’s account of realization in terms of constitution as general and schematic like Poland’s “in virtue of” formulation. The same can be said for Gillett’s account, as well as the version he finds in Kim and Shoemaker. Kim (1998) argues that supervenience is a relation that holds within levels of organization rather than between levels. On Kim’s view, the liquidity of the soup or the hardness of the diamond supervene not on the molecules that compose them, but on a macro-object that the molecules mereologically compose. Supervenience is an intralevel relation, not an interlevel relation as philosophers widely assume. Kim also argues that supervenience is a general logical relation, so cases of supervenience demand explanation; in particular, we need to account for the metaphysical relation or relations that explain why supervenience holds. According to Kim (1998), realization is the metaphysical relation that explains supervenience. It follows that realization is also an intralevel relation on Kim’s view. Although Kim does not explicitly formulate a realization relation, Gillett (2002) makes a compelling case for interpreting Kim as holding something like Shoemaker’s (2000, 2001) account of realization. Acknowledging that “realization” has a broader use, Shoemaker sets out to discuss functional realization specifically. The account he offers is nevertheless of the broad sort.
122 Chapter 4
Shoemaker proposes: “property X realizes property Y just in case the conditional powers bestowed by Y are a subset of the conditional powers bestowed by X” (2001: 78). Gillett (unpublished) dubs this the “subset model.” Shoemaker’s account of realization has some features of interest. He explicitly disengages the tasks of specifying functions and giving an account of the realization relation (2001: 84).8 So Shoemaker’s theory looks like a version of what I earlier referred to as the orthodox view of realization, according to which different notions of function can be plugged into a general schema. This is not quite right, however. In fact, Shoemaker has a specific kind of function in mind: His account applies only to realization of causal-role functions (ibid.). But Shoemaker does not think of causal-role functions as one sort among many; rather, he takes this sort of function to be quite ubiquitous (1981b, 2001). Perhaps it is preferable to think of Shoemaker, like Poland and Heil, as providing a general schema for specific realization relations. Gillett (2002, unpublished) is critical of Shoemaker’s and Kim’s views of realization on the grounds that they neglect some common cases of realization. On the “subset model,” the realizing and realized properties occur within the same objects. Thus the liquidity of the soup is realized by a property of the molecular structure of the soup, but not by properties of the molecules considered individually. Realization and supervenience are intralevel relations. Gillett criticizes this treatment of the realization relation, which he sometimes calls the “flat” view, because it fails to cover cases of interlevel realization. According to Gillett, physical sciences are committed to interlevel realization relations. In the case of the hardness of a diamond: “The sciences have illuminated how the causal powers of the properties/relations of the carbon atoms nonetheless result in the causal powers of hardness in the diamond, but not vice versa. Thus the properties/relations of the carbon atoms apparently ‘play the causal role’ of H, but not vice versa, and, consequently, it is plausible that H is realized by the relations/properties of the carbon atoms” (Gillett 2002: 319). Gillett maintains that it is the properties of the atoms and their relations, not the properties of some mereological combination of the atoms, that realizes the properties of the
123 Realization and Generic Functionalism
diamond. If so, then the realizing and realized properties belong to different individuals, and Kim and Shoemaker must be wrong. I am not going to adjudicate between Gillett, Kim, and Shoemaker concerning the properties of diamonds. But if Gillett’s account applies directly to functional realization then it would indeed be in tension with my analysis. For I claim that for x to realize a functional state is for x to have a function, which implies that the same individual or object x has the function and thereby realizes the functional state. So my account of realization relations, like that of Kim and Shoemaker, looks “flat” schematically. On the other hand, I have not set out to explain the hardness of a diamond. It may be that hardness is a functional property, but I will not assume that it is. It may simply be the case that functional realization in my sense is not a species of Gillett’s realization relation, which he takes to be a fundamental metaphysical relation (2002: 322, fn. 8). There is a feature of Gillett’s critique of the “subset” or “flat” model that suggests this interpretation. Gillett submits that Kim and Shoemaker might have been led astray by taking too literally the idea that realizers play functional roles: “One explanation is that proponents of the Flat account have taken a quite literal view of a realizer X ‘playing the casual role’ of a realized instance Y, thus assuming X must contribute the very same powers as Y ” (ibid.: 321). For my part, I certainly mean to take seriously the idea that realizers of functional states occupy or play functional roles. This strikes me as a truism about functional states. Gillett’s approach, then, looks to be entirely at odds with my own if we are talking about the same “realization” relation. But he does not mean to deny the aptness of talk of roles in the case of functional states. So a better conclusion is that we are not examining the same relation.9 If there is a mistake in my approach it is not that I take the idea of functional roles too literally, for playing of functional roles is explicitly part of the functionalist theories that I mean to understand. If that is a problem according to Gillett’s account of realization generally, then it is a problem for those functionalist theories but not for my attempt to understand them. That being said, Gillett’s own preferred account of realization does not exclude the relation appealed to by Shoemaker and Kim, nor
124 Chapter 4
even the varieties that I will propose in the coming chapter. Gillett’s “dimensioned” view of realization holds: “Property/relation instance(s) F1–Fn realize an instance of a property G, in an individual s, if and only if s has powers that are individuative of an instance of G in virtue of the powers contributed by F1–Fn to s or s’s constituent(s), but not vice versa” (2002: 322).10 While allowing that realizing and realized properties may not occur in the same individual, Gillett’s account does not rule out cases in which they do. Like Poland’s schema, Gillett’s formulation leaves us wondering how it is that a property can have its powers, that is, its causal powers, “in virtue of” some other property or set of properties. Perhaps “in virtue of” is all that we can hope for if we seek a variety of realization that is quite general and metaphysically basic. I have not offered direct critiques of the accounts of realization offered by Heil, Wilson, Kim, Shoemaker, and Gillett. I’ve argued that their accounts do not explain the specific relation of functional realization; they discuss a more general and perhaps more basic relation. This is only a problem to the extent that they assume that the general account explains all that we need to know about functional realization. I have not had anything to say about which if any of their accounts of a primitive realization relation should be preferred. Each in its own terms could be acceptable, and all appear to be compatible with my approach to functional realization.11 All that I have insisted on is that we should not assume that general accounts of realization provide the whole story about realization relations. Heil, Wilson, Shoemaker, and Gillett should acknowledge the need for realization theories, as Poland does, and they should recognize the need for explanations of functional realization in particular. If they allow this much, then my proposal is no hazard to their projects. Likewise, it is not a fault of my account that it does not answer the general questions. I am trying to understand the specific relations that do the work for metaphysical functionalism; later we can wonder whether there is a more general relation that usefully covers the liquidity of soup and the hardness of diamonds, as well as one or more kinds of functional states. I can imagine an argument against functionalism that begins by establishing a general requirement on naturalism or physicalism of
125 Realization and Generic Functionalism
the sort that Poland investigates. Perhaps there is a general notion of realization such that all states, objects, and properties are either fundamental constituents of the natural universe or else recursively realized by fundamental natural states, objects, and properties. One could then try to argue that functional states and properties violate that principle. Perhaps the reason would be that so-called functional realization fails to be a case of realization properly understood. I will leave that argument for someone else to pursue. My goals are somewhat more modest and yet perhaps more remarkable. I will indeed argue that functionalism fails to satisfy some important criteria, but those criteria are not general metaphysical principles. Instead, I will focus my attention on whether realization relations can do the work that is demanded of them by metaphysical functionalism itself. I’ll argue that functional realization cannot live up to the promises of functionalism. The next task, then, is to see just what demands functionalism makes on functional realization. 3 Generic Functionalism According to my account, realizing a functional state is having a function. There are many notions of function, and having a function involves different things for different notions of function. So realizing a functional state involves different things for different notions of function. We need to examine each notion of function and its corresponding realization relation to see if they do the job for metaphysical functionalists. Functionalists are asking the realization relation to do some clever tricks, and we want to know whether any particular notions of function and realization are up to the task. To answer that question we need to have a better idea of what the metaphysical functionalist expects out of functional realization. There are nontrivial constraints both on realization relations and on the functional states that are to be realized. If I am right these are part of the metaphysical functionalist’s picture, not extra demands imposed by critics. One might worry about whether a theory that has these features can stand up to other sorts of challenges that are external to the theory itself. At the moment we are merely concerned with the internal constraints that are typical of
126 Chapter 4
metaphysical functionalism. Four desiderata come to mind: First, functionalism is intended to be a constructive theory of mind, rather than deflationary, eliminativist, or epiphenomenalist. So functionalism should give us mental states that are causally efficacious. Second, if mental states are to have causal powers, then they must be “out there” in the world, at least to the extent of brains, bricks, and snakes. This seems to require that the realization relation be, in some sense, an objective relation; in particular, it should not be a matter of interpretation whether something is a mental state or not, or whether it has causal powers or not. Similarly, if functionalist mental states are to have causal powers in virtue of the powers of their realizers, then it seems likely that the realizers will have to be present to contribute those powers. If the sensation of heat is going to be causally efficacious in my now withdrawing my hand from the stove burner then the realizer that contributes causal powers to my sensation must also be present now. This suggests that the realization relation should at least be relatively synchronic. The second and third desiderata are corollaries of the requirement for causal efficacy. The fourth characteristic is required if functionalism is to contend with identity theory in the usual manner. Namely, functional mental state kinds must be biologically abstract; this is needed so that they are be multiply realizable to the degree usually advocated by functionalists. Together these four characteristics describe what I will call Generic Functionalism. Generic Functionalism ought to be familiar. It has the textbook features that are invoked by functionalists against identity theorists, eliminativists, and if there is a difference (as I think there is), “reductionists.” Needless to say, some self-avowed metaphysical functionalists will not agree to one or more of the features of Generic Functionalism, or to their conjunction. But functionalists ought to endorse Generic Functionalism. Generic Functionalism is a good idea; I will argue that it does not work, but it’s a good idea nonetheless. If you’re going to be a functionalist, your theory should fit the Generic Functionalist schema. The motivation behind my formulation of Generic Functionalism is to make explicit the features that functionalism must have for standard arguments in its favor to work as advertised. Since I am going to use the four features of Generic
127 Realization and Generic Functionalism
Functionalism as a metric for metaphysical functionalism it is reasonable to expect that I explain each criteria and provide some reasons for accepting it. Moreover, these should be explanations and reasons that are generally acceptable to metaphysical functionalists. We need to understand what these features contribute to a functionalist theory so that we can see what is sacrificed by giving up any of them. Then we will be able to see why these four features are so important to the defense of functionalism. Causal Efficacy Perhaps most important, instances of functional states must be causally efficacious. They cannot be epiphenomenal in what Owen Flanagan and I have called the “strict metaphysical” or “causal-role” senses (Polger and Flanagan 2002).12 That is, they must have some causal powers, and those causal powers must be effective in the states’ contributions to the causal powers of the systems of which they are parts. In particular, mental states must have causal powers that are effective in controlling the movements of creatures like human beings. My visual sensations, for example, ought to contribute to the control of my behavior, such as walking about without running into furniture and so forth. It remains for psychologists and neuroscientists to discover just how much of our behavior is typically caused by conscious mental states. Some recent evidence suggests that the amount is much less than we might have expected.13 And some philosophers believe that the world does not have room for the distinctive causal powers of conscious mental states and therefore either accept that such states have no causal powers (e.g., Chalmers 1996a) or else conclude that these sorts of mental states simply do not exist (e.g., Churchland 1983). But identity theorists and functionalists alike typically eschew both epiphenomenalism and eliminativism about the mental. So there is nothing unusual in the basic claim that the causal efficacy of mental states is part of Generic Functionalism. There is an extremely minimal version of this requirement. It demands only that instances of functional states have some effects in the systems of which they are functional parts; but it does not require
128 Chapter 4
that the property of being a functional state itself has distinctive causal properties. This weak form of the requirement would be met by functional states even if the functional properties themselves were not causally efficacious. Frank Jackson and Philip Pettit (1990) defend the weak requirement when they argue that mental states (and macroscopic entities in general) may be “causally relevant” while not in fact being causally efficacious. And something like the minimal requirement seems to be what Fred Dretske (1988) establishes for representational states. A stronger requirement is that instances of functional kinds must be causally effective in virtue of their being of the functional kinds that they are. This is more like what Horgan (1989) calls “quausation.” He argues that common claims about the causal efficacy of the mental should be understood as claiming that mental states have quausal powers: “we believe not merely that mental events and states are causes, but that they have the effects they do because they instantiate the specific mental properties they do” (1989: 51). Where Jackson and Pettit replace the causal claim with an explanatory claim, Horgan augments the causal claim with an explanatory claim. For mental states to have quausal efficacy it is not enough that they cause actions; in addition, that they are mental states must be explanatorily relevant to their causing the actions.14 That is, my mental states qua mental states are the causes of my behavior. A view that grants less than quausal powers to mental states, writes Horgan, “would constitute a version of epiphenomenalism hardly less offensive to common sense than is the version which denies that mental events have any effects at all” (ibid.: 47). No doubt there are subtleties not captured by my coarse distinction between weaker and stronger claims of efficacy.15 Though the distinction lacks precision, we can get a general sense for what is at stake. Functionalists typically intend for mental states to be causally efficacious in the stronger sense. Despite some dissonance, it seems fair to say that functionalists usually think of mental states as having distinctive causal powers that are genuinely effective in causing behaviors in the same way that common macro-objects and properties are causally efficacious. Why did the window break? Because I hit it with a brick. The brick qua brick caused the breaking of the window
129 Realization and Generic Functionalism
qua window. This object’s being a brick, and that object’s being a window, are causally effective to their interaction, to the breaking. And all things considered, the functionalist would like to show that functional states meet the strong requirement for causal efficacy. One wants it to be the case that mental states cause behaviors, for example, in the way that bricks cause windows to break.16 If mental states do not have their causal powers because of the mental kinds that they are, then being a mental state is prima facie epiphenomenal. The second and third features of Generic Functionalism are corollaries of the requirement of causal efficacy. Objectivity and synchrony are each designed to secure the strong causal efficacy of functional kinds. For example, it has proven extremely difficult to see how a nonsynchronous realization relation could have any hope of yielding functional kinds that satisfy the strong causal efficacy requirement. So one of the major obstacles that any argument for weakening the synchrony requirement faces is to convince us that a nonsynchronic realization relation could meet the strong causal efficacy requirement, or that weak causal efficacy is sufficient. Dretske (1988) and Jackson and Pettit (1990), for example, are satisfied with the weaker causal claim. Of course if there is no causal efficacy in the stronger sense, or no such causation except among microphysical entities, then there would be no problem in accepting that limit on the causal powers of mental states as well. But if there is any strong macrocausation at all then we would like mental states to have it. The identity theory promises to deliver strong causal powers; it identifies the causal powers of minds with those of brains. So it is hard to see how the functionalist could settle for less. Objectivity The notions of function and realization must be objective. The purpose of this constraint is to ensure that the realization relation and the resulting states, properties, or events have (relatively) mindindependent existence in the world. This seems to be a prerequisite for having causal powers in the strong sense, and perhaps for having causal powers at all.
130 Chapter 4
The notion of an objective relation invoked here is contrasted with those of conventional or arbitrary relations.17 Suppose that I have here a piece of paper. I take it that this piece of paper has certain causal powers, and I take it that it has those powers in virtue of being the sort of macroscopic object that it is, on account of its size and shape, and so forth. Suppose further that this particular piece of paper happens to be a five dollar bill. Although its being a five dollar bill endows it with certain economic powers, that property does not contribute to or detract from its causal powers. This bit of paper would have exactly the same causal powers if it were not a five dollar bill but only an extremely good forgery, or if I were handling it after it had been removed from circulation. In those cases it would have quite different economic powers, but its causal powers would be just the same. Modifying just the conventional properties of the dollar bill will not change its causal powers. To change the causal powers of the bill I would have to do something to the piece of paper itself. I can change its shape (by folding it, or even by cutting or tearing it in certain ways), or its texture (by crumpling it), or its propensity to be recognized by vending machines (by writing on it). But I cannot make changes in the causal powers of the bill merely by external conventions, such as passing a law governing what will or will not count as currency in the United States. The causal powers of a five dollar bill are relatively objective, whereas its economic powers are relatively nonobjective because they are highly conventional. Functionalists usually claim that functional properties are relatively objective; they do not claim that whether something has a mind is a matter of convention or interpretation. The motive for including objectivity in Generic Functionalism is to guarantee that the specification of functional kinds and the identification of realization relations is neither arbitrary nor relative to an arbitrary description of the circumstances. Functional ascription and realization may nevertheless be relativized, for example to an explanatory framework, so long as that framework is not itself arbitrarily selected. The main point is that there should not be ad hoc or post hoc specification of functions and instances, or of the systems to which they are relativized.
131 Realization and Generic Functionalism
I do not pretend that this is a precise specification of the objectivity requirement. Indeed, it invites the same objections that face any other I-know-it-when-I-see-it criteria. And working out the details of objectivity may be tricky. Despite this, it is useful to identify— however roughly—the outline of the desiderata. Functionalists have not typically defended the view that whether something realizes a mental state (whether it has a mind) is a matter of interpretation.18 Although the requirement for objectivity was not always acknowledged and there are some who yet resist, the troubles for nonobjective notions of function are great. The truth that realization is relative to function, and so to functional ascription, can easily lead to the idea that whether something realizes a mental state is just a matter of whether it can be described as such. I suspect that much of the science fiction that has served as argument in philosophy of mind could have been avoided if a requirement for objectivity of function had been recognized. I’ll have more to say about the source of this confusion in the next chapter. Once we understand how realization works, we’ll see that we need not concern ourselves with the mental states of thermostats (Chalmers 1996a), aliens who fly around in microscopic spaceships mimicking atoms (Block 1978), the nation of China attempting to pass the Turing test (Block 1978), rocks (Putnam 1988; Chalmers 1996b), walls (Searle 1992), or buckets of water (Hinkfuss, described in Lycan 1987) whose molecular motion temporarily mimics a finite state machine. There is of course nothing wrong with thermostats, fleets of alien spaceships, the nation of China, walls, or buckets. These systems may even realize some functional properties; but that does not show that they realize mental states. The objectivity of the realization relations that might make it seem that these systems have minds is suspect. Synchrony A third requirement for metaphysical functionalists is that the realization relation should be synchronic. That is, it must be the sort of relation that obtains of a state while and insofar as it has certain properties, dispositions, or whatever. “[A]ny object that has the
132 Chapter 4
higher-level property, or is an instance of the higher-level type, does so in virtue of simultaneously having one of the lower-level properties or types that realizes it” (Tye 1995: 41, italics in the original). The reasoning behind the synchrony requirement is plain enough: If realizers contribute to the current causal powers of realized states they must do so in virtue of their current causal powers. It stands to reason that the current causal powers of my mental states can only be the current causal powers of their realizers. Realizing a functional state should be something that a system can do, as it were, right now and in virtue of its current states, properties, and dispositions. Realization may, of course, begin and cease. And it needn’t be instantaneous; it may stretch across time. But the realization of a mental state by a physical state ought to be in virtue of the properties of the physical state through the duration of realization. Something is an instance of water in virtue of synchronic properties, in virtue of having the structure H2O. Being a Bordeaux wine, on the other hand, is not synchronically realizable; it requires having a certain history or appellation. The molecular structure of H2O presumably accounts for the causal powers of water. So too for other liquids and their molecular structures. If being a Bordeaux wine contributes causal powers to a liquid, it is only because causal factors in the past have arranged the molecules of the wine such that they currently have certain powers, for example, to stimulate gustatory and olfactory receptors in certain ways. Of course we might cite the wine’s origin in an explanation of how it came to have the molecular structure that it does. But it is the synchronic molecular structure, not the history, that grounds the causal powers of the wine. It seems that the causal history of the wine can affect its current causal powers only by structuring its synchronic realizers. In Jackson and Pettit’s (1990) terminology, being a Bordeaux wine may be causally relevant, but it is not causally efficacious. The synchrony requirement is more controversial than the requirement for objectivity, for as we shall see it straightforwardly rules out some kinds of function from completing metaphysical functionalism. But the synchrony requirement is clearly part of Generic Functionalism, for several reasons. The first, mentioned above, is that synchronous realization seems to be necessary for mental states to
133 Realization and Generic Functionalism
have causal powers in the strong sense, or perhaps at all. The second is that functionalism is attractive in part because it promises not only that we might discover things quite different from us that have minds, but moreover that we might someday build such things. As Jerry Fodor writes, “Talk about the function of Xs is, in short, related in a variety of ways to talk about artificial Xs, substitute Xs, impaired activity of Xs, pathological Xs, normal Xs, and so on and on” (1968: xx). The very possibility of artificial minds presumes something like the synchrony constraint; such is the nature of construction. A third consideration is that our states of consciousness certainly seem to be states that we now have. Indeed the possibility of introspection seems to require that my mental states are current states of me. This sort of observation carries little weight among some philosophers, so I mention it only after the more substantive reasons. Yet I submit that there is at least a prima facie presumption in favor of synchrony. This observation or intuition may prove incorrect. The synchrony requirement might be relaxed or entirely ignored. But there would have to be a mighty good argument to convince us to give up the idea that something is a conscious mental state in virtue of its contemporaneous qualities. Both objectivity and synchrony are desiderata for functionalism because they appear to be required if mental states are to be causally efficacious, or so it has seemed to many. In this sense, objectivity and synchrony may be viewed as secondary to the requirement for causal efficacy. The fourth feature of Generic Functionalism is not required to support causal efficacy, but it is no less central to the functionalist program. Biological Abstractness The functionalist wants more than just causal efficacy, objectivity, and synchrony. Metaphysical functionalism typically claims that the functional kinds to be identified with mental kinds are more general or abstract than biological kinds. Metaphysical functionalism, remember, is supposed to be a more general alternative to identity theory; so it should not be that they universally pick out the same kinds. (They might sometimes happen to pick out the same kinds,
134 Chapter 4
of course.) The functionalist is committed to multiple realizability, to the claim that mental kinds can have multiple realizations. In fact Generic Functionalism demands a bit more. It is custombuilt to allow at least standard MR, if not also radical MR. If it should turn out, as some argue, that biological kinds are themselves functional kinds in some sense, that would be an interesting result, but it would not secure the thesis of metaphysical functionalism as it appears in philosophy of mind. Functionalism requires that the realizers be potentially variable not only from the point of view of physics or chemistry, but moreover from the point of view of biology. As Robert Van Gulick notes, “If physiological properties and other physical properties of many sorts can all be interpreted as functional properties, then the functional nature of psychological states cannot be taken as evidence against the identity thesis. Moreover, the very thesis of functionalism itself is in danger of losing interest, for its appeal lay in picking out some supposedly distinctive characteristic of psychological properties” (1982: 190). The revolutionary claim of functionalism is that mental states are functional states in relation to biological states, that is, that mental states are biologically multiply realizable. Generic Functionalism therefore must include the view that mental kinds are not identical to biological kinds—even if biological kinds turn out to be functional kinds of a sort. Putting the requirement in terms of biological abstractness emphasizes that the multiple realizers may be biological or nonbiological. In chapter 1, I argued that we should not expect mental states to be multiply realizable in the way that functionalists typically claim. Any such reservations are entirely irrelevant to the present considerations. Regardless of whether or to what extent the multiple realizability of the mental is a good idea, it is a basic tenet of functionalism. Indeed, commitment to multiple realizability is what leads many philosophers to functionalism. If functionalism could not deliver on the multiple realizability of the mental vis-à-vis the biological, its allure would be significantly diminished. So there can be no doubt that Generic Functionalism includes the idea that mental states can be realized by various biological states, and perhaps by some nonbiological states. Mental states, according to functionalism, are biologically abstract.
135 Realization and Generic Functionalism
4 Generic Functionalism versus Identity Theories The picture of metaphysical functionalism that I am painting is one according to which mental states are causally potent, are objectively and synchronically realized, and are of kinds more abstract than biological kinds. These are the features of what I have been calling Generic Functionalism. Generic Functionalism is the sort that textbooks are peddling when they speak without qualification about “functionalism” in the philosophy of mind. Generic Functionalism is not really a version of functionalism, of course. It is only a sketch of a what a good theory of mind would look like—one that is good according to functionalists. My goal has merely been to enumerate some familiar claims that functionalists advertise and depend on; I am not making demands that functionalists will not recognize. So far I have made the case in terms of desiderata of common versions of functionalism. I claim that the functionalist wants a theory that satisfies the outline provided by Generic Functionalism. This might make it seem that some functionalists may simply decline to accept these characteristics, saying, “maybe others would endorse Polger’s Generic Functionalism caricature, but my functionalist theory need not have those features.” That is a fair reply. I’m happy to admit that some functionalists challenge one or more of the characteristics I’ve attributed to Generic Functionalism. Certainly I am in no position to dictate what functionalists believe. Indeed I’ve hinted at some reasons for resisting one or another of the aspects of Generic Functionalism, and we’ll encounter some alternatives in the next chapter. But it’s important to notice that departing from the framework of Generic Functionalism is not without costs. Someone who takes this route will not be able to appeal to the standard arguments for functionalism and thus will owe us some novel argument. The standard functionalist argument is that some version of Generic Functionalism is more plausible than any identity theory. Specifically, as we saw in chapter 1, the main argument claims that functionalism is more general than any identity theory because functional states are multiply realizable. A theory that can satisfy all four features of Generic Functionalism would be attractive indeed. After
136 Chapter 4
all, the identity theorist wants mental states to be causally efficacious and wants being a mental kind to be an objective and synchronous property. The identity theory secures these features by identifying mental state kinds with biological kinds, that is, brain or nervous system state kinds. In so doing identity theorists deny that mental state kinds are biologically abstract. And as a result we must concede that the identity theory is not as general as functionalism. Earlier I argued that psychology is not as general as functionalists typically suppose, and therefore that the generality of functionalism is not in itself a good feature. But now suppose that I am wrong, and that we accept standard MR or radical MR as desiderata on a theory of mind. In that case, if a version of functionalism can fulfill the four promises of Generic Functionalism, it will have all the central merits of identity theory while being more general. If that generality is apt, such a theory should be heralded. Indeed, this is the crown that functionalists have claimed. If the functionalists’ mantle is well earned it must be because psychology demands a degree of multiple realizability that is incompatible with the identity theory and some version of functionalism can deliver that degree of multiple realizability while otherwise matching the claims of identity theory—that is, while having the features of Generic Functionalism. Earlier I impugned the former claim. I tried to carve out some space for an identity theory by arguing that the degree of generality associated with functionalist theories is not desirable. Now we see that even if those efforts were to fall short, the functionalist would still have to show that some concrete version of metaphysical functionalism can satisfy Generic Functionalism. Remember that identity theory has the first three features: It gives us mental states that have causal powers, at least to the extent that brains have causal powers; the identity relation is as objective a relation as we could hope for; and identity is synchronic. The only feature of Generic Functionalism that identity theory lacks is biological abstractness. This sets up the familiar multiple realizability arguments, according to which identity theories are less general than functionalism. At the moment we have set aside misgivings about the multiple realizability requirement. If functionalism has an advantage over identity theory, it is surely its generality. But the generality argu-
137 Realization and Generic Functionalism
ment goes through only if functionalism is otherwise comparable to identity theory. That is why Generic Functionalism is important; it includes the features that functionalism must have to be otherwise comparable to identity theory. Unless some version of functionalism can satisfy all of the features of Generic Functionalism, standard functionalist arguments fall short. Yet Generic Functionalism is only a sketch of a theory, or a framework for one. Generic Functionalism describes the work that metaphysical functionalism must do. It outlines a theory that, if brought to fruition, would be a serious contender. What remains to be seen is whether any specific version of metaphysical functionalism can live up to the promises of Generic Functionalism. This inquiry at last brings us back to the varieties of functionalism. Recall the three questions that structure my taxonomy of functionalism: (1) What kind of functionalism is it? (2) To which mental phenomena does it apply? (3) What notion or notions of function does it use? I have said that I will largely ignore the second question; I am interested in functionalism about conscious mental states. And the first taxonomic principle was discussed in the previous chapter. We are focusing on just one project, that of metaphysical functionalism. The third question remains to be addressed, and we now see that it is absolutely crucial. Choosing notions of function and functional specifications is not just a matter of tinkering with the details of the theory. Realizing a functional state is a matter of having a function. Since realization relations are themselves bound up with notions of function, the fundamental nature of the metaphysical theory rides on this tinkering. There must be an appropriate realization relation for each notion of function. The question is: Can any notions of function and functional realization do all the work that Generic Functionalism requires of them?
5 Functional Realizations
In the last chapter I argued that realizing a functional state is a matter of having a function, and I identified four features that standard versions of functionalism claim to exemplify. The four features—causal efficacy, objective and synchronous realization, and biological abstractness—together characterize what I called Generic Functionalism. I argued that these features are central to standard ways of thinking about and arguing for functionalism. The question, then, is whether any particular notion of function and functional realization can animate a version of functionalism that lives up to Generic Functionalism. In this chapter I consider the different notions of function to which functionalists have sometimes appealed. My strategy is simple. In each case I determine the realization relation appropriate to each notion of function and consider how it fares on the criteria of Generic Functionalism. The results are not encouraging. In each case, functionalism comes up wanting. I do not thereby conclude that metaphysical functionalism is entirely untenable. That would be far too hasty. If I am right, functionalists have some serious work ahead of them. Indeed, my investigation in this chapter and the last may be viewed as a guide to what a metaphysical functionalism must do to fulfill its promises. But, as we shall see, the prospects for such a theory are not bright.
140 Chapter 5
1 In the Beginning, There Were Machine Functions I’ll begin with an old friend, machine functionalism. I have been focusing on machine functionalism because it is familiar and because it seems reasonably straightforward. But machine functionalism turns out to be a more complicated theory than almost anyone suspects. Still, most everyone is eager to distance him- or herself from machine functionalism. Lycan (1987) uses it only as a starting point. As noted earlier, Putnam says, “I do not know of anyone who ever actually held the one-computational-state version of functionalism to be right” (1988: 84). And Daniel Dennett writes, “I am not a Turing machine functionalist, but then I doubt that anyone ever was, which is a shame, since so many refutations then have to go to waste” (1991: 460). So it might seem prudent to move along to its successors. But machine functionalism is more than a historical curiosity. Functionalists trace their ontological pedigree to Putnam and to the concepts of machine functionalism; it is the fountainhead of all variations of metaphysical functionalism. According to the orthodox picture that I rehearsed at the start of the previous chapter, current functionalist theories have the same basic ontological form as the old. On that view the evolution of functionalism is a smooth progression from Putnam in the 1960s up to the latest versions favored by Lycan, Dretske, Tye, and many others. In particular, the realization relation is assumed to remain the same from Putnam onward— making machine functionalism (whether or not anyone ever held it) an unwitting accomplice to the idea that varieties of functionalism invoke a single realization relation. That is the idea I am putting under scrutiny. If we’re going to make any sense of realization for any notion of function, we are obliged to begin with the notion used by machine functionalism. According to machine functionalism, being a mental state is being a realization of a functional state of a machine: “To be in a mental state M is to realize or instantiate machine program P and be in functional state S relative to P” (Lycan 1987: 8). Earlier we considered machine functionalism as an example of metaphysical functionalism in contrast to, say, semantic functionalism. Now we are concerned with the notion of function that machine functionalism employs and
141 Functional Realizations
thus with the question: How does a physical state realize a functional state relative to a machine program—what is the realization relation for machine functionalism? To begin to answer this question, we need to know more about “functional states relative to machine programs.” Machine programs are usually explained by reference to Turing machines. The traditional way of describing a Turing machine is in terms of a device that consists of a finite number of internal states, a tape divided into distinct regions in which appear symbols from a finite alphabet, and a scanner capable of moving along the tape and of detecting and changing the symbols on the tape. Each of the states the device could be in has the “function” of mediating the inputs read from the tape, outputs written to the tape, and movements of the scanner along the tape. There is a finite number of states that the device can be in, of symbols that it can encounter on the tape, and of actions it can perform. So the complete repertoire of behaviors the device is capable of can be specified by a finite set of statements of the form “If in state w and encounter symbol x then do action y and go into state z.” The set of all such statements specifying possible behaviors of the device constitutes a program for the machine. Although machine programs can be written out as lists of statements, they are more efficiently specified in the form of a machine table. Figure 5.1 illustrates a machine table for addition of numbers represented on the tape in unary (“prisoner’s tally”) notation. According to Putnam: “A ‘machine table’ describes a machine if the machine has internal states corresponding to the columns of the table, and if it ‘obeys’ the instruction in the table in the following sense: when it is scanning a square on which symbol s1 appears and it is in, say, state B, that it carries out the ‘instruction’ in the appropriate row and column of the table” (1960, in 1975c: 365). The internal states of a machine that is described by a machine table are the “functional states” of the machine, relative to the machine table. The relations mediated by a particular functional state together compose the functional role that is occupied by that state; functional state kinds are constituted by their functional roles. According to machine functionalism, mental state kinds are functional state kinds of exactly this sort. Jaegwon Kim, for example,
142 Chapter 5
A
B
C
1
print 1;
print 1;
print [blank];
move right; enter state A
move left; enter state B
move left; stop
+
print 1; move left;
print +; do not move;
print +; move left;
enter state B
stop
stop
print [blank]; do not move;
print [blank]; move right;
print [blank]; move left;
stop
enter state C
stop
[blank]
Figure 5.1 Machine table for addition. Adapted from Putnam (1960).
writes, “It isn’t merely that anything with mentality has an appropriate machine description; machine functionalism makes the stronger claim that its having a machine description of an appropriate kind is constitutive of its mentality” (1996: 91). Metaphysical functionalism identifies mental kinds with functional kinds; for this reason it is sometimes called a functional state identity theory. If mental state kinds are identical with functional state kinds, the question of what makes a particular state the mental state it is is the same question as that of what makes a particular state the functional state it is.1 What makes a particular state a functional state of a machine? When a device is described by a machine table it realizes the kind of machine for which the table is a program, and states of the device realize the functional states of machines of that kind. Many otherwise careful thinkers are therefore overcome by the temptation to believe that realization is the relation of being described or describable by a machine program. For example, Jaegwon Kim writes: “We will say that a Turing machine M is a machine description of system S, relative to a given input/output specification, just in case S realizes M relative to that input/output specification. Thus, the relation of being a machine description of is the converse of the relation of being a realization of ” (1996: 87). But Kim gets off on the wrong foot when he
143 Functional Realizations
identifies a Turing machine with a machine description (“a Turing machine M is a machine description . . .”). If we were interested in how to realize descriptions then maybe describability would be the relation we’re looking for.2 But we are not interested in instances of descriptions; we are interested in instances of kinds of machines and of the functional states of those machines. Kim conflates Turing machines with machine programs, making it seem that realization is the relation between machine programs and physical states. But it is not. We may talk loosely about “realizing programs” as long as we are careful to remember that it is only shorthand for “realizing the kinds described by programs.”3 Caution is called for lest we mistakenly think that realization is describability. Realization is the relation between a particular state and the kind of state described by a machine program. Describability, however, is not a relation between particulars and kinds; it is a relation between particular things and particular descriptions, like machine programs. That is not to say that we are not interested in machine programs, only that we must keep them in their proper place. A machine program describes states of a functional kind by specifying the properties that constitute the identity conditions for that kind. To be a member of the kind is to have those properties.4 Something that has those properties will, of course, be describable by the program; but it is having the specified properties that makes something an instance, not the describability. So describability is only standing in for the having of some other relations. Moreover, if the relations specified by machine tables are not all of a sort—as may be the case— then the relations that underwrite both describability and realization are also various. If this much is correct, then it easy to understand why “functional equivalence” accounts of realization fail to tell us what we want to know. Consider Hartry Field’s proposal: “Suppose that the specifically psychological primitives in [a theory] H are T1, . . . , Tn; then we can write H as (T1, . . . , Tn; x, t). For simplicity, I will assume that an n-tuple [P1, . . . , Pn] of properties realizes H in an organism X at time t if and only if the formula (Y1, . . . , Yn; x, t) is true of [P1, . . . , Pn, X, t]” (1978, in Block 1980b: 91–92; original typography
144 Chapter 5
adapted with standard quantifiers). This formalism says nothing more than that something has a property M if and only if we can truly assert “M” of it. It tells us that a certain complicated predicate— a psychological theory—will apply to all the instances of a functional kind.5 But this is not even enough to ensure that the kinds are functional in Van Gulick’s (1982) and Shoemaker’s (1981b) “strong” sense. That is, it does not ensure that not only is every mental state itself a functional state, but that no state is “anchored” to a nonfunctional state (Rey 1997). If all the relata picked out by the theory H are not themselves functional states, then although H is a functionalist account in a sense, it does not identify the kind of state with which metaphysical functionalism is concerned (Shoemaker 1981b). H would be a functional theory only in a “weak” sense; some states in H could fail to be completely specified by their functional roles. Metaphysical functionalism makes the stronger claim that “type identity conditions for mental states can be specified purely in terms of their mutual interconnections and relation to stimulus conditions and behavior” (Van Gulick 1982: 186; see also Shoemaker 1981b). Functional equivalence formulations of realization do not by themselves guarantee that a state is a functional state in Van Gulick’s and Shoemaker’s strong sense. To understand realization we need to know the relation between particulars and functional kinds—we need to know what makes a functional predicate true of a particular system. Equivalence formulations explain only the relation between instances, not the relation between kinds and their instances. As Fodor writes, It might be supposed that the following principle, at least, can be taken for granted: if m1 and m2 are functionally equivalent mechanisms, and if e1, e2, . . . en are the series of effects of m1, then somewhere in that series there must be an ei such that ei . . . en are effects of m2. This, presumably, is what we have in mind when we say of functionally equivalent systems that they do the same things. Functionally equivalent systems must somehow contrive, in the long run, to produce the same states of affairs. Even this sort of suggestion is, however, relatively unhelpful for an attempt to clarify the concept of functional equivalence. For it appeals to the concept of “same state of affairs” that is unclear in much the same way that sameness of function is. (1968: xviii–xix; italics removed)
145 Functional Realizations
We must already understand how to characterize the ways in which states are alike if a notion of equivalence is going to be at all helpful. If states are supposed to be alike with respect to their functional properties, then we need to know what it is for a system to realize a functional property. But the mere assertion of equivalence does not tell us that. Functional equivalence formulations of realization fail to answer the basic question about realization relations or metaphysical functionalism. They are not metaphysically deep. We can’t cash out the idea that two systems are functionally equivalent until we understand what it is for a system to realize a function. What about machine tables? Don’t machine programs describe classes of functionally equivalent systems? Each such system is identified by its relational properties. A machine table specifies the properties had by instances of a kind of machine; and to be an instance of that kind of machine is to have those properties. To understand the realization relation for machine functionalism, we need to know what kinds of properties and relations are named by Turing machine programs. That might seem an easy enough matter. But it is not, for a well-known ambiguity infects functionalist treatments of Turing machines in philosophy of mind from the first.6 Sometimes Turing machines seem to be physical devices, as when Putnam writes, “Any machine that is described by a machine table of the sort just exemplified is a Turing machine” (1960, in 1975c: 365). But Turing machines are also said to be abstract entities to be explicated in terms of the metaphors of states, tapes, scanners, and inputs and outputs: “a given ‘Turing machine’ is an abstract machine which may be physically realized in an almost infinite number of different ways” (ibid.: 371). Putnam sometimes marks this distinction by using scare-quotes when he is talking about the abstract Turing machine and its “states,” “tapes,” and “scanners.” Both uses have currency in the philosophical literature. The ambiguity infects all the parts of Turing machines: states, symbols, scanners, machine tables, and so on. The relations specified by the machine table in figure 5.1, for example, are relentlessly causal (print, move, enter a new state). They call for causal interactions, state changes, and movements. It is a machine table for a type of physical device. The functional states specified by that table are
146 Chapter 5
physical state kinds, not abstract kinds. Of course one might say that the relations are abstract in the sense that they can be fulfilled by more than one physical state, or even more than one kind of physical state—that is, they are multiply realizable. But that does not make the states abstract in the manner of an abstract Turing machine. Probably it would be better to think of the states specified by the machine table in figure 5.1 as incompletely specified physical states. Turing machines may be of that sort; or they may specify transformations that are abstract in the traditional sense—as with relations between “objects” like numbers and mathematical functions. According to machine functionalism, mental states are to be identified with functional states of systems that realize certain machines. The question therefore arises whether mental states are to be identified with states of abstract machines or states of physical machines.7 Of course functionalists hold that token mental states are token physical states. But that says nothing about whether mental state kinds are physical kinds or abstract kinds. We have every reason to suppose that realizing an abstract kind (being an instance of an abstract kind) is different from realizing a physical kind (being an instance of a physical kind). So if we are going to understand machine functionalism, we have to determine whether it takes mental states to be instances of abstract machine states or instances of physical machine states. Rather, since both are possible views, we need to assess whether either is a good candidate for Generic Functionalism. Let us begin with abstract machine functionalism. What is the realization relation if machine functionalism is the view that mental states are instances of functional states of machine programs for abstract Turing machines? A machine table for an abstract machine would have to specify abstract entities, for example, rather than symbols, and state-space transitions (or some such) rather than causal relations. How could a physical device have states that stand in the relations—occupy the roles—specified by an abstract machine program? A common idea is that a physical device realizes or instantiates an abstract machine indirectly. Robert Cummins uses the example of an adding machine:
147 Functional Realizations
We may think of the button-pressing sequences as arguments to a function g that gives display states as values. An adding machine satisfies g ; that is, the arguments and values of g are literally states of the physical system. Addition, as was remarked above, relates numbers, not physical states of some machine, so a physical system cannot literally satisfy the plus function. What an adding machine does is instantiate the plus function. It instantiates addition by satisfying the function g whose arguments and values represent arguments and values of the addition function, or in other words, have those arguments and values as interpretations. (1989: 89)
On Cummins’s story a physical device (an adding machine) realizes an abstract function (addition) by “satisfying” a function g that relates physical states (button presses, displays). This is the picture that seems to show that realization is transitive. A physical system realizes a causal function, which in turn realizes an abstract function, so the physical system realizes the abstract function by transitivity. Some authors take transitivity to be a central feature of realization (Horgan and Tienson 1996). Cummins says that to “satisfy” a relation is literally to have that relation; the adding machine satisfies function g. It has that function. So Cummins’s satisfaction is a realization relation in my sense. But satisfaction is not the realization relation for an abstract functional kind like addition, for no physical thing can literally have those relations. This difference explains why Cummins invokes two relations. Cummins’s satisfaction and instantiation are two different realization relations for different notions of function, one for causal and the other for abstract functions. Realizing function g is a matter of satisfying the function. But realizing the adding function is not a matter of satisfaction. The realization relation for mathematical functions is something like having those arguments and values as interpretations. Cummins is arguing that the adding function is realized by whatever represents the values of addition, by whatever has those “values” as “interpretations” of its inputs and outputs. If this is correct, then function g does not itself realize addition—g does not itself have inputs and outputs (physical states such as button presses, displays) that represent or have as interpretations abstract “values.” There may be cases of the transitivity of realization, but Cummins’s example is not one of them because function g does not itself realize addition.
148 Chapter 5
Nevertheless a physical device may realize addition, and it may do so directly, rather than transitively or indirectly. The physical device realizes the abstract relations by having them as its interpretation: “Instantiation of such a formal machine description requires roughly that there be some mapping from the formal states, inputs, and outputs of the abstract machine table onto physical states, inputs, and outputs of the instantiating system, such that under that mapping the relations of temporal sequence among those physical items are isomorphic to the relations of formal succession among the machine table items” (Van Gulick 1988b: 80). A physical device can both realize Cummins’s function g and realize addition; but function g is at most instrumentally useful in the realization of abstract functional kinds. If the situation is properly arranged, a device that satisfies function g will have addition as an interpretation. In this sense, the system realizes addition by realizing g. The abstract functional realization relation is having such-and-such (mathematical or abstract) arguments and values as interpretations. This account makes sense of how a physical state can realize an abstract state. But from the view of metaphysical functionalism it has drawbacks: Although having an interpretation is presumably synchronic and computational kinds are no doubt biologically abstract, it is doubtful that having an interpretation is an objective relation. This raises serious questions about whether the resulting functional kinds are causally efficacious in the strong sense. But this result is not at all surprising if we are considering abstract functions. Since we were investigating the realization of abstract objects and properties, we got the correct result. Of course realizations of abstract objects and properties do not have causal powers! What this shows is only that abstract functions are not suitable for metaphysical functionalism. This should have been obvious all along but for the conflation of abstract and causal Turing machines. Having an interpretation (which really can be realization for abstract machine functionalism) should not be confused with the describability formulations of realization. Whereas it may be relatively subjective whether or not some system has numerical values as its “interpretation,” it will not be a matter of interpretation whether a system is “describable” as having causally specified relations.8
149 Functional Realizations
According to some intentional and semantic functionalists, for a system’s state to have intentional content is for them to map onto the world—for there to be isomorphism between the system and the world. One might then say that the relations between the system’s states are in some abstract way functionally equivalent to the relations in the world, or that they have semantic values as “interpretations.” If we combine this view with something like Brentano’s thesis that “intentionality is the mark of the mental,” then it will seem that such abstract functional equivalence is all that is needed for a system’s states as a whole to realize mental states. Whether a system is a mind is just a matter of whether its states have the world as an interpretation. So easy! The trouble is that the notion of functional equivalence is only well defined for specific functions. Metaphorical appeal is not enough; “functional equivalence to the world” has been given no sense. There is no isomorphism an sich. Before we can assess equivalence we have to know what sort function to use. In the example at hand, I suppose we’d have the know the “world function,” whatever that could mean. Putnam himself has recently advanced this observation as part of a general argument against functionalism (1988: 84; 1999).9 The lesson, again, is that we’re not going to get an account of minds from abstract relations alone. We’re going to need some relations that give functionalism a metaphysical grip. 2 Causal-Role Functions It seems that machine functionalism is not the doctrine that a mental state is a state of a system that realizes an abstract machine program, after all. What, then, is the realization relation if machine functionalism is the view that mental states are instances of functional states of machine programs for physical Turing machines? Realizing a physical machine is a matter of having the relations picked out by a machine table such as the one in figure 5.1, which names causal relations. We might call this view computational functionalism, because it appeals to the sort of functions that ordinary programmable computers are thought to perform.10 machine functionalism, viewed this way, is a special case of causal-role functionalism.
150 Chapter 5
Functionalists these days typically take mental states to be causalrole functional states. Often causal-role functionalism is treated as the only sort. Kim assumes as much when he writes, “[a]ccording to functionalism, a mental kind is a functional kind, or a causal-functional kind, since the ‘function’ involved is to fill a certain causal role” (1996: 76). Similarly, Fodor claims that functional states “are typeindividuated by reference to their (actual and potential) causal relations; you know everything that is essential about a functional state when you know which causal generalizations subsume it” (1985, in 1990: 13). That functionalism in philosophy of mind is best understood in terms of causal functions is the orthodox interpretation of the theory, itself the received view in philosophy of mind. The best-known treatment of causal functions is Robert Cummins (1975). His goal is not an understanding of function per se; he writes, “it is the analytical style of explanation, especially as applied to complex capacities, that interests me, not the proper explication of the concept of function” (1983: 195, fn. 2). Despite protests, Cummins is widely taken to have provided an account of the causalrole notion of function: “x functions as a f in s (or: the function of x in s is to f) relative to an analytical account A of s’s capacity to y just in case x is capable of f-ing in s and A appropriately and adequately accounts for s’s capacity to y by, in part, appealing to the capacity of x to f in s” (1975: 762). The causal-role function of a thing is the effect it has in a system that helps explain the total capacities of the system. The function of a certain gear in a watch is the effect it has that figures in an explanation of how the watch keeps time, for example. Causal-role functions are therefore functions relative to a system and an explanation. Metaphysical functionalism, interpreted thus, claims something like: To be a mental state M is to realize causal-role functional state (in the strong sense) F relative to system S and explanation Q of S ’s capacity to y. To explain what it is to be a mental state is to explain how it is that appropriate functional states are realized. And Cummins’s treatment of causal-role functions provides a way to explain how mental properties are realized: To realize a causal-role function is to have the
151 Functional Realizations
appropriate causal relations in a system, relative to an explanation of the system’s capacities. The kind of explanation that led us to this point is that provided by physical machine programs. It is a form of causal-role functionalism that is relativized to the causal explanation a physical machine table gives of the system’s behavior: “[W]e can think of the Turing machine approach as a special case of the Ramsey-Lewis method, in which the psychological theory is presented in the form of the Turing machine table with the internal machine states, the qs, corresponding to the predicate variables, the Ms” (Kim 1996: 107). A version of causal-role functionalism so constrained as to be silly might further restrict the kinds of systems we consider to those consisting of tape heads, paper tapes, and the other rudimentary Turing machine components. But functionalists usually suppose that there will be few if any such limits on the realizers of mental systems. This feature, after all, is what secures claims of standard or radical multiple realizability. Turing machine programs are only one special kind of causal explanation.11 If constraints are relaxed—if any causal explanation can be treated as a program, and there are no restrictions on what objects the so-called program appeals to—then computational functionalism (physical machine functionalism) is simply causal-role functionalism. This is no complaint, for causal-role functionalism is a flexible and potentially powerful doctrine. Yet important questions must be answered. There is still the matter of determining which are the appropriate systems and explanations. Peter Godfrey-Smith notes that “[i]f functions rest on explanation, then it makes a difference how explanation is understood” (1996: 17). Even among causal notions of function, then, there are choices to be made. The trouble is that “The existence of a ‘system’ can be a matter of choice or convention, as can the capacities of the system deemed worthy of explanation” (ibid.). Godfrey-Smith has in mind general questions about what constitutes an explanation; but his point applies to less general concerns as well, such as which causal explanation of mind is the right one—or even, the right sort—for ascribing causal-role mental functions. The conventional aspect of causal-role functions threatens their objectivity and thus may undermine Generic Functionalism.
152 Chapter 5
A great deal rests on the explanatory foundation that functionalists choose. The explanation to which a causal ascription is relativized determines the kinds of relations that constitute a functional role and therefore the kinds of objects apt for filling the role. Different explanations thus yield variations among causal-role functions. By adjusting the particular causal relations and system boundaries one can generate endless assortments of functional-role explanations. Some are entirely unsuitable for theories of mind, such as the implausibly strict Turing machine functionalism mentioned earlier. Which causal explanations, if any, are good candidates to flesh out Generic Functionalism? Although we’ve arrived at the topic from somewhat different considerations, the issue we are now considering is akin to Ned Block’s (1978) “problem of inputs and outputs.” Functional states, in Van Gulick’s and Shoemaker’s strong sense, are constituted entirely by their relations; so the identity of each state is a matter of, speaking loosely, its “inputs” and “outputs.” Block (1978) uses a now familiar litany of examples to argue that any specification of causal-role function will result in either too many kinds of systems being counted as having minds (liberalism) or too few kinds (chauvinism). He poses the question: How can functionalists specify functional relations so as to avoid liberalism and chauvinism? Block argues that there is no satisfactory solution, and that the problem is fatal for functionalism. But a difficulty with Block’s argument is that it brazenly presumes that we already know how to identify those things that have minds, that we are merely looking for the theory that also ascribes mentality to all and only those things. I don’t think we have this knowledge with much confidence about all familiar creatures, much less outlying cases. And I certainly do not know how we could settle the matter for all the possible creatures we might encounter without the aid of a theory about minds. Whereas it is a defect in Block’s argument that he assumes he knows which systems are minded, this problem does not arise for the questions we are now pursuing about functions. We are not worried whether a variety of functionalism that appeals to that kind of function will be too liberal or too chauvinistic. (Of course we might later determine that our theory ascribes or fails to ascribe conscious states
153 Functional Realizations
of an appropriate sort to some particular system—ourselves, computers, automobile gauges, thermostats, rocks, or buckets of water whose molecular motion temporarily mimics a finite state machine— in which we have a particular interest. We might then decide that our theory is not the right theory on those grounds.) For now we are not concerned with which creatures a functionalist theory applies to. Rather, we want to know whether it can do the metaphysical tricks that Generic Functionalism requires. The present concern is that because causal-role functions are relativized to explanations we need a relatively objective explanatory schema to characterize mental functions. We are interested in any proposal for such a privileged explanation against which to specify the causal-role functions. If we can find any, then later we can concern ourselves with whether some are to be preferred on account of being more or less chauvinistic or liberal than others. But if a functionalist theory of minds cannot at least satisfy the demands of Generic Functionalism, then it won’t matter whether the theory is too liberal or chauvinistic. Remember the promise of Generic Functionalism. Any theory giving us strongly causal, objective, synchronic, and biologically abstract functional kinds would be exciting. No doubt some versions of causal-role functionalism can satisfy some aspects of Generic Functionalism. The question for functionalists is whether there is a single version of functionalism that can do all the tricks at once. Causalrole versions of functionalism get off to a good start. Realizing a causal-role functional state—having a causal-role function—is a matter of having certain causal relations. Since having causal relations is something that a system does, as it were, in the present tense, the realization relation for causal-role functions is synchronic. This is progress toward the robust realization relation that functionalists need. But causal-role functions are relativized to causal explanations: “What distinguishes a function from a mere effect is causal/explanatory importance and no more” (Godfrey-Smith 1996: 19; see also Polger and Flanagan 2002). So whether the resulting states are objective, causally efficacious, and more abstract than biological states—whether causal-role functionalism can satisfy these other characteristics of Generic Functionalism—depends on the explanations to which functions are relativized. Objectivity, in particular,
154 Chapter 5
looks to be a problem for an account of functionalism built on causalrole functions. At the very least functionalists must have some nonconventional way of selecting a variety of causal-role function, and thus of selecting the explanation to which functions are relativized. They need some relatively objective way of specifying systems, and thus a relatively objective kind of an explanation. Psychological theory is the obvious candidate for a relatively objective theory that could be the basis for causal-role functionalism. If psychology provides explanations in terms of causal relations, then a psychological theory might be the sort of nonarbitrary explanatory scheme that could ground functional ascriptions. Even if this is correct, we need a way of choosing the right causal psychological theory for the job.12 The basic candidates are familiar: commonsense “folk” psychology, analytic “folk” psychology, and scientific psychology. Consider Lewis’s commonsense psychological theory and Shoemaker’s analytic psychological theory. We briefly encountered both of these in chapter 3 when we considered versions of theoretical functionalism. Lewis’s (1970, 1972) proposal is to construct a “commonsense” psychological theory out of familiar “platitudes” about mental states. The general idea is that mental state terms are theoretical terms of this “folk” theory, so their meanings can be defined by their typical uses, in particular their uses in truistic folk psychological explanations. Lewis suggests that the generalizations might have the form, “When someone is in so-and-so combination of mental states and receives sensory stimuli of so-and-so kind, he tends with so-and-so probability to be caused thereby to go into soand-so mental states and produce so-and-so motor responses” (1972: 257–258). Perhaps the generalization that “someone who is pricked with a pin will likely feel pain and flinch” counts as such an explanation. Shoemaker, however, worries that because Lewis’s folk psychological theory may include a posteriori claims it is unsuited for functionalists’ purposes: Assuming that our “common-sense platitudes” are synthetic, if one of them should be contradicted by scientific findings we presumably would not want to include it in the theory to which the Ramsey-Lewis technique is applied; we do not want to define our mental terms in terms of a false theory. And
155 Functional Realizations
if our functional “definitions” are not meant to capture the meaning, or sense, of the mental terms, why should the information included in them be limited to facts that are common knowledge? (Shoemaker 1981b, in 1984: 273)
To avoid this pitfall, Shoemaker proposes to limit the functionalist theory to the analytic truths of folk psychological theorizing.13 This theory might not include the alleged generalization about pin pricks, but perhaps it would include such rules as “painful stimuli are typically avoided.” Whether we adopt Lewis’s or Shoemaker’s version, we will treat the theory as fixing the meanings of its mentalistic terms, thought of as theoretical terms. If this were right then we would have strong reason indeed to privilege the explanations offered by those theories. That would be objectivity enough, I suppose. What then of the other requirements of Generic Functionalism? The answer to this question leads us into dicey territory. Lewis (1969) himself argued that his account implies species-specific identity claims; such states might not be more abstract than biology. On the other hand, Lewis (1980) contends that the term “pain” applies to states of Martians as well as to states of human beings and octopi. In chapter 1 I argued that we do not typically ascribe empathetically the same mental states to creatures that are quite different from us in their biology. If I am right, it is not a platitude of commonsense psychology that humans, dolphins, octopi, and Martians can all have empathetically the same pain states, or that we will all respond similarly to similar stimuli. Certainly we do not regard such claims as analytically true. Of course I have nothing but my own linguistic intuitions to put up against Lewis’s and Shoemaker’s. But it seems to me that it can be credible that folk or analytic psychology fixes the meanings of mentalistic terms only if those terms are also restricted to relatively local species, and in fact to human beings. I don’t think that our commonsense psychology contains generalizations, analytic or otherwise, about robots or aliens. In short, neither folk nor analytic psychology seems to be biologically abstract.14 Perhaps more important, it is by no means obvious that folk or analytic psychology is functional in Van Gulick’s and Shoemaker’s strong sense. That is, these theories may involve some causal relations between mental states of kinds that are not themselves
156 Chapter 5
constituted only by their causal relations. Here I have in mind the usual sorts of “qualia”-based objections to functionalism, among others.15 According to these arguments, two creatures might be exactly the same from the point of view of functional psychology but differ in the experiences that they have. We seem to be able to imagine cases in which all the commonsense or analytic generalizations about pain are satisfied by some state that we are not inclined to count as a pain state, or in which a pain state satisfies some generalizations other than those that are supposed to be definitive of pain. Indeed the usual tactic against functionalism has been to press such cases directly in an attempt to show that functionalist psychology is incomplete. I have been pursuing a somewhat different strategy, and so I will not lean heavily on these sorts of considerations. Yet they cannot be ignored, for they raise serious questions about whether functionalist psychology is exhaustive. If it is not, then psychological theories will fall short as the explanations to which functions are relativized, for there will be some nonfunctional mental states and any resulting theory would not be a version of metaphysical functionalism. There is some reason to doubt, then, that either folk or analytic psychological functionalism are metaphysically “strong” enough or biologically abstract enough to satisfy Generic Functionalism. But the real trouble lies with Lewis’s hypothesis about the meanings of theoretical terms. If psychology gives the meanings of mental terms, then that is good reason to think that relativizing functional states to folk psychological explanations is objective. But only a few functionalists maintain Lewis’s claim about the meanings of mentalistic terms. The reasons for this are many and complicated and would lead us far from our present considerations; but I will mention two. First, Kripke (e.g., 1972/1980) and Putnam (e.g., 1975a) were largely successful in persuading philosophers to abandon the Millian theory of linguistic meaning that Lewis assumes. They argued that the references of natural kind terms—mentalistic terms included—are determined by a causal chain of usage. If this is correct, the question of what pains are cannot be settled by “defining” pains in terms of the current usage of the word “pain”, or of the typical characteristics of states we call “pains”. The nature of pain is a topic of empirical inquiry rather than linguistic analysis. Thus Putnam’s so-called psy-
157 Functional Realizations
chofunctionalism, which we will come to in a moment, bases its analyses on empirical psychology instead of commonsense platitudes. As Shoemaker explains the difference: “These two versions of functionalism do not differ in the sorts of ontological claims they make. Both assert that mental states are identical to . . . states expressible by functional predicates containing no mental vocabulary. . . . We might say that the analytical functionalist looks for the functional characterizations that give the ‘nominal essence’ of mental states, while the Psychofunctionalist looks for functional characterizations that give the ‘real essence’ of such states” (1981b, in 1989: 272–273). So the difference between the functionalisms is as much a difference in theory of reference as it is in substantial thesis about minds. The second reason for forgoing Lewis’s meaning hypothesis predates the dispute about reference, and it is also related to the “qualia” objections mentioned in passing earlier. Lewis holds that it is a contingent fact that sensations are brain processes, but that it is a necessary truth that, for example, pinpricks normally cause flinching. The relationship between pains and their typical characteristics is necessary because it is definitional, on Lewis’s account. But some philosophers, including Smart (1959), claim that the symptoms of mental states are contingent. Smart denies that expressions in terms of brain states are translations of or have the same meaning as mental state talk (1959: 144). This stance is further enabled by Kripke’s arguments, which show that the descriptions Lewis takes to be definitive of mentalistic terms could be used to fix their reference without fixing meaning. In such a case, it will be contingent that pain is the state that has the causes and effects that in fact typify it; it is possible for some other state to have those characteristics and for pain to have different characteristics.16 But this contingency is in conflict with the meaning thesis; for if mental states are defined in terms of their functional roles then the fact that they have those roles will be necessary. Lewis himself aims to accommodate these sorts of cases (1980). He offers an account that mixes aspects of functionalism with a species-specific identity theory; it is not functional in the strong sense, so it is not metaphysical functionalism. Much remains to be said. Even if folk or analytic psychology fixes the meanings of mentalistic terms, doubts linger about the biological abstractness of the resulting accounts. Most of the
158 Chapter 5
advocates of causal-role functionalism follow Putnam and argue that the causal-role functions come from scientific psychology. According to Putnam mental states should be understood as functional states, and their constitutive relations are the laws and generalizations of empirical psychology. “Psychofunctionalism,” as Block dubbed it, was widely influential. Before long it became the dominant view in philosophy of mind. An attractive feature of Putnam’s functionalism is that it takes full advantage of empirical psychology. To many, the growth of cognitive psychology in the 1970s and the so-called cognitive sciences in the 1980s appeared to offer boundless support for functionalism. These empirical sciences reversed the behaviorist trend and studied mental states as causally efficacious internal states. They proceeded in a manner largely independent of underlying biological and neurosciences, and took as their objects not only human beings but animals and sometimes computers. And they were computational always in spirit and often explicitly in their theorizing; the notion of “information,” though rarely examined and assumed to be grounded in computational theory, was central to almost every cognitive psychological explanation. Psychologists and philosophers seemed to be speaking a common tongue. Marr’s (1982) theory of vision, mentioned earlier, was often cited as illustration of the convergence. In short, it seemed that empirical psychology had borne out Putnam’s bold hypothesis, even as he himself demurred (1988). Although I do not share Putnam’s reasons for hesitation, I think his prudence is keen. Recall the many kinds of functionalism canvased in chapter 3. There is no doubt in my mind that many varieties of functionalism are compatible with much recent empirical work in cognitive psychology, cognitive neuroscience, and other cognitive sciences. Furthermore, I acknowledge that some functionalist theses have indeed been borne out in empirical theories. Functionalists are right to boast the support of the cognitive sciences. But I submit that many of the functionalist successes accrue to methodological, explanatory, and theoretical functionalisms. Metaphysical functionalism is another matter. I do not think that psychology aims for, nor has it stumbled on, a theory that fits Generic Functionalism. In chapter 1, for example, I argued that much theorizing in psychology
159 Functional Realizations
is interpendent with theorizing in neuroscience, suggesting that psychology will not be biologically abstract in the way that functionalists expect. For similar reasons, it is not at all clear that empirical psychological theories are functional in Van Gulick’s and Shoemaker’s strong sense. There is no doubt that empirical psychology is chock-full of broadly functionalist explanations, theories, and methodologies. But is empirical psychology metaphysically functionalist? Does it treat those explanations and theories as exhaustive characterizations of mental states? I think the answer is no. If psychology is indeed working in tandem with neuroscience to discover the mechanisms of mind, then the functional explanations and theories of empirical psychology should not be regarded as accounts of the nature of mental states. It is preferable to think of the functionalist theories as providing reference-fixing descriptions of mental mechanisms rather than as defining the real essences of such states. Empirical psychology may be to a large degree methodologically and explanatorily independent of neuroscience, but it is not metaphysically independent. I tentatively conclude that empirical psychology is not suited to the task of grounding metaphysical functionalism, but I recognize that I owe a more thorough argument. I’ve provided no evidence that empirical psychology is not functionalist in the strong sense. And even if it is not, one might count that only as a defect of current psychological theories; I haven’t provided an argument showing that psychology shouldn’t be metaphysically functionalist. My tentative conclusion is supported in part by the sorts of observations about psychological and neuropsychological theorizing discussed in chapter 1. Those considerations support the reference-fixing interpretation of psychological theories. But much more remains to be said. In the next chapter I will explore in greater detail the kinds of functional explanation that seem to support metaphysical functionalism. There I will argue that psychological explanation need not be metaphysically functionalist, and that perhaps it should not be. Where does that leave us now? I have been considering whether some variety of functionalism that appeals to causal-role functions is a version of metaphysical functionalism that satisfies the require-
160 Chapter 5
ments of Generic Functionalism: causal power, objectivity, synchrony, and biological abstractness. My argument has been that because causal-role functional states have functions only relative to some explanation, the explanation will have to be privileged in some way or else there is little hope of satisfying the requirement that the realization relation and resulting kinds be objective. The best candidates are causal-role functions that are relativized to a psychological theory; but that approach does not guarantee that the resulting theory will be metaphysically functionalist in the strong sense or (even if it is) that it will be biologically abstract. This verdict is not uncontroversial, nor are my reasons for reaching it. Moreover, my wrangling with the variations of causal-role functionalism may seem perverse since there is a well-known argument that seems to provide trouble for causal-role functionalism regardless of the quarrelsome technicalities we have be sorting through. The argument is simple: Realizing a causal-role functional state is a matter of standing in certain causal relations. But standing in causal relations does not endow a state with any new causal powers. A state has its causal powers whether or not it happens to be embedded in a system covered by some explanation. Thus having a function—realizing a functional state—does not endow a state with any distinctive causal powers. In particular, the property of being a functional state is impotent. So if mental states are functional states, then mental states are epiphenomenal. This is a version of what Jaegwon Kim (e.g., 1998) calls the causal exclusion argument, for it seems to show that the causal powers of a physical state exclude causal powers from being attributed to the functional state that it realizes.17 Since metaphysical functionalism identifies mental states and properties with functional states and properties, the exclusion argument concludes that mental states are not causally efficacious. This will be the result whether mental properties are identified with realized or realizer properties. If mental properties are identified with the higher-order properties that are realized by lower-order properties, the exclusion argument applies directly; the causal powers of the higher-order property are excluded
161 Functional Realizations
by those of the lower-order realizer. On the other hand, if mental states are identified with first-order realizer properties, then they have the same causal powers as the realizer. But in that case they have causal powers qua realizer properties but not qua realized properties. The causal powers of the first-order realizer are simply those it would have whether or not it was acting as a realizer. Again the functional properties do no causal work. So the causal exclusion argument shows that functional states and properties can have causal powers in at most a weak sense. Many worry that this weak kind of causal efficacy is no more than epiphenomenalism (Kim 1996, 1998; Horgan 1989, 1997, 2001; Antony and Levine 1997; Antony 1999b). Opponents of the exclusion argument often respond that the causal exclusion argument proves too much. They argue that if the causal exclusion argument applies to mental states, then it generalizes to other states like brickness or hurricaneness. And that is absurd, for we know that bricks and hurricanes have causal powers. One response to generalization concerns is to accept a deflationary approach to macrocausation. The causal exclusion does apply to bricks and hurricanes; it’s just that we do not care if bricks and hurricanes have causal powers qua bricks or qua hurricanes in the same way that we are invested in mental states having causal powers qua mental states. But as long as the causal powers of mental states are on the same footing as the causal powers of bricks and hurricanes, we will have to learn to live with the deflationary result.18 Philosophers of mind, after all, are not obliged to solve all the puzzles of mereology and causation. If bricks don’t have causal powers then neither do minds, regardless of whether the functionalist or identity theorist is right. This unflinching stance is not very attractive, however. I claimed that I would be satisfied to show that mental states have as much causal efficacy as brain states and brick states; but I presumed that they would all have some causal powers, not that they would all be equally epiphenomenal. Kim (1998) has recently advocated a somewhat different response to what he calls the generalization argument. Causal-role functional states, according to functionalists, are higher-order states. That is, a causal-role functional state is the state of standing in such and such
162 Chapter 5
causal relations. (Or if you prefer: A causal-role functional property is the property of standing in such and such causal relations.) To realize a functional state is to have those relations; no new entity or property is created. But for exactly that reason, realizing a functional state seems superfluous, for the lower-order objects and their properties are doing all the causal work. In contrast, brickness and hurricaneness are higher-level properties; they are properties of macrophysical entities that do not exist in the microphysical world. Solidity is not a property of swarms of molecules, it is a property of objects like bricks. A brick is a mereological whole; brickness is a firstorder property of a higher-level (composite) object. In Kim’s (1998) terminology, brickness is based on but not realized by the properties of the molecules that compose the brick. Contrary to the common wisdom of functionalists like Lycan (1987), realization does not create or explain levels, because it is an intralevel relation; realization generates orders, not levels. Distinguishing between levels and orders is supposed to cut off the generalization argument by showing that higher-level properties and higher-order properties are not equally susceptible to the exclusion argument. The reason is that higher-order properties, such as functional properties, are instantiated in the same objects as their lower-order bases or realizers. This is true of causal-role functions: “x functions as a f in s . . . just in case x is capable of f-ing in s and . . .” (Cummins 1975: 762). The same object x has the first-order causal property and the higher-order functional property. Because one and the same object or system has the lower-order and higherorder properties, the two sets of properties seem to be in competition. The lower-order properties exclude the higher-order properties, since they are causally sufficient to account for the behavior of the object or system. The situation for levels is different. Higherlevel properties belong to different objects than their bases. Solidity is a property of a brick, but not of the molecules that compose the brick. Because the higher-level and lower-level properties belong to different objects (bricks and molecules, respectively), they are not in competition and do not exclude one another. Thus higher-level properties are immune from the exclusion argument (Kim 1998).
163 Functional Realizations
Kim argues that the causal exclusion argument applies to higherorder states but not to higher-level states. If this maneuver to block the exclusion argument from generalizing works, he can grant the efficacy of macrocausation while maintaining the exclusion argument vis-à-vis functional properties. Kim’s distinction turns his opponents’ reductio ad absurdum argument on its head. We know that bricks and hurricanes have causal powers; that’s exactly why we know that causal exclusion argument does not generalize. It is worth noting that the causal exclusion argument is not just another objection to causal-role functionalism, or to the causalefficacy of functional properties. I have been satisfied to argue that causal-role functions cannot do the work that functionalists demand of Generic Functionalism. That would be enough for my purposes. But the exclusion argument is different. It threatens to show that the features of generic functionalism cannot be jointly satisfied. Generic Functionalism requires mental-cum-functional states to be biologically abstract. Brains are composite objects; their properties are higher-level properties. A state or property that is more abstract than biology would not be a higher-level state or property, that is, one that is composed of brain states or properties. It would have to be a higher-order state or property that could be instanced by the first-order states or properties of brains or other entities. But if the exclusion argument is correct then no such higher-order object or property can have causal powers. The exclusion reasoning suggests that nothing can be both biologically abstract and causally efficacious. If the exclusion argument is correct then I have been too modest. I claim only that no current version of functionalism satisfies the demands of Generic Functionalism. The exclusion argument might show that Generic Functionalism is internally inconsistent. But I am not ready to claim that victory. Although I am partial to Kim’s distinction between orders and levels, I argued in the previous chapter that the functional realization relation is not a basic metaphysical relation like mereological composition; and from the first I have claimed that composite objects like brains and bricks have causal powers. So I think that Kim is on to something. Nevertheless, I admit
164 Chapter 5
that there are serious questions about the exclusion argument, and about whether the levels/orders distinction can be used to block its generalization.19 I cannot hope to settle the matter here, or even to explore what that would require. For this reason I have made a point of showing that causal-role functionalism faces difficulties other than causal exclusion. I will settle for the modest conclusion if I can get it. And that requires looking at a few more notions of function. 3 Teleological Functions Even some philosophers who believe that causal exclusion worries can be dismissed nevertheless acknowledge that causal-role functions cannot suffice for Generic Functionalism. In growing numbers functionalists have sought a notion of function more objective than plain causal-role functions. For this they look to biology: Most recent philosophy of mind has been “functionalist” in some sense or other. We can distinguish two basic forms of functionalism in philosophy of mind. First there is the more orthodox view which I will call “dry functionalism.” This view understands function in terms of causal role, and it identifies mental states in terms of their typical causal relations to sensory inputs, other mental states, and behavioral outputs. . . . Second there is “teleofunctionalism.” This view makes use of a richer, biological concept of function more closely allied to traditional teleological notions, a concept often analyzed with the aid of evolutionary history. . . . For the dry functionalist, one essential property of any mental state is the pattern of behavioral outputs which the state, in conjunction with the rest of the system, tends to cause in various circumstances. For the teleo-functionalist, what is essential to the mental state is not what it tends to do, but what it is supposed to do. (Godfrey-Smith 1996: 13)
As noted earlier, the idea that things in the natural world—and biological things in particular—have purpose, telos, goes back at least as far as Aristotle. It is in recognition of this heritage that biological notions of function are collectively known as teleological functions. The allure of teleological notions of function is obvious. Causalrole functionalism falters in part because it cannot select an objective explanatory account to which functions are relativized—it cannot objectively specify the inputs and outputs that constitute the functional roles. So functionalists set out to find a more objective
165 Functional Realizations
notion of function: “What we are looking for are functions involving a system of natural signs . . . that is not parasitic on the way we exploit them in our information-gathering activities, on the way we choose to interpret them. We need, then, some characterization of a system’s natural functions” (Dretske 1986, in Lycan 1990: 135). Teleological notions of function specify not just what a thing happens to do, but what it is supposed to do. And the force of the “supposed” comes not from human interests but from evolutionary history. If there is an effect that something is supposed in evolutionary terms to have, then that could provide a privileged and relatively objective notion of function. The idea that biological normativity solves the problem of inputs and outputs looks promising.20 A straightforward application of teleological function to metaphysical functionalism would claim that being a mental state is a matter of realizing a teleological functional state: To be a mental state M is to realize teleological functional state (in the strong sense) F in system S. Among those who have seemed to endorse this view are Millikan (1989, 1993), Dennett (1991), Dretske (1995), Lycan (1987, 1996), Van Gulick (1982), and Flanagan (1992). The diffidence with which I offer this roster is not merely rhetorical; I will argue that some of these philosophers do not intend to advocate a straightforward application of teleological notions of function to functionalism. (Some do—viz., Millikan and sometimes Dretske, as we shall see.) But the philosophers on my list are all widely thought to hold the straightforward teleological view. So it is worthwhile to examine whether a straightforward application of teleological functions to metaphysical functionalism could get us what we want out of Generic Functionalism. I call the straightforward application of teleological functions to metaphysical functionalism direct teleofunctionalism.21 For convenience I will divide teleological accounts of function into two basic groups: etiological functions, and all the others. Both sorts have a common shortfall from the point of view of Generic Functionalism. It should be immediately evident that all versions of direct teleofunctionalism, depending as they do on biological notions of function, will have
166 Chapter 5
little or no hope of generating functional state kinds more abstract than biological kinds. So direct teleofunctionalists cannot avail themselves of the standard multiple realizability arguments for functionalism; they will have to produce alternative reasons for preferring it over identity theory. The different varieties of teleological functions do not have the same consequences for the other characteristics of Generic Functionalism. It is fair to say that among philosophers of mind the most favored account of teleological functions is the etiological account due to Millikan and Neander, mentioned in chapter 3. Millikan (1989) calls this idea proper function. According to the etiological or proper function account, the function of a trait is the effect it has that explains its presence or maintenance due to evolution by natural selection (Millikan 1989; Neander 1991; Godfrey-Smith 1994). Exactly how to formulate this notion of function is a matter of some dispute; in addition, some philosophers maintain that the account is a theoretical proposal, whereas others maintain that it is an analysis of the notion of function as it has been used in biology (cf. Millikan 1989 and Neander 1991). These details are important for understanding functions in biology, but they can be set aside for present purposes. The properties of etiological accounts of function that are relevant to the assessment of metaphysical functionalism are shared by all the variations. Let us take as our exemplar a formulation due to Karen Neander: “It is the/a proper function of an item (X) of an organism (O) to do that which items of X’s type did to contribute to the inclusive fitness of O’s ancestors, and which caused the genotype, of which X is the phenotypic expression, to be selected by natural selection” (1991: 174). Neander’s account is reasonably concise while still giving us a flavor for the complexities that are at issue among philosophers of biology. Her version is sometimes singled out by Lycan (1987) as the notion of function he has in mind, and it is mentioned favorably by Dretske (1995). So Neader’s etiological account is a credible representative of what direct teleological functionalists have in mind. Millikan is the clearest example of a direct teleofunctionalist. For her, etiological functionalism is the correct way to understand the presence of all biological traits, mental states being only one feature
167 Functional Realizations
of human animals: “The position is that psychological classification is biological classification, hence proceeds by reference to teleofunction. This means that categories such as belief, desire, memory, and purposive behavior are biological-function categories—very broad and general ones, of course. Compare the categories limb, hormone, circulatory system, eye, visual system, etc.” (Millikan 1993: 172–173.)22 Mental states are states of brains, but there are also states of livers, hearts, and so forth. Millikan and Dretske both argue that the etiological notion of function lies behind the capacity of organisms to have states that represent and are about the world—to have intentional states.23 Dretske (1995) explicitly extends a similar account to conscious mental states such as sensations, though I will suggest that his version adds some important wrinkles. But the etiological account is entirely unsatisfactory for the purposes of the metaphysical functionalist. To say that something has an etiological function tells us nothing at all about its causal powers. A major impetus behind notions of etiological function is to be able to account for malfunction: “Function categories are essentially categories of things that need not fulfill their functions in order to have them” (Millikan 1993: 23). A trait may regularly fail to perform its function, but it is nevertheless said to have the function. Millikan points out, for example, that mating displays and predatory behaviors in animals “characteristically fail numerous times for every success” (ibid.: 175). A trait might only rarely perform its etiological function: “Consider, for example, the ability human babies are rumored to have of instant hibernation when submerged in very cold water. Surely it is a rare baby whose life has been saved by this capacity. But there have been enough, apparently, to . . . confer a biological function on this disposition” (ibid.: 172). Millikan herself is quite clear that attributing proper function to a trait tells us nothing about its current capacities or dispositions: “To describe the biological function of an item is not to describe its dispositional capacities” (ibid.: 171). Since having an etiological function will not ensure that a state has any particular causal dispositions at all, functional states or objects as conceived by direct teleofunctionalism would certainly not be causally potent in the strong sense. A good deal of effort has gone to show that direct teleological states can have causal effects in the
168 Chapter 5
weak sense (see, e.g., Dretske 1988). But this is small comfort given that direct teleological functionalism is not biologically abstract, so will fail to meet the full demands of Generic Functionalism. One reason that etiological function does not secure causal efficacy is that etiological functions are not realized synchronically. To have an etiological function is to be a thing of a kind that has a certain evolutionary history. That is, the realization relation for etiological functions is having an evolutionary history. And having an evolutionary history is not something that is done synchronically. Of course, if something has an etiological function then it has a certain evolutionary history, and it has that history for the duration of its having the function. But having an evolutionary history—or more properly, being of a kind that has a certain evolutionary history—is not a synchronic relation between things during a period of time. Having an evolutionary history requires standing in certain diachronic relations, relations to things and events in the past, including things and events that occurred prior to the existence of the physical state in question, and sometimes prior to the existence of the kind in question. Realization for etiological functions is not a synchronic relation. There is some temptation to assimilate etiological function into causal-role function, which is synchronically realizable (Griffiths 1993; Davies 2000).24 One way to do this is to treat evolution by natural selection as the explanatory account according to which causal-role functions are ascribed. This instinct to combine etiological and causal notions of function lies behind the teleological accounts of Lycan, Dretske, Flanagan, and Van Gulick, for example. I will formulate a more sophisticated account on behalf of those philosophers in the next section. If we merely use evolutionary theory as an explanatory account, we are able to assimilate etiological function to causal-role function only by changing the objects that are taken to have the functions: The sort of object that has a causalrole function vis-à-vis evolutionary explanation is an object that is extended in evolutionary time and space.25 In the case at hand, the object is probably a kind of trait. This way of combining causal and etiological functions fails to allow particular physical or biological states to realize particular mental states synchronically.
169 Functional Realizations
Etiological functionalism is relatively objective.26 But it fails to secure the causal efficacy of mental states in part because the realization relation for etiological functions is not reasonably synchronous. And, like all teleological accounts, it is not biologically abstract. Other notions of teleological function fare somewhat better in terms of synchrony and causal power; the most promising is John Bigelow and Robert Pargetter’s propensity account. Bigelow and Pargetter (1987) find the etiological account of functions lacking precisely because it fails to secure causal and explanatory power for functions.27 They hold that to ascribe a function to something is to make a claim about how it will behave in the future, about its propensities. Bigelow and Pargetter’s propensity account of function is: “Something has a (biological) function just when it confers a survivalenhancing propensity on a creature that possesses it” (1987, in Allen, Bekoff, and Lauder 1998: 252). On the propensity account, if a thing has a function it has certain properties and dispositions. So the propensity account is able to satisfy the requirement for synchronic realization. And it may be even better situated than causal-role accounts in terms of causal efficacy, for it identifies having a function with having a characteristic that “confers” a propensity—probably a causal property. Some teleological functionalists (e.g., Lycan 1996) have suggested that the propensity account of functions might be more suitable than the etiological account. Unfortunately for teleological functionalists, the propensity account is embattled in its own right. Millikan announces, for example, “I will argue that the propensity theory is, in fact, entirely empty, and hence clearly useless for anyone’s purposes” (1993: 32). Millikan argues that the notion of a “a survivalenhancing propensity” as such is entirely without substance: “There is really no sense at all to the question of how much, if at all, your monogamy, your nose, or your eyes ‘enhance’ your fitness, without first an answer to the question, Enhances it over what?” (ibid.: 40). Bigelow and Pargetter seem to think that they can avoid this problem by relativizing fitness and function to a normal environment, a creature’s “natural habitat” (1987, in Allen, Bekoff, and Lauder 1998: 253). Millikan is suspicious of this sort of too-easy ceteris paribus move: “Are normal conditions for a mouse, perhaps, just conditions
170 Chapter 5
that mice, on average, are in? Then if we tossed all mice but Amos into outer space, our listing of Amos’s “dispositions under normal conditions” would have to change, the main one being left to him being, I suppose, to explode” (1993: 27). Of course, Bigelow and Pargetter acknowledge that a creature’s natural habitat may change, in which case functions would also change. But this means, for propensity-based direct teleofunctionalism, that whether a creature may be said to have mental states at all is subject to the stability of environmental circumstances such that certain states remain fitness enhancing. And that is more than a bit counterintuitive. What seems not quite right is the possibility that an organism, without any change to itself, could come to lose its mental states. But if direct teleofunctionalism is understood in terms of Bigelow and Pargetter’s propensity functions, this very change could occur (retroactively?) if a current state later turns out not to be adaptive. Claims about whether something now has mental states are, on this view, conjectures about the future adaptiveness of those states. If aliens arrived we could only conjecture that they have minds; the final verdict could not be reached until we see if they have states that are adaptive in their new Earthly environment. This is very odd indeed. In contrast, the etiological function of a trait does not change when the environment changes. But the conundrums it produces when applied to questions about minds are familiar. If two otherwise physically identical creatures have different evolutionary histories then their traits will have different functions. If mental states are identical to etiological functional states, two physically identical creatures may differ with respect to their mental states. And this, too, is quite counterintuitive. It is this aspect of etiological function that produces puzzles about so-called Swampman (see Davidson 1987, and the March 1996 issue of Mind and Language). Millikan herself embraces this consequence with respect to biological traits in general and intentional states in particular (1993); and Dretske (1995) explicitly accepts the result with respect to conscious experience. The need to find a relatively objective notion of function by appealing to biology leads them to give up on some of the traditional features of functionalism as characterized by Generic Functionalism.
171 Functional Realizations
Direct etiological functionalists sacrifice synchronic realization and strong causal efficacy in order to naturalize their functionalist theories. Of the four characteristics of Generic Functionalism, etiological functionalism has only one: objectivity. Etiological functionalism fails to yield states that are causally efficacious in the strong sense, that are synchronically realizable, or that are more abstract than biological states. Finally, both etiological and propensity accounts of function are unsuitable for use in metaphysical functionalism because they do not ascribe functions independently of prior individuation of kinds. Paul Davies points out that accounts of function must give the conditions of identity over time for traits “without employing the very theory of functions being explicated and defended” (1994: 363). The point is not just that etiological and propensity accounts presume the availability of something like causal-role functional ascriptions.28 That might be acceptable if we were interested only in explicating teleological notions of function. But the etiological notion says that the function of a trait is an effect that things of its kind did in the past, and the propensity interpretation says that a function of a trait is an effect that things of its kind will have in the future. Both of these accounts presume that we can individuate kinds independently of teleological functional attribution. This is more than just biologically suspect (see, e.g., Amundson and Lauder 1994). It is fatal if we are trying, as the metaphysical functionalist is, to provide a theory of the identity conditions for things of a kind in the first place. The metaphysical functionalist isn’t committed to the claim that functions are merely important to the identity conditions of mental states. The metaphysical functionalist holds the strong view that functions exhaust the identity conditions for mental state kinds. It seems that direct teleofunctionalism will not do the job. At best it fails to satisfy all the requirements of Generic Functionalism. And it seems that the best case still rests on prior individuation of kinds, so it is not even a variety of metaphysical functionalism. Teleological functions by themselves are not promising; but maybe they don’t have to go it alone.
172 Chapter 5
4 Hybrid Functions I indicated earlier that, despite what they each sometimes seem to say, Lycan, Van Gulick, Flanagan, and Dretske do not advocate direct teleofunctionalism. What they desire is some way of building metaphysical functionalism on a combination of teleological and causal notions of function—a hybrid view. For this reason I distinguished direct teleofunctionalism from the view that these philosophers attribute to themselves, which they sometimes call teleological functionalism.29 Owen Flanagan nicely summarizes the motivation for such a hybrid teleofunctional view: “I favor a form of functionalism that analyzes input-output relations in terms of the processes that mediate and subserve them in the normal biological cases, not in any possible cases whatsoever” (1992: 6). It is causal-role functions (“input-output relations”) as understood in the context of biological functions (“normal biological cases”) that interest hybrid teleofunctionalists. Teleological function brings natural objectivity to the specification of causal-role functions. It is not just any causal relations that constitute mental states, it is causal relations that are, in some sense, biologically normal. Van Gulick writes: There is more than punning similarity in the teleological and nonteleological senses in which informational states (or psychological states in general) may be thought of as functional states. For the interdependence of information possession and goal directedness entails that it will not be possible to spell out the relevant causal behavior determining roles in virtue of which some feature (or state) functions to realize the possession of information, without also characterizing the systems states vis-à-vis teleological functions. . . . Thus with respect to information processing systems the notion of a functional state as specified in terms of its contribution to behavior within a systematic context of interacting components of necessity merges with the notion of a functional state specified in terms of its adaptive value. (1980, in Lycan 1990: 128, n. 16)
Selection, natural or otherwise, is a causal process. If a trait has been selected for or will be selected for, if it has a teleological function, then things that are of its biological kind must have had causal effects in the past. We can surmise that there must be some explanation of
173 Functional Realizations
some system such that it may be said of a biological state that has a teleological function that it now has a causal-role function. In other words, selection requires that the trait not be outside the closed bounds of causation. That isn’t much of a connection between teleological and causal-role functions. Van Gulick is claiming more. One failing of direct teleofunctionalism, recall, is that it does not secure causal powers for functional states and so fails to satisfy Generic Functionalism. Direct teleofunctionalism fails to secure causal powers because a thing can sometimes or frequently fail to perform its teleological function. Teleological functions are not statistically normal—they are not causal dispositions, propensities, or properties. Rather, the teleological function of a thing is what is normal in a design or purposive sense; it is what things of a kind are supposed to do. Hybrid teleofunctionalism attempts to secure causal powers for mental states by identifying these states with functional states that are apt or disposed to do what it is their teleological function to do.30 Lycan says that he takes his biological functions “very seriously and literally: as honest-to-goodness natural teleology”: “I propose to type-identify a mental state with the property of having such-and-such an institutionally characterized state of affairs obtaining in one (or more) of one’s appropriate homunctional departments or subagencies. . . . To be in pain of type T, we might say, is for one’s sub- . . . sub-personal f-er to be in characteristic state ST(f), or for a characteristic activity AT(f) to be going on in one’s f-er” (ibid.: 41). On Lycan’s account, the functional states are identified in terms of biological (e.g., etiological) functions. It is biology that determines that some organ is the f-er rather than the y-er. But to be in a mental state is to be in a state that is characteristic of a f-er—a state that is apt to or disposed to f. So Lycan’s is a hybrid theory. The general idea behind teleofunctionalism is that causal-role functions are circumscribed by teleological functions: Being a mental state is a matter of realizing a causal-role function that is specified with respect to teleological functions. The trouble is that the two notions of function do not readily allow themselves to be linked together. It isn’t that causal and teleological functions are incompatible; rather, they are simply unrelated. As Peter Godfrey-Smith
174 Chapter 5
notes, “the only thing the two concepts of function share is that in both cases functions are explanatorily salient effects of components of systems” (1996: 18). Teleological functions help explain why a trait has come to exist; causal-role functions tell what a trait does or is apt to do. If we know the etiological function of a thing then we know something about its history or future and about the capacities of its ancestors, or some of them at any rate. But this tells us nothing about its current capacities, nor about either the capacities or the history of other things that are physically like it. If I have a defective heart, it nevertheless has the etiological function of circulating blood. Likewise, a preserved heart in a jar of formaldehyde has the teleological function of circulating blood. But neither of these things, nor other things like them—artificial defective or artificial pickled hearts— have the capacities or dispositions to circulate blood. No instance of a trait is guaranteed, even statistically, to produce as its causal-role function the effect that is its etiological function. Hume, famously, claims that we cannot derive ought from is. In like manner, it is unclear that the hybrid teleofunctionalists can derive disposed to from supposed to—that teleological evolutionary normativity can be converted into statistical causal normativity. The challenge to the hybrid teleofunctionalist is to show that the two notions of function can be linked together in a way that is not ad hoc. Let us consider two ways of linking teleological and causal-role functionalism. First: (H) To be a mental state M is to realize a causal-role function F in system S, where F contributes to S’s capacity to R and R is a teleological function of S. (H) captures the idea that a mental state has a causal function that contributes to a system in a teleologically specified way. It seems to be what Lycan (1987) and Flanagan (1992) have in mind. But teleological functions apply only to the systems that earn the full teleological ascription directly, for example by having a certain evolutionary history. The two notions of function have not been linked in (H) except insofar as they both apply to particular cases. What (H) asserts is that mental states in fact have the effects that are biologically normal for their kind. It adds a modest success condition
175 Functional Realizations
to the direct teleofunction, but it is nevertheless necessary that every state in fact have the relevant history. As such, (H) is a quite reasonable version of direct teleofunctionalism. But it is merely direct teleofunctionalism, for it does not escape the requirement that each and every minded system itself have a particular evolutionary history, namely, the one that justifies the etiological functional ascription.31 The trouble is that having the current functional organization and having the evolutionary history seem disconnected. Why don’t other states with the same causal dispositions but without the particular history nevertheless count as mental states? Such states have the same causal-role functions, even if those effects are not telelogically characteristic for them. So evolutionary history looks to be epiphenomenal when it comes to current causal powers. Hybrid teleofunctionalism needs to somehow extend a teleologically grounded ascription to systems that do not themselves directly have the grounding selection history. For example, a second possibility: (H¢) To be a mental state M is to realize a causal-role function F in system S resulting in effect E, where S is physically like kind N that has the teleological function of E-ing. (H¢) begins with the direct teleofunctional account and then claims, more or less, “and this applies to other states physically like these teleological functional states, too.” (Let us set aside the puzzle of how to specify when two systems are physically alike.) The trouble is that teleological function does not license such extension of the functional ascription; it is entirely ad hoc. After determining that a creature has a trait that confers on it a capacity and that the capacity is a teleological function, (H¢) would have us conclude that other morphologically similar traits have the same teleological function. But this is not so. Viceroy butterflies are tasty to birds, and they have similar wing coloration to monarch butterflies, which are poisonous to birds. Viceroy wing patterns are disposed to produce similar effects to those of monarch butterflies, for example, that the butterflies bearing them are not generally eaten by birds. But viceroy and monarch wing patterns do not thereby have the same teleological functions. They might have—do have—entirely different
176 Chapter 5
etiologies. Teleological function distinguishes those things that actually have or will have certain causal histories from similar things that do not or will not.32 The only attempt that I know of to formulate a hybrid view in detail is due to Fred Dretske (1988, 1995), who subscribes to a version of Brentano’s thesis that he calls the representational thesis: “(1) All mental facts are representational facts, and (2) All representational facts are facts about informational functions” (1995: xiii, italics removed). Given this commitment, when Dretske argues for a hybrid functionalist theory of representation, he thereby articulates a version of metaphysical functionalism. Dretske distinguishes teleological function from a kind of causalrole function having to do with conveying information. The information-bearing causal-role functions are relative to an explanation of a system that is doing whatever it is its teleological function to do, “doing its job” (1995: 2). Dretske writes that “a system, S, represents a property, F, if and only if S has the [teleological] function of indicating (providing information about) the F of a certain domain of objects” (ibid.: 2). Thus, “[r]epresentation is here being understood to combine teleological with information-theoretic ideas” (ibid.: 4).33 Dretske formulates his account with respect to problems of representation, and that makes all the difference. He is interested in how mental states represent and misrepresent. They represent when they have the function of indicating; they misrepresent when they fail (in certain ways) to do what it is their function to do. But that means that being a mental state—being a representational state, on the representational thesis—falls to having a teleological function. True, it is the teleological function of indicating something or other, for example, that I should not eat that butterfly or that it is a martini that the woman in the corner is drinking. But the representing—the proper informational function—is entirely dependent on teleological function. A system without that teleological function is not even a candidate for representing. So Dretske’s representational account, taken as a version of metaphysical functionalism, is a version of (H); it turns out to be a version of direct teleofunctionalism after all. Each version of hybrid teleofunctionalism, (H) and (H¢), invariably collapses into causal-role or direct teleofunctionalism. If we limit
177 Functional Realizations
the instances to those things that have certain causal histories as in (H), then we are back to giving up biological generality and causal efficacy. We end up with direct teleofunctionalism, and the causalrole functions are doing no metaphysical work. On the other hand, if we extend the possible realizers to any systems causally like a system with certain teleological functions, as in (H¢), then appeal to teleological function is ad hoc. Teleological functional kinds are not always or often causal-role functional kinds. If we extract the causal roles and designate them as the ones we are interested in, it cannot be because the teleology compelled us to do so. Teleological functions are had only by systems that have certain histories, not systems that have certain histories and others physically like them. Then the teleology is doing no metaphysical work. The hybrid teleofunctionalist needs some natural way of using teleological function to select causal-role functional kinds, a way that is not ad hoc or post hoc. No good candidates present themselves. 5 Prospects for Generic Functionalism The history of functionalism in philosophy of mind is woven of many projects, with many goals, using many notions of function. In the last three chapters I have tried to be reasonably clear about how the miscellaneous functionalisms work, where and when they intersect, and where and when they do not. I am not merely performing an autopsy on functionalism; I am also sketching part of its biography. But to recount the entire history of functionalism from beginning to end is more than I can here undertake. I have dedicated my attention to one strand, metaphysical functionalism. I distinguished a number of ways that metaphysical functionalism could be developed. The variations considered are the most familiar ones, and they are also milestones in the evolution of functionalism as a distinctive theory of mind. The story that I am telling is one according to which functionalists took themselves, at each step, to be refining a single core theory. Putnam himself recalls, “When it became clear that the formal properties of [computational] states are quite unlike the formal properties of psychological states, the original idea of functionalism quickly was replaced by an appeal
178 Chapter 5
to the notion of an ideal ‘psychological theory.’ But this ideal psychological theory was conceived of as having just the properties that formalisms for computation theory possess” (1999: 34). Putnam describes the orthodox functionalist project just as I characterized it earlier: as trying to find the correct functional specification (psychological theory, computational theory, etc.) while assuming that the core relations (the “formalisms,” like realization) could be taken for granted. We have now seen that this is a mistake.34 The versions of metaphysical functionalism are not variations on a theme, unified by a common metaphysical thread. This is obvious once we look carefully at the realization relation and see that there is no common metaphysical relation shared throughout the functionalist lineage. We find instead progressions: from abstractness to naturalness and objectivity, from synchrony to etiology. With the realization of abstract functional states we get great generality without causal power or objectivity. Causal-role functionalism pursues the possibility of causal efficacy by giving up some degree of generality, and it still does not ensure objectivity. Direct teleological functionalism secures objectivity at the cost of abstractness and causal power. The hybrid teleological functionalist only wants what functionalists have wanted all along: an objective theory of the metaphysics of mind that occupies a moderate position between dualism and identity theory. But there is growing evidence that the middle ground is unstable. Hybrid teleological functionalism tries to have it all, but the difficulty of jointly satisfying the desiderata of Generic Functionalism only recapitulates itself in the problem of how to formulate the hybrid view. There are four options for devotees of metaphysical functionalism. One is to try to work out a version of hybrid teleological functionalism. The second is look for some nonteleological way of naturalizing causal-role functions. The third is to hope for an altogether novel notion of function with a robust realization relation. Finally, one might abandon Generic Functionalism and the generality argument, seeking another path. For my part, if I had to be a metaphysical functionalist I would be a hybrid-teleological functionalist. I would look hard for a way to make it work. But I would be prepared to retreat to directteleological functionalism; and, having abandoned the generality
179 Functional Realizations
argument, I would need to develop an alternative argument against identity theory. Many philosophers evidently regard their options from this standpoint. They believe that they have to be functionalists of some sort, and they are trying to figure out which kind of functionalism is least unpalatable. But we do not have to believe some version of functionalism. And if we do not have to be functionalists, then there is no reason to run hither and yon looking for hybrid teleofunctions. We might, instead, contemplate an identity theory. We can consider the alternatives unless, that is, we have a nearfanatical prior commitment to metaphysical functionalism. Consider Lycan’s declaration: “Functionalism is the only positive doctrine in all of philosophy that I am prepared (if not licensed) to kill for. And I see the ‘role’/‘occupant’ distinction (some say obsessively) as fundamental to metaphysics” (1987: 37). Such a commitment outstrips Putnam’s claim that functionalism is more plausible than identity theory. It must derive from some other argument in favor of functionalism. Let us now consider what such arguments might be.
6 Functional Properties and Mechanistic Explanations
I am trying to make room for an identity theory of conscious mental states. I don’t claim to have established that an identity theory must be correct, but I have argued that it is a considerably more appealing option than is generally recognized. I began by arguing that the theory can accommodate the sorts of considerations raised by Putnam and Kripke. This is important for it shows that the theory is not already dead in the water, as is often supposed. Then I set out to raise some questions about the viability of the main competing theory, functionalism. Functionalism is relevant because it seems to be the received view of mind among philosophers and cognitive scientists. My concerns about functionalism are not the familiar ones: I do not complain that functionalism fails to properly handle this or that science-fiction example or thought experiment. Instead I argue that the functional realization relation cannot do the work that a functionalist theory demands of it. In the preceding chapters I advanced a series of arguments for the conclusion that there are no available versions of metaphysical functionalism that live up to all functionalists’ advertisements, and that there is little reason for optimism. It is clear what a satisfactory version of metaphysical functionalism would look like. But it is not at all likely that a metaphysically functionalist theory can be formulated without stretching credibility. By my measures the identity theory looks very good indeed. Although I have cast doubt on whether a workable version of metaphysical functionalism will be forthcoming, the considerations of the
182 Chapter 6
previous chapters do not go so far as to rule out the view. It is open to the defenders of functionalism to provide reasons for believing that, however unlikely, some workable version of metaphysical functionalism must be available. Two such strategies come to mind. First, Sydney Shoemaker (1981b) and William Lycan (1987) argue that functionalism must be the correct theory of mind because nature is “functions all the way down.” Shoemaker considers an argument to the effect that metaphysical functionalism is entailed by a familiar thesis about properties in general, while Lycan urges a related line of reasoning based on the ubiquity of teleofunctional kinds. A second approach, due to Jerry Fodor (1974) and Putnam (1975b), argues that we are committed to functional explanations in general (and in psychology specifically) and that as a consequence we are committed to metaphysical functionalism. To these arguments I now turn. In each case I argue that the considerations do not entail the truth of metaphysical functionalism. In the end I suggest that, rather than supporting functionalism, the two lines of reasoning support a broadly mechanistic view of mind. This will take some explaining, but understanding mechanism will allow us to recognize that the identity theory has some of the merits that are too often thought of as unique to functionalism. 1 Functions, Functions Everywhere If all states and properties are functional states and properties, that is prima facie reason for thinking that some version of metaphysical functionalism is correct. Suppose, following Shoemaker, that we have a Ramsified theory of all properties: If we could specify all of the causal features of a property in a set of propositions of finite length, then using that set of propositions as our “theory” we could use the Ramsey–Lewis technique to construct a functional predicate which is true of a thing, in all worlds having the same causal laws as the actual world, just in case it has that property. The functional property expressed by that predicate could be called the “functional correlate” (or the “actual world functional correlate”) of that property. (1981b, in 1984: 274)
Many accounts of properties maintain that in the actual world properties are individuated by their causal relations: “it is necessary and
183 Functional Properties and Mechanistic Explanations
sufficient for the identity of properties A and B in the actual world that A and B share all of the same causal features” (ibid.: 275). If we also take it that properties are individuated across possible worlds by their causal features, then we have arrived at what Shoemaker calls the causal theory of properties (CTP). If, following Shoemaker’s suggestion, we translate the causal theory into a Ramsified functional theory of properties, every property will be identified with its Ramsey “functional” correlate. Since the theory contains only causal relations, the resulting theory will be functional in the strong sense. Applied to mental properties, the causal theory of properties says that mental properties are functional properties in the strong sense—it claims that mental properties are metaphysically functional properties. Shoemaker calls this view minimal functionalism or CTPfunctionalism: “the view that (a) every mental property is identical to its functional correlate, and (b) the functional correlate of every mental property is an SS-[strong sense-]functional property. Since to hold this is to hold that CTP holds for mental properties, I shall refer to it as CTP-functionalism” (ibid.: 276). Shoemaker believes that CTP-functionalism is entailed by the causal theory of properties, but he concedes that he does not have a decisive demonstration of the entailment. If the causal theory of properties is false or does not entail CTPfunctionalism, then Shoemaker’s line of reasoning is a nonstarter. It will not compel us to accept some version of metaphysical functionalism in philosophy of mind. But Shoemaker’s lingering doubts stem instead from the possibility of nonphysical properties and realizers (ibid.: 275–278), a possibility that the naturalist sets aside. For the moment let us suppose with Shoemaker that the causal theory of properties is reasonable and that it entails CTP-functionalism. If so, is CTP-functionalism enough for functionalists? The answer is not clear. It remains to be seen whether CTP-functionalism can be used to support any of the candidates for functionalism as a distinctive theory of the mind. We can begin by asking whether CTP-functionalism entails Generic Functionalism. It is clear that it does not. Shoemaker presumes that the notion of function at stake is a causal notion of function. If we understand these causal functions in terms of Cummins’s (1975) account, then any version of functionalism entailed by
184 Chapter 6
CTP-functionalism will have the standard shortcomings of causal-role functionalism: It will not specify mental states or properties in an objective way, it will not guarantee biological abstractness, and it will be subject to the causal exclusion argument. Shoemaker is not tied to Cummins’s account, however. CTP-functionalism helps itself to an extremely prolific notion of causal function: Every causal relation is a functional relation. (“If we could specify all of the causal features of a property . . . [t]he functional property expressed by that predicate could be called the “functional correlate” . . . of that property.”)1 Suddenly, functions are everywhere! This is part of the appeal of Shoemaker’s view. It seems that if all properties are functional properties then this ought to at least lend some support to the claims of Generic Functionalism. But it does not. For example, that all properties are functional properties does nothing to support the contention that mental states and properties are more abstract than biological states and properties. Of course that is a problem only if Shoemaker’s variety of functionalism needs to satisfy Generic Functionalism. Generic Functionalism is designed to support the generality argument based on multiple realizability. But if CTP-functionalism is the basis for a distinct argument for functionalism that does not depend on the generality argument from multiple realizability, then it would not be required to satisfy Generic Functionalism. If Shoemaker can show on general grounds that mental states must be functional states, then he does not need to appeal to the multiple realizability argument and so can eschew Generic Functionalism. In that case, lack of biological abstractness will be irrelevant. What can be said about this sort of minimal or CTP-functionalism? Remember that we have been operating on the assumption that the causal theory of properties is correct, and that mental states are functional states in the strong sense. CTP-functionalism’s reliance on the causal theory of properties may cut against metaphysical functionalism in three ways. The obvious problem is that the causal theory might be wrong, in which case Shoemaker’s argument never gets off the ground. I’ll have nothing to add concerning this worry. But, second, if the causal theory is correct, then it seems to have the surprising consequence that a substantial thesis about the mind is
185 Functional Properties and Mechanistic Explanations
entailed by a general metaphysical theory. Third, though the causal theory alleviates the need for an objective way of specifying functional properties, a similar problem immediately reappears because we still need a way of specifying which of the functional properties are mental properties. Let’s begin with the second worry. Putnam, among others, seems to think that the question of functionalism is separable from the details of a theory of properties. After all, he takes functionalism to be an empirical hypothesis (Putnam 1967). But if Shoemaker is right, CTP-functionalism about the mind is not a substantial or empirical thesis at all, but is entailed by the conjunction of the causal theory of properties and the truth of naturalism (materialism). That result is surprising. But that it is unexpected should not itself be an objection. The problem, rather, is that CTP-functionalism fails to discriminate between functionalism and the identity theory. After all, on Shoemaker’s view, brain properties are CTP-functional properties just like all the others. Noting this feature of CTP-functionalism, some functionalists might simply declare success. But the identity theorist might as easily claim the victory: After all, CTP-functionalism could be true if mental states and properties are type-identical to physical states and properties, namely, those of brains. What should we make of this standoff? Some functionalists are suspicious of this route to victory. Again Van Gulick’s worry is apt: “If physiological properties and other physical properties of many sorts can all be interpreted as functional properties, then the functional nature of psychological states cannot be taken as evidence against the identity thesis. Moreover, the very thesis of functionalism itself is in danger of losing interest, for its appeal lay in picking out some supposedly distinctive characteristic of psychological properties” (1982: 190). The nagging concern is that CTP-functionalism somehow fails to capture the distinctive claims of metaphysical functionalism about the mind after all. Later in this chapter we will return to this worry, but other questions remain. CTP-functionalism also faces the challenge of saying which, among the myriad functional properties picked out by the causal theory of properties, are the mental properties. The problem faced by the more common forms of causal-role functionalism, based on
186 Chapter 6
Cummins’s account, was to say which among the causal relations are the definitive functional properties of mental states. If all causal properties are functional properties, the question is merely transformed. Declaring that the causal properties are themselves functional does nothing to answer the question of which are definitive of mental properties. Shoemaker’s causal functions are not explicitly relativized to theories or explanations as are Cummins’s causal-role functions, but they nevertheless demand a kind of theoretical choice. The issue is analogous to one with which we are familiar from earlier consideration of varieties of functionalism and the realization of causal-role functions: Which theory of mental properties goes into the giant theory of all properties that yields the causal theory of properties? CTP-functionalism has no resources of its own for solving this problem; it must rely on some other account, such as one based on a psychological theory. But which? Will it be folk or analytic psychology, empirical psychology, or some sort of “physiological” (Shoemaker 1981b) or neuroscientific theory? And what considerations should we use to make the choice? Shoemaker himself is primarily concerned with whether CTPfunctionalism, if true, is compatible with or entailed by his analytic functionalism and Putnam’s psychofunctionalism (Shoemaker 1981b, sections VI–VII, passim). Shoemaker argues that empirical psychofunctionalism is superior to the neuroscientific variant, and that analytic functionalism is still preferable to either, on the grounds that the others are more parochial or chauvinistic (ibid.: 280). In short, Shoemaker retreats to the generality argument after all. If this is correct, then Shoemaker’s argument is not independent of the generality argument from multiple realizability and therefore must sustain a variety of functionalism that satisfies Generic Functionalism. 2 Functions All the Way Down Despite my apprehensions, it surely seems like it ought to be a good thing for functionalism about the mind if all properties are functional properties. Perhaps there is a way of cashing out the basic idea without depending on the causal theory of properties. William Lycan suggests an argument that aims to support functionalism by showing
187 Functional Properties and Mechanistic Explanations
that nature is “functions all the way down” (1987: 48), or at any rate “as far down as could possibly be relevant to psychology (well below neuroanatomy, for example)” (ibid.: 45, italics removed). Lycan argues that functions, specifically teleological functions, come in degrees. Everything, or most everything, is functional (teleofunctional) “to some however small degree” (ibid.).2 Since pretty much everything in nature is teleofunctional to some degree, Lycan argues, the distinction between function and structure itself “fails to get a grip on human psychology where it lives” (ibid.). According to Lycan, standard objections to functionalism generate their intuitive pull by acting as though there were a single functional level of description—and a single structural level of description—a dogma that he calls “Two-Levelism” (ibid.: 57). For example, he argues that it is Two-Levelism that makes Block’s problem of inputs and outputs, of avoiding chauvinism and liberalism, seem intractable. But Lycan rightly observes that it is a mistake to think in terms of the structural level and the functional level. Once we see that the functionalist can theorize at many levels, that one need not choose a single functional level of description, the threat of chauvinism and liberalism is mitigated. There will be many teleofunctional levels of organization to which the functionalist can appeal. Moreover, Lycan argues that once we give up Two-Levelism’s rigid view of the structure/function distinction, the very distinction between functionalism and identity theory itself gives way: [I]f we also accept my claim that Homunctional characterizations and physiological characterizations of states of persons reflect merely different levels of abstraction within a surrounding functional hierarchy or continuum, then we can no longer distinguish the Functionalist from the Identity Theorist in any absolute way. “Neuron,” for example, may be understood either as a physiological term (denoting a kind of human cell) or as a (teleo-)functional term (denoting a relayer of electrical charge); on either construal it stands as an instantiable—if you like, for a role being played by a group of more fundamental objects. Thus, even the Identity Theorist is a Functionalist— one who locates mental entities at a very low level of abstraction. (1987: 58–59)
There are two suggestions in this passage. The second is that identity theory is committed to Two-Levelism, to an absolute
188 Chapter 6
structure/function distinction, and that it locates mental states at a single low (or lowest) level of abstraction. But there is no reason to so restrict identity theory. An identity theorist need not theorize at a single neurobiological level. Identities may occur at many levels. The difference between identity theory and functionalism can be the difference between structure-relative-to-level and function-relativeto-level. Just because the distinction between structure and function goes relative does not mean that it goes away. The fact that the structure/function distinction fails to get a grip in any one place in the hierarchy of nature is no more a problem for identity theory than it is for functionalism. My guess is that Lycan thinks of identity theory as committed to Two-Levelism because he does not make Kim’s (1998) distinction between higher levels and higher orders, discussed in the previous chapter. Lycan (1981, 1987) is a proponent of the view that orders generate hierarchical levels—his is just the sort of view that Kim is challenging. Using Kim’s distinction we can see the source of the confusion: An identity theory works at a low order of abstraction, but it is not thereby constrained to operate at lower levels (fine grains) of organization. Because Lycan presumes that levels and orders go together, he concludes that the identity theorist must operate at the lowest levels. But we now see that this is a mistake. Just as there is no single “functional level” there is no single “neuroscientific level.” The identity theorist is free to operate at many levels. This leads us back to Lycan’s first proposal from the passage above. He makes the tantalizing suggestion that the identity theory is itself a version of functionalism. According to Lycan, the idea that they are different is a side-effect of Two-Levelism: “There is an idea, brought on by blind Two-Levelism, that Functionalism differs somehow conceptually or structurally from the Identity Theory. . . . As I have said, the Identity Theory is just an empirically special case of Functionalism, one that (implausibly) locates all mental states at the very same low level of institutional abstraction—the neuroanatomical” (1987: 59). Notice the relation of this claim to the point already discussed. If the difference between functionalism and identity theories is not structural or conceptual then it must be a difference in the levels at which they theorize. If Lycan is correct that there is no structural or
189 Functional Properties and Mechanistic Explanations
conceptual difference between functionalism and the identity theory, then once we give up Two-Levelism the remaining conclusion is that there is simply no difference at all between identity theory and functionalism. For reasons already mentioned with respect to Shoemaker’s argument, it is not obvious that it is a good thing for functionalism if it turns out that identity theory is a version of functionalism, or is indistinguishable from functionalism. We could just as well adopt the conclusion contrary to Lycan’s, namely, that functionalism is a version of the identity theory.3 But that is not my view. In chapter 3 I argued that there is indeed a structural or conceptual difference between the two kinds of theories. Although the identity theorist and the functionalist generally identify mental states and properties at the same levels of organization, they differ about whether mental properties are essentially functional. The metaphysical functionalist maintains that mental states are functional in Van Gulick and Shoemaker’s strong sense—and this is exactly what the identity theorist denies. The difference between identity theory and functionalism is conceptual or structural after all. I have been evaluating Lycan’s proposals to see if they compel us to adopt metaphysical functionalism. I am suggesting that they do not. And really this should be no surprise. As I understand the debate, Lycan intends to defend his functionalism from sciencefiction counterexamples of the familiar sort, absent and inverted qualia, and so forth (Block 1978). His claim is that functionalism is no worse off than identity theory with respect to these science-fiction cases and thought experiments. That is all his arguments are supposed to show; they are not supposed to defeat identity theory. But if the theories are both subject to the same objections, what is the reason for preferring functionalism? The reason is the standard one: The theories are otherwise comparable but identity theory is chauvinist, or at least more so than functionalism. That is, Lycan depends on the generality argument based on multiple realizability. This is doubly worrisome. First, if he depends on the generality argument then (like Shoemaker) he does not have an independent argument for functionalism based on the thesis that the world is functions all the way down. Second, the particular variety of functionalism that
190 Chapter 6
Lycan prefers is not well supported by the multiple realizability argument; it sacrifices biological abstractness by tying itself to biological functions.4 That is why it seemed in the previous chapter that the teleofunctionalist, in particular, would need an independent defense for his variety of functionalism. Some argument might give us reason to suppose that Lycan’s hybrid teleological functionalism must be true, but generality cannot be that reason. If all properties are hybrid teleofunctional properties, then mental states are hybrid teleofunctional properties to be sure. But I do not think we have any reason to suppose that all properties are hybrid teleofunctional properties. Even Lycan, in his “lucid” moments, is prepared to concede that this may be too much “panteleology” (1987: 45) to be credible. And the functionalist who wants to make the argument that nature is rife with hybrid teleofunctional kinds will still have to show how hybrid teleofunctions can work. He or she must contend with the problems of integrating teleological and causal notions of function, or else introduce an entirely novel notion of function. The admittedly paradoxical conclusion that I have reached is that even if nature is (in some sense) functions all the way down this does not show that the claims of metaphysical functionalism can be satisfied. This is because metaphysical functionalism about the mind claims more than that mental states are in some sense functional; it claims that they are functional relative to physiological or brain states. And that result is not assured merely because functional states or properties are abundant. The fact, if it is a fact, that functional properties are ubiquitous does not by itself provide reason for thinking that some satisfactory version of metaphysical functionalism is forthcoming. 3 Functionalism and the Autonomy of Psychology We have been considering whether there are arguments that show, despite questions raised about its tenability, that some version of metaphysical functionalism must nevertheless be correct. The “functions all the way down” arguments came up short. They couldn’t assure that mental states were functional with respect to biological
191 Functional Properties and Mechanistic Explanations
states. Correspondingly, they turned out not to be independent of the multiple realizability argument after all; so they are not suitable answers to our earlier concerns. There is an argument for functionalism about the mind that is compatible with the generality argument based on multiple realizability. I suspect that this argument lies behind the temptation to think that metaphysical functionalism about the mind is guaranteed if the world is functions all the way down. The argument takes as its premises the legitimacy of psychology as a science, along with some principles concerning explanation and ontology; it yields the conclusion that functionalism must be correct because the alternatives fail to adequately ground psychological explanation. The central idea in the argument is that psychological explanations, and the mental states and properties that they pick out, are autonomous. Putnam seems to be responsible for introducing the idea that the autonomy of the mental is what matters. He argued that it was a mistake to care about what stuff—“copper, cheese, or soul”— minds are made of; what we care about is the “autonomy” of minds (1975b). What Putnam dubs the “autonomy of our mental life” is simply the legitimacy of mentalistic psychological explanation, and with it the legitimacy of the mental states and properties picked out by such explanations. Similarly, Fodor (1974) argued that being materialists does not require us to be identity theorists and that as a consequence psychology can go about its business without a worry as to whether psychological kinds correspond to physiological kinds. This is fortuitous since Fodor, like Putnam, believes on the basis of multiple realizability concerns that identity theory is a nonstarter. They each criticize the identity theorist for making a mistake about what is important in philosophy of mind. What is important is not the stuff but the legitimacy or “autonomy” of psychological explanation. According to advocates of autonomy, “a property is real (or autonomous) just in case it is essentially invoked in the characterization of a regularity” (Antony and Levine 1997: 91).5 Call this the autonomy thesis. By extension, an explanation or theory is autonomous only if it characterizes a unique regularity, that is, a regularity that cannot be captured in other terms. Since we don’t want
192 Chapter 6
our explanations to traffic in objects that are not real, we need our explanations to be autonomous. (Likewise, since we don’t want to commit ourselves to objects and properties that aren’t doing any work, we accept only those that we need for our explanations and reject those that we can make do without.) Thus to show the reality of mental states and properties one has to show that explanation in psychology is not only useful but that it is compulsory. If we are confident in the reality of mental states as described by psychology then we can conclude that psychological explanation is autonomous. And if psychology is autonomous then we can use that fact to argue for functionalism. It is often supposed to be, and sometimes claimed to be, a consequence of identity theory that we could in principle explain everything about minds without using distinctly psychological language. If so then we could dispense with psychological theories and ipso facto with the bothersome psychological states, events, and properties that they traffic. In that case, it might seem that we have no basis for accepting the reality of psychological states, events, and properties. On this way of thinking, the identity theory entails, or at least invites, a brand of eliminativism about the mental. So if we are to give a constructive account of psychological states and properties it must not be an identity theory; it will have to be some account that provides the basis for autonomous psychological explanations. Functionalism claims to fit the bill. If psychological kinds are functional kinds then there is hope for the autonomy of the mental. Because functionalist psychology construes mental states as biologically abstract (multiply realizable) it purports to characterize regularities that cannot be captured by explanations in neuroscientific terms. Functionalism would thus salvage the autonomy of psychology and the reality of mental states and properties. If functionalism and identity theory exhaust the naturalistic alternatives then functionalism must be correct, for the identity theory is self-defeating. Ironically, by failing to secure the autonomy of psychology, it undermines the reality of that which ostensibly is its subject matter. Therefore the need for autonomous psychological explanation at least invites (and may require) metaphysical functionalism about the mind. This is the autonomy argument for functionalism.
193 Functional Properties and Mechanistic Explanations
As a positive argument for functionalism the autonomy argument has some obvious gaps. First, it must assume at least one of two controversial premises. It could assume that functionalism and identity theory exhaust the options, so that the failure of identity theory leaves only functionalism. Alternatively, the autonomy argument could assume that only functional explanations are autonomous of fundamental physics, so that the requirement for autonomy directly entails functionalism. With respect to the first, the dichotomy is rhetorically convenient, but as a premise it is too strong. I have been arguing that the failings of functionalism leave more room for identity theories than has been supposed; but that is clearly insufficient to prove the identity theory correct, and I have not claimed to do so. Identity theory and functionalism are the main naturalistic contenders, but the field is not so empty that the defeat of one would ensure victory to the other. The alternative premise is hotly debated under the guise of the exclusion argument, which contends that functional properties are not autonomous after all. So neither assumption is on strong footing. Second, the autonomy argument is not independent of the multiple realizability argument; if functionalist psychology is autonomous then this is so in virtue of its biological abstractness.6 This objection returns us to the familiar debate over the empirical claim of multiple realizability. If mental states are not multiply realizable then the autonomy argument is in no position to supplement the generality argument; but if mental states are multiply realizable then the generality argument may be sufficient on its own. These concerns about the autonomy argument cannot be brushed aside. But there is a more basic difficulty with the autonomy argument that I have not yet considered. So far I have stated the autonomy argument and objections to it without using the customary language of “reduction.” Although in philosophy of mind, “reductionism” and “identity theory” tend to be used almost interchangeably and to be contrasted with “functionalism,” it is a great mistake to conflate them. This is why I insisted, in the in troduction, that I am advocating identity theory but not “reductionism.” “Reductionism” is first and foremost an explanatory or theoretical thesis, namely, the thesis that psychology is not autonomous.7 The
194 Chapter 6
thesis of autonomy is sometimes formulated explicitly as the rejection of “reduction,” as when Fodor offers: “Simply to have a convenient way of talking, I will say that a law or theory that figures in bona fide empirical explanations, but that is not reducible to a law or theory of physics, is ipso facto autonomous; and that the states whose behavior such laws or theories specify are functional states. (In fact, I don’t know whether autonomous states are ipso facto functional. For present purposes all that matters is whether functional states are ipso facto autonomous)” (1997: 149). Although Fodor aims to draw an ontological conclusion, autonomy itself is thought of first in explanatory terms. In contrast, the identity theory plainly makes a metaphysical claim. The merits of the identity theory do not depend on any claim about what other explanatory resources are or are not available. In particular, it can be accepted or rejected independently of the autonomy of psychological explanation. It is true that some identity theorists have argued for the thesis by arguing that psychology is “reducible” to neuroscience and therefore that psychological kinds should be identified with neuroscientific kinds. In my view this is a mistake that has only contributed to confusion and hindered understanding the identity theory itself. At the time that “identity theory” and “reduction” came into use in philosophy of mind, it was thought that theoretical “reduction” and metaphysical identification were doctrines that necessarily go together. That is, it was thought that the “reducibility” of psychology to neuroscience entails the identity theory; and that the truth of the identity theory would ensure the “reducibility” of psychology to neuroscience. Correspondingly, it was thought that multiple realizability was an obstacle to the “reduction” of psychology to neuroscience, and ipso facto to the identification of mental states with brain states. (On this view, multiple realizability ensures the autonomy of psychology by blocking “reduction.”) If this were correct, then the identity theory would stand or fall with the question of whether psychology is “reducible” to neuroscience. So it has seemed to many identity theorists who have argued that psychology is “reducible,” and to many functionalists who have argued that psychology is autonomous. But if I am correct, the identity theory does not depend on the “reducibility” of psychological explanation.
195 Functional Properties and Mechanistic Explanations
The reader has no doubt noticed that I persist in using the annoying device of scare-quotes when speaking of “reduction.” One reason is that the very notion has itself come in for a good deal of scrutiny of late, leaving serious doubt as to whether the term is being used univocally. Some persist in using “reduction” as though it were unproblematically ontological and equivalent to identity, even while others have been sensitive to the development and debates concerning theoretical and explanatory “reduction” in the sciences. For example, it is clear that debates in philosophy of mind concerning whether or not psychology can be “reduced” to neuroscience are typically cast in terms of a notion of “reduction” (due to Nagel 1961) that is widely dismissed in biology and philosophy of biology, not least of which because that notion is based on a model of explanation (due to Hempel and Oppenheim 1948) that has been long since abandoned.8 This has inspired some to call for the revision of the archaic conception of theoretical “reduction” (Churchland 1985; Bickle 1996, 1998, 2003). Bickle, for example, argues that, contrary to Putnam, Fodor, and many others, psychology is “reducible” to neuroscience, and allegations of multiple realizability are no obstacle to that claim. Yet “new wave reductionists,” including Bickle, tend to be reluctant to assert that their “reductionism” entails the identity theory or any metaphysical position at all.9 Rather than thinking of this as bad news for the theory, I see it as yet another hint that the question of “reduction” is the wrong issue on which to rest any ontological conclusion. Note that the autonomy thesis as presented by Fodor is insensitive to the different kinds of “reduction.” One might think of this as a defect or ambiguity in the thesis; certainly most authors concern themselves mainly with classical Nagelian “reduction,” and we might perhaps rebuke them for so doing. But this is not quite right. Instead, it seems that the complementary requirement of autonomy (viz., that the reality of psychological kinds depends on their being essential to some explanation), explicit in the formulation by Antony and Levine, is best understood in contrast to any and all kinds of “reduction.” If that is right then any sort of “reduction” of psychology would threaten the autonomy argument by denying the premise that psychology is in fact autonomous. So if, contrary to my suggestion, the
196 Chapter 6
relative merits of identity theory and functionalism depend on whether psychology is, in some sense, “reducible” to (or autonomous from) neuroscience, then obviously it will be important to properly understand “reduction” (and autonomy). In that case it will also be extremely important to examine the details of psychological and neuroscientific theories and explanations to find out whether the former are or are not “reducible” to the latter according to whatever model of “reduction” is taken to be relevant.10 But is that the proper approach? The question is whether the dispute between identity theory and functionalism is best understood as a disagreement about the autonomy of psychological explanation. I think that it is not. The identity theory need not require the “reduction” or even the “reducibility” of psychology. In the next two sections I describe two approaches to explanation that are compatible with identity theory and with psychological explanation, but that do not demand that explanations be autonomous. 4 Mechanism and Mechanistic Explanation Peter Machamer, Lindley Darden, and Carl Craver (2000) have recently defended a model of multilevel mechanistic explanation that neither requires nor welcomes the autonomy thesis.11 Their model allows for the kind of “upward- or outward-looking,” contextual, higher-level explanation that functionalists trumpet, but without the suggestion that the availability of those explanations depends on their autonomy. And the model also allows for the kind of “downward- or inward-looking” explanations of things in terms of their composition, but without the worry that this will lead to the “reduction” or elimination of the entities or phenomenon so composed. Craver and Darden (2001) apply this model to the case of explaining memory in terms of neurobiological mechanisms, just the sort of case that worries the advocates of autonomy. And Craver (2001) explicitly contrasts mechanistic explanation with Cummins’s notion of functional analysis. According to Machamer, Darden, and Craver: “Mechanisms are entities and activities organized such that they are productive of
197 Functional Properties and Mechanistic Explanations
regular changes from start or set-up to finish or termination conditions” (2000: 3). They argue that the methodologies of neuroscience and molecular biology are best understood as processes for discovering mechanisms, and that explanation in those sciences often involves locating entities and activities within a hierarchy of mechanisms. Machamer, Darden, and Craver are mainly concerned with how it is that describing mechanisms can be explanatory and why it is that discovering mechanisms can be a substantial scientific achievement, though they offer that “Thinking about mechanisms presages new ways to handle some important philosophical problems and concepts” (ibid.: 23). There are five features of the Machamer, Darden, and Craver account of mechanism and mechanistic explanation that we should notice. First, mechanistic explanations are causal, and they are typically multilevel and hierarchical. Second, ideal mechanistic explanation requires both contextual and constitutive explanations, and it does not suppose that they are in tension with one another. Third, mechanistic explanations may be useful even when they are incomplete. Fourth, the notion of levels invoked in mechanistic explanation is epistemic and not absolute. Finally, mechanistic explanations are silent concerning the ontological nature of the entities whose activities they describe. Mechanistic explanations succeed when they identify the entities and activities that compose a system and account for its activities. If the system is a simple mousetrap then the entities are the parts of the mousetrap (the example is from Craver and Bechtel, forthcoming). In the “loaded” position, the spatial organization of the parts is such that a catch holds back a spring-loaded bar. When the trigger is pressed, it releases the catch, thereby allowing the bar to swing freely under the tension of a coiled spring. The bar is sized and positioned so that when it swings it will quickly strike any object that is located on the trigger. The activities in this system are causal: The catch holds the bar, the trigger releases the catch, and so forth. The entities may be proper parts of the mechanism, as in the mouse trap example. But the entities may also be organized in a context of a system of which the mechanism is a part. The mousetrap may be a part in some Rube Goldberg apparatus for ironing shirts, say.
198 Chapter 6
Mechanistic explanation of the carburetor discussed in chapter 1 would involve not only describing the parts of the carburetor and how they are organized to mix air and fuel (figure 1.2), but also locating the carburetor in the operation of a combustion engine of which it is a part (along with lubrication and exhaust systems, among others), and the engine in the operation of the drive system (along with the transmission, etc.), and so on. At each organizational level, the mechanism consists of entities and their activities. An ideally complete mechanistic explanation of some system describes it in three ways.12 The system is described at its characteristic level; this will be an ordinary causal explanation that describes the system’s “input-output” relationships (level 0). In addition, the system will be situated in the context of some other system(s) of which it is a part and to which its activities contribute (level +1). And the system will be decomposed into its constitutive parts (level -1), so that we understand how they enable the input-output relations that are characteristic of the system at the original level of interest (level 0). Mechanistic explanation involves integrating the organization of entities and activities across the levels (figure 6.1): “The goal might be put in the form of a directive. An activity (F) is fully integrated into a multilevel mechanism when (i) the activity has been fit into the organization of a higher (+1) level mechanism, (ii) the isolated (0-level) activity has been adequately described, and (iii) the activity has been explained in terms of its lower (-1) level mechanism” (Craver 2001: 67–68). Mechanistic explanation is inherently multilevel. Needless to say, ideally complete mechanistic integration is difficult to achieve, and ideally complete mechanistic explanations are correspondingly scarce; even partial integration is quite demanding. This is why discovering mechanisms is a scientific achievement (ibid.; Craver and Darden 2001). Mechanisms are typically discovered in a piecemeal manner. Often, either because of the state of the science or because of our explanatory interests, we make use of incomplete mechanistic explanations. In those cases it may be useful to formulate a mechanism sketch or mechanism schema (Machamer, Darden, and Craver 2000: 15–18; Craver and Darden 2001: 120). A mechanism sketch contains gaps that cannot presently be filled in. Those gaps are locations for
199 Functional Properties and Mechanistic Explanations
contextual (+1) mechanism S of ψ-ing, composed of φ-ing Xs
isolated φ-ing of an X
constitutive (-1) mechanism for Xs φ-ing, composed of σ-ing Ps
Figure 6.1 The hierarchy of mechanisms. Adapted from C. Craver, “Role functions, mechanisms, and hierarchy,” Philosophy of Science 68 (2001): 53–74. Used by permission of Carl Craver and the University of Chicago Press. Copyright 2001 by the Philosophy of Science Association. All rights reserved.
additional research. Mechanism schemata are also incomplete in a way; but rather than containing gaps that cannot presently be filled in, they are abstractions from what is known that can be completed in more than one way. Mechanism schemata include one or more roles that can be “instantiated” (Machamer, Darden, and Craver 2000) or realized by one or more mechanism. The open role can be specified by the contextual description that tells us how it and its activities contribute to a containing system. Such a description looks very much like functional specification of the sort advocated by Cummins (1975, 1983); as noted earlier, Cummins himself is mainly
200 Chapter 6
concerned to explicate a model of explanation rather than to defend an analysis of functions or an ontological thesis. However, there are two key differences between Cummins’s “functional analysis” and the model of mechanistic explanation that we are now considering. One is that Cummins emphasizes the explanation of an entity and its capacities by describing its “function” in a containing system, which corresponds to the contextual mechanistic explanation advocated by Craver (2001). But the mechanistic model emphasizes the multilevel integration of the contextual explanation with isolated and constitutive explanations. “Functional analysis” in Cummins’s sense is only one part of mechanistic explanation. The second is that, because it settles for identifying mechanistic roles, functional analysis typically results in mechanism sketches rather than schemata or full-blown mechanistic explanation. Causal-role functions give the input-output profile for an isolated entity. A functional analysis of a car might include the role specification of the “airfuel mixer” that must take certain inputs and generate certain outputs. But that functional entity—the air-fuel mixer—is a black box about which no more can be said. Considered in isolation, that role could be occupied by a carburetor or a fuel injector. But integrating the isolated role description into a multilevel mechanism may well provide constraints—for example, spatial and temporal constraints—that provide information about the lower-level mechanisms that fill the role in a particular system. On the Machamer, Darden, and Craver picture, which level or levels we want to explain will depend on which activities or phenomenon we hope to understand. The three kinds of mechanistic description that are required for complete integration are “not to be taken as divisions in the furniture of the world. . . . Instead, they are distinct perspectives on an activity in a hierarchically organized mechanism” (Craver 2001: 67). The talk of levels, then, is entirely epistemic; what counts as a level depends on our explanatory interests. Levels are understood to be relative: there is no such thing as “the 0 level” or “the +1 level.” As far as the mechanistic explanatory model is concerned, there is no absolute bottom or top level. Explanation in some scientific discipline might “bottom out” with entities that are located at the -1 or +5 level with respect to some other
201 Functional Properties and Mechanistic Explanations
explanatory interests. There is no suggestion that the legitimacy of a level of explanation either depends on the entities it mentions being “reducible” to or autonomous from the entities and activities at adjacent levels. The mechanistic model need not make commitments about the nature of the entities that figure in the various levels of mechanistic explanation. The picture of multilevel, hierarchically organized mechanisms is reminiscent of that suggested by Lycan’s (1987) idea that the world is functions all the way down. But whereas Lycan takes the hierarchical arrangement of functional levels to tell us something about the ontology of nature, namely, that it is essentially functional, the mechanistic view remains ontologically neutral. It is compatible with functionalism; but it is also compatible with the identity theory. Comparing his view with Lycan’s, Craver writes, “I would rather put the distinction like this: see the world as a mechanistic hierarchy, and the distinction between a contextual role (+1), an isolated activity (0), and its constitutive mechanism (-1) goes relative to a perspective on an activity at a given level in a mechanistic hierarchy” (2001: 67). The distinction between the two approaches is perhaps best viewed in the context of the autonomy argument. If the requirement for autonomy must be met then the legitimacy of (relatively) higherlevel explanations depends on their picking out autonomous regularities. Some entities may be autonomous because they are constituted by their contextual (+1) relations. This is what the functionalist claims about mental states: “I propose to type-identify a mental state with the property of having such-and-such an institutionally characterized state of affairs obtaining in one (or more) or one’s appropriate homunctional departments or subagencies” (Lycan 1987: 41). However, if the autonomy requirement can be rejected, then accepting higher-level explanations does not depend on their naming autonomous, institutionally or contextually—that is, functionally—characterized entities. Mechanism and mechanistic explanation seem to allow us to adopt this alternative viewpoint. From the perspective of mechanism, what the functionalist view gets right is that the notion of a level or role is relative. But the functionalist goes wrong in supposing that one can get a complete explanation of a phenomenon at any single level. Mechanism regards
202 Chapter 6
0-level descriptions and explanations as incomplete and as potentially enriched by (but not in competition with) descriptions and explanations at levels +n or -n. Functionalism also goes wrong in supposing that these levels are ontological. Mechanism does not require us to adopt any particular stance about the nature or autonomy of the entities and properties that figure in explanations at any arbitrary level. It is compatible with the existence of some functional entities, but it does not entail them; it is also compatible with identity theory. If Machamer, Darden, and Craver are right, the legitimacy of psychological or neurobiological explanations do not depend on the “reduction” or autonomy of psychology. Mechanism is an explanatory model that rejects the autonomy thesis. 5 Explanation without Autonomy Mechanism is not the only approach to explanation that rejects the autonomy thesis. Consider an anecdote from Wesley Salmon (1989). A “friendly physicist” encountered a child on a plane. The child was holding a balloon on a string, and the physicist asked the child what would happen to the balloon during the takeoff. The child answered that the balloon will move toward the back of the cabin. The physicist says that he thinks the balloon will move toward the front of the plane; lo and behold, it behaves as he predicts. (As the story goes, the friendly physicist ends up winning a little bottle of Scotch from the flight attendant, who did not believe his prediction.) Salmon asks, Why did the balloon move toward the front of the cabin? Two explanations can be offered. . . . First, one can tell a story about the behavior of the molecules that made up the air in the cabin, explaining how the rear wall collided with nearby molecules when it began its forward motion, thus creating a pressure gradient from back to front of the cabin. This pressure gradient imposed an unbalanced force on the back side of the balloon, causing it to move forward with respect to the walls of the cabin. Second, one can cite an extremely general physical principle, Einstein’s principle of equivalence, according to which an acceleration is physically equivalent to a gravitational field. Since helium-filled balloons tend to rise in the atmosphere in the earth’s gravitational field, they will move forward when the airplane accelerates, reacting as they would if a gravitational field were suddenly placed behind the rear wall. (1989: 183)
203 Functional Properties and Mechanistic Explanations
Salmon takes this example to show the complementarity of two views of explanation whose rivalry he chronicles. The first kind of explanation Salmon calls causal/mechanical, the second unificationist (ibid.: 183–185). Causal/mechanical explanations describe the causal processes involved in a phenomenon, such as the balloon’s motion; unificationist explanations appeal not to particular causal mechanisms but to general principles, such as the principle of equivalence (ibid.). According to Salmon’s way of thinking about different explanatory criteria, both functionalism and identity theory provide causal/mechanical explanations of mental states. Both explain what makes a state a mental state by appealing to the mechanisms that produce the phenomena of interest. Both, that is, are causal/mechanical; they differ only over which causal mechanisms they cite. Psychological explanation is a unificationist alternative to functionalism and identity theory; it explains behavior by appealing to general psychological regularities. Salmon argues that both causal/mechanical and unificationist explanations can be correct (ibid.).13 Of course, two explanations of some occurrence may come into conflict; but that is not necessitated merely by the multiplicity of available explanations. If Salmon is correct, then autonomy is too strong a requirement on the acceptability of an explanation, for the acceptability of each kind of explanation does not depend on excluding other explanations of the phenomenon. Salmon’s friendly physicist example is instructive because it bears striking resemblance to Putnam’s (1975b) example of a square peg and round hole. Yet Putnam arrives at a quite different conclusion; indeed, Putnam takes his example to illustrate the autonomy of special sciences explanations. He imagines a board with square and round holes cut in it (“region 1” and “region 2”), and a square peg (“system A”) just smaller in height and width than the two holes: “We have the following very simple fact to explain: the peg passes through the square hole, and it does not pass through the round hole” (1975b: 295). Like Salmon, Putnam considers two explanations. First: a microphysical deduction “from just the laws of particle mechanics or quantum electrodynamics that system A never passes through region 1, but that there is a trajectory which enables it to pass through
204 Chapter 6
region 2” (ibid.: 295–296). Second: “that the board is rigid, the peg is rigid, and as a matter of geometrical fact, the round hole is smaller than the peg, the square hole is bigger than the cross-section of the peg. The peg passes through the hole that is large enough to take its cross-section, and does not pass through the hole that is too small to take its cross-section” (ibid.: 296). Putnam contends that if we want to explain why the peg goes through one hole and not the other, then we must or should use the second sort of explanation. He claims that the microphysical deduction is not an explanation at all, or it “is just a terrible explanation, and why look for terrible explanations when good ones are available” (ibid.). Putnam thinks that this example demonstrates that microphysical explanation cannot replace the macrophysical explanation—that macrophysical explanation of the peg’s behavior is autonomous. He believes that the observation generalizes, showing that psychological explanation is also autonomous. Why is Putnam tempted to deny that there can be microphysical explanations of the behavior of pieces of inert matter? Perhaps we should conclude that he says this only because he believes that there must be some facts left unaccounted for by the microphysical account such that a proper explanation essentially involves the macrophysical objects, thus validating their reality. Likewise Putnam does not simply claim that we do not or prefer not to explain psychology in microphysical terms; he goes so far as to claim that “Whatever our mental functioning may be, there seems to be no serious reason to believe that it is explainable by our physics and chemistry” (ibid.: 297).14 Otherwise, Putnam seems to fear, square pegs and mental states will turn out to be unreal. But if Salmon is right, then the acceptability of an explanation does not depend on the unavailability of alternative explanations. Salmon’s pluralism differs somewhat from the variety licensed by the mechanistic model of Machamer, Darden, and Craver. Whereas the mechanistic model is mainly concerned with explanations at different levels, Salmon is concerned about the general possibility of accepting various distinct explanatory criteria and claims. But both offer ways of thinking about explanation that reject the autonomy thesis.15
205 Functional Properties and Mechanistic Explanations
6 Mechanism and Functionalism, Revisited The mechanistic model is not a “reductionist” style of explanation. In this respect it agrees with functionalism. Although mechanism permits multiple levels of causal-mechanical explanation, it is not “antireductionist,” either, because that view still clings to the autonomy thesis. Classical “reductionist” inward- and downward-looking explanations emphasize the constitutive part of mechanistic integrations and neglect contextual explanation. In contrast, functional analysis emphasizes outward- and upward-looking contextual explanation and neglects constitutive explanation. Both, therefore, are incomplete from the point of view of mechanism. Understanding mechanism can help us to avoid another source of confusion, as well. As we have seen, philosophers who are concerned about autonomy typically hold that the legitimacy of higher-level explanations depends on their being functional explanations ranging over autonomous functional entities. This amounts to counting a whole range of ordinary causal/mechanical explanations as functional explanations. Indeed, if Shoemaker (1981b) is correct, all causal relations are functional relations in this sense. Once one assimilates all causal relations into the class of functional relations, denying that functional relations are what matter in a naturalistic theory seems to be nonsense. To deny that the mind is functional is then to deny that it is causal, to sentence the mind to epiphenomenalism or exile it from nature. As a result, an identity theorist who denies that a mental difference (e.g., that between conscious and nonconscious states) is a matter of functional role is accused of harboring residual Cartesian intuitions (e.g., Dennett 1991, 1995). So functionalism must be true! But the mistake, rather, is to think one can import all causal relations into the functional. These unfettered “functional” relations have none of the distinctive qualities of more robust—and more restricted—notions of function, nor their distinctive ontological commitments: “Without such relativization and addition of detail, the claim that psychological states are functional states is incredibly trivial. Every psychological state has causes and effects, and if we look carefully enough, we can always find a causal difference between any
206 Chapter 6
two different psychological states” (Sober 1985: 190). The functionalist, by loosening the notion of a functional relation until it encompasses all causal relations, thus receives illicit support from the plausibility of mechanism.16 This case of mistaken identity has abetted functionalism from the start. For example, Putnam, writing in favor of his hypothesis that mental states are probabilistic automaton states of whole organisms, suggests that “to investigate this hypothesis is just to attempt to produce ‘mechanical’ models of organisms—and isn’t this, in a sense, just what psychology is about?” (1967, in 1975c: 435) My guess is that it is a failure to distinguish functional and mechanistic explanation that leads Shoemaker and Lycan to argue that the world is functions all the way down. On those views the conflation of causal/mechanical entities and functional entities becomes doctrine. As Lycan puts it: “Structure” is (when you think about it) a surprisingly organizational, I would say almost explicitly teleological, term; a structure is an organized collection of elements, somehow held in place and/or serving to hold other things in place for some purpose or other. It does not contrast markedly with “function,” even though it is not synonymous with it (and even though a structure normally serves a function). How might we better express the notion of brute, primitive realizing-stuff that does or is supposed to contrast with the functional? We might try “functional”-as-opposed-to purely mechanical. “Mechanical”? Hardly—mechanisms are functional items par excellence. (1987: 47)
Lycan’s idea, recall, is that there is no structural or conceptual difference between functionalism and the identity theory, for everything or nearly everything is teleological (i.e., functional) to some degree. But I suspect the allure of the idea that everything (or nearly everything) is functional is a result of conflating functional entities with mechanisms. The world is full of mechanisms, but is it full of functions? If we have in mind etiological functions then the answer is surely no. Causal-role functions are plausibly ubiquitous, but their being so is dependent on our employing them in explanations and is not a basic feature of the world. Only if we think of the class of functional relations as including all causal relations does it become plausible that the ontology of the world is univocally “functions all
207 Functional Properties and Mechanistic Explanations
the way down.” This is just what Shoemaker proposes with his causal theory of properties, and what Lycan seems to be headed toward.17 At that point they have entirely conflated functional entities and mechanisms. Consider Lycan’s claim that mechanisms are functional items par excellence. He takes the supposedly thick notion of a function to be more basic than the thinner notion of a causal mechanism. This seems to me to put the cart before the horse; functional entities are mechanisms par excellence (Polger and Flanagan 2001: 122). One might be worried that my appeal to multilevel mechanistic explanations is incompatible with the causal exclusion argument discussed in chapter 4 and thus provides support for functionalism in the final analysis. After all, once we admit a plurality of explanations, what punch is left to the claim that one excludes another? The question is important, and I have no general answer as to when explanations are or are not compatible with one another. But if we examine the cases discussed above we can see how the present measure of pluralism is not incompatible with functional exclusion argument. Recall that one case in which plural explanations were acceptable was Salmon’s story of the friendly physicist. There the two explanations considered are compatible because they are different kinds of explanations with different explanatory criteria. One is a causalmechanical explanation, and the other a unification explanation. One reason, though not the only reason, that Salmon judges that these two approaches to explanation are “not incompatible” (1989: 183) is that he takes causal/mechanical explanations to be ontic (i.e., they reveal hidden mechanisms), whereas he takes unification to be a merely epistemic conception of explanation (i.e., it organizes and systematizes a general worldview). Thus the two kinds of explanation need not make competing ontological claims. Of course there are other approaches to explanation, and some of them may make competing claims. A second case of explanatory pluralism considered was Putnam’s example of the square peg and the round hole. In this example it is not obvious whether the explanations are of different sorts. Perhaps they are not different in that way. We might think of each as a uni-
208 Chapter 6
fication explanation: one subsuming the event under microphysical laws (e.g., “perhaps one could deduce from just the laws of particle mechanics that system A never passes through region 1, but there is at least one trajectory that allows it to pass through region 2” [Putnam 1975b: 295]) and the other under some more mundane principle (e.g., a peg will pass through a hole that is large enough to fit its cross-section). But one might equally well think of both explanations as causal-mechanical, one citing particles and the other blocks. What should be plain is that under both interpretations of the case (unification, or causal/mechanical), the two explanations are operating at different levels: One appeals to microphysical objects and their properties (molecules, charge), the other to macrophysical aggregates and their properties (blocks, diameter). As discussed in chapter 4, there is no exclusion when the causal powers are attributed to different objects. Perhaps the friendly physicist and the square peg do not add to cases of exclusion, but what about the alleged exclusion cases already considered? Can they be maintained? These are the cases where the competing causes are the functionally realized and realizer properties. Can they be rescued by our newfound pluralism? They cannot. In the exclusion cases, two explanations will be of the same sort (causal mechanical), and they will be operating at the same level and appealing to the same objects. The identity theory and functionalism both offer causal/mechanical explanations, for example, of the removal of a cork from a bottle with a corkscrew. The two causal explanations appeal to the shape of a thing and to its being a corkscrew, respectively.18 Kim (1998) argues for exclusion in those cases in which competing causal explanations appeal to the same level of explanation—functional explanations are higher-order but not higher-level. A thing’s being a corkscrew is a property of it (a particular aggregate, if you like) and not of its parts considered individually. (Considered jointly they are of course one thing, the aggregate.) Being a corkscrew is a higher-order property, but not a higher-level property. A thing has the capacity to open the bottle because of its shape, and it (the very same item) has that capacity (the very same capacity) because it is a corkscrew. Thus we have a straightforward case of causal overdetermination. Rejecting the
209 Functional Properties and Mechanistic Explanations
autonomy thesis will not solve that problem. Rejecting autonomy allows only that explanations do not always compete; it does not have the consequence that they never compete.19 If anything, the pluralism we achieved by rejecting autonomy helps to explain why the exclusion argument does not generalize across levels. 7 A Plea for Mechanism I am urging that we adopt mechanism rather than “reductionism” or functionalism as a framework for explaining the mind. This involves rejecting the autonomy thesis, and with it the autonomy argument for functionalism. Whether or not functional explanations are essential to capturing psychological regularities is irrelevant as long as they are adequate causal/mechanical explanations. Likewise, even if psychological explanations fail to be autonomous (i.e., are “reducible”) this will not itself lend any support for an identity theory.20 Both identity theory and functionalism can ground causal/mechanical explanations. Mechanistic explanation does not entail or require a distinctive ontology; it is neutral about the nature of the entities that figure in mechanistic explanations. And it grounds a hierarchical explanatory framework without suggesting that explanation at any particular level must be either “reducible” to or autonomous from the others. As far as mechanistic explanation is concerned, the relationship between an entity at a level of interest and those at higher and lower levels may be type-identity, realization, or simply unknown. This model lends psychology a kind of methodological freedom that is closer to the colloquial connotation of “autonomy.” The interlevel constraints of mechanism are both weaker and more defensible than those of the autonomy thesis.21 Mechanistic explanation is compatible with the identity theory—and it is also compatible with functional explanation and even functionalism as long as they are not understood to presuppose the autonomy thesis. But this compatibility means that the identity theory can make use of explanatory strategies that are usually thought of as unique to functionalism. The revelation that those tools are available to the identity theorist is quite important because it relieves the worry that identity theory, even if it were
210 Chapter 6
ontologically adequate, is explanatorily limited. This is not the case. The identity theory in fact has more explanatory resources than functionalism because it makes use of both contextual and constitutive explanations. Please, take a minute (or more) to do some soul-searching. Why not recognize functionalism as a useful and appropriate corrective to “reductionist” fervor, but one that has run its course? Why not leave behind the restrictions of functionalism as an explanatory and metaphysical doctrine in favor of the richness of mechanism and identity theory? Think about our old friend, the carburetor. It is true that carburetors have the causal-role function of, crudely put, mixing air and fuel, relative to some explanation of the capacities of a combustion engine to power a vehicle. Does that mean that to be a carburetor is to realize the function, crudely put, of mixing air and fuel? No. Fuel injectors mix air and fuel. Some carburetors (viz., broken and defective carburetors) do not or are not apt to mix air and fuel. Carburetors are mechanical devices par excellence. But isn’t it important, you might wonder, that carburetors are supposed to mix air and fuel? If that is correct, then we have a different notion of function on our hands—one that is relativized to an explanation not of what the carburetor does for the car but perhaps of how the carburetor came to be part of the car, how it came to have venturi, and so forth. That it is supposed to be a carburetor, however, plays no part in the explanation of what the device—the mechanism—in fact does for the car of which it is a component. Far from telling us what kind of thing a carburetor is, the etiological functional explanation presupposes that we already know how to individuate objects. We must know what kind of thing an object is before we can given an explanation of its behavior in terms of what things of such-and-such type are supposed to do. The explanation will appeal to, for example, what things of that type did in the past that explains how they came to be here, and to have the features they do. Can’t we connect the two notions of function? To ask that question is to seek a hybrid teleofunctional account. I am not optimistic that any such account can be found. I argued earlier that the alter-
211 Functional Properties and Mechanistic Explanations
natives seem invariably to collapse into either causal-role or direct teleological functionalism. Here is a litmus test to see if you are a metaphysical functionalist. In chapter 4, I argued that to realize a functional state is to have a function. What do you think of that claim? (Make sure you are using the notion of function univocally in “functional state” and “have a function.”) If that seems wrong to you—if you do not think that functional realization is having a function—then you are not a metaphysical functionalist. If you do not think that functional realization is a matter of having a function or you think that most or all causal relations are functional, and you nevertheless insist that you are a metaphysical functionalist rather than a mechanist, then I am puzzled. In what way is your theory functionalist rather than mechanist? Perhaps a more detailed example would help. In the next chapter I will examine how failing to distinguish functional explanation from causal/mechanical explanation plays a role in Daniel Dennett’s thinking about consciousness.
7 Dennett’s Challenge
In the last chapter I argued that some prominent arguments for functionalism are based on a mistake. The mistake is to confuse functional explanation and its distinctive ontology with mechanism, a general doctrine about explanation that does not have a distinctive ontology. Identity theory is a metaphysical theory of mind that is compatible with a mechanistic model of explanation. In this chapter I will examine one example of a philosophical thesis that depends on the conflation of functional and mechanistic models: Daniel Dennett’s argument that two systems that are functionally identical must be alike with respect to their conscious mental states.1 The dispute is often cast in terms of philosophical fictions known as zombies. Zombies are imaginary creatures that are stipulated to lack consciousness despite being otherwise identical in one way or another to human beings or other conscious creatures. In an essay entitled “The Unimagined Preposterousness of Zombies,” Dennett laments the sad state of philosophy such that there is serious debate over whether or not this fictional kind of being is possible: “Sometimes philosophers clutch an insupportable hypothesis to their bosoms and run headlong over the cliff edge. Then, like cartoon characters, they hang there in mid-air, until they notice what they have done and gravity takes over. Just such a boon is the philosophers’ concept of a zombie, a strangely attractive notion that sums up, in one leaden lump, almost everything that I think is wrong with current thinking about consciousness” (1995: 322). Dennett fears that by taking
214 Chapter 7
zombies seriously, even in order to answer the critics, naturalistic philosophers are conceding too much. He won’t even allow that zombies are a useful fiction, a tool for thinking about theories of consciousness. Zombies are simply absurd. He sets down a challenge: “If the philosophical concept of zombies is so important, so useful, some philosopher ought to be able to say why in non-question-begging terms. I’ll be curious to see if anybody can mount such a defence, but I won’t be holding my breath” (ibid.: 326). I agree that many claims about zombies are unsupported or even question-begging. But zombies may nevertheless may be useful fictions. Dennett’s distress is unwarranted. Zombies worry him so only because he is conflating functions and mechanisms. To see this, we’ll carefully examine the idea of a zombie and Dennett’s arguments against the idea. 1 Zombies: A Taxonomy Zombies are stipulated to be creatures identical in some way to human beings, but that lack consciousness. They are stipulated to be at least behaviorally identical to human beings or other conscious creatures. They may also be identical in other ways, as we shall see. The main question about zombies is whether they are possible. The zombie problem, like the problems of absent, inverted, alien, and dancing qualia, is just one way of pushing questions about consciousness (cf. Block 1978; Shoemaker 1982; Chalmers 1996a). These thought experiments, and many others, serve the purpose of putting questions about the nature and causal efficacy of consciousness in a particularly salient form. By facing us with entities that exemplify our theories, thought experiments force us to think carefully about proposed explanations of mind and consciousness. Of course, no thought experiment is to be conducted in a vacuum. The goal of such considerations is to draw out the commitments, conditions, and caveats of various theories about mind and consciousness. Are zombies possible? The answer depends on the details of how the zombies are stipulated, what kind of possibility is in question, and what sort of theory of consciousness you hold. As Güven Güzeldere writes, “playing with the idea of zombies could turn into playing with philosophical fire. But precisely for that reason, it is
215 Dennett’s Challenge
important to pay attention to the particulars in using zombies as a tool of imagination in thought experiments” (1995a: 327). Güzeldere (1995a) provides a taxonomy of zombies based on the question of how zombies are said to be identical to conscious creatures: Are they supposed to be behaviorally, functionally, or physically2 identical to conscious beings? In earlier chapters we have seen that functions are not all of one sort; it follows that “functional identity” may have different interpretations. But abstract and etiological functions are not what are at stake when it comes to zombies, at least not those that concern Dennett. In the present context, some sort of causal function is assumed, that is, a notion of function that has defining relations that are more abstract than basic physical relations but less abstract than mere behavioral relations. In addition to Güzeldere’s distinction between the ways in which zombies may be similar to conscious creatures, Flanagan and I (1995a) drew a modal distinction: Are zombies logically, metaphysically, or naturally possible?3 Taken together, these two distinctions form the axes of the Zombie Scorecard (figure 7.1). The Zombie Scorecard is a way of organizing nine questions about zombies.
behavioral
identity functional
physical
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
metaphysical logical
possibility
natural
(1)
Figure 7.1 The Zombie Scorecard.
216 Chapter 7
(Q1) Is it naturally possible that there be zombies that are behaviorally identical to human beings? (Q2) Is it naturally possible that there be zombies that are functionally identical to human beings? (Q3) Is it naturally possible that there be zombies that are physically identical to human beings? (Q4) Is it metaphysically possible that there be zombies that are behaviorally identical to human beings? (Q5) Is it metaphysically possible that there be zombies that are functionally identical to human beings? (Q6) Is it metaphysically possible that there be zombies that are physically identical to human beings? (Q7) Is it logically possible that there be zombies that are behaviorally identical to human beings? (Q8) Is it logically possible that there be zombies that are functionally identical to human beings? (Q9) Is it logically possible that there be zombies that are physically identical to human beings?4 I assume that logically possible is just not contradictory. There is some question about whether there is an even weaker sort of possibility, something like conceivability (Horgan 1987) or epistemic possibility (Kripke 1972/1980), but we can ignore that idea since it does not play a part in the present discussion. Naturally possible I take to be something like compatible with all and only the actual substances and laws of nature.5 There is no general agreement about the status of metaphysical possibility. Whether or not metaphysical possibility is connected to conceivability is a point of contention, although there is de facto (if grudging) consensus that conceivability is our best guide to metaphysical possibility.6 The lack of agreement about metaphysical possibility makes it difficult to fill in the middle row of the Zombie Scorecard. Behaviorally identical zombies make all the overt movements and utterances that conscious creatures do, but they may have any internal structure and be composed of whatever material. Behaviorally identical zombies needn’t be hollow shells; they could be
217 Dennett’s Challenge
quite sophisticated. However, in considering behaviorally identical zombies we leave their internal organization unspecified. Functionally identical zombies not only make the movements that conscious creatures do, but also have an internal organization in some sense the same as that of conscious creatures. (For this reason we take the “function” of functionally identical zombies to be a variety of causal function.) Physically identical zombies are identical to conscious creatures cell for cell, molecule for molecule, or atom for atom. The logical possibility of physically identical zombies—Robert Kirk’s (1974) original zombies—was introduced as a problem for materialism. The strongest claim represented in the Zombie Scorecard is of the natural possibility of physically identical zombies. The weakest claim is of the logical possibility of behaviorally identical zombies.7 If the zombie questions are asked in the form, “Is it y-ly possible that there be creatures that are x-ly identical to human beings but that lack consciousness?” (where y is a mode of possibility and x is a degree or kind of identity), then someone might object that this begs the question as to whether human beings are conscious. I believe that human beings are conscious. But to avoid any appearance of impropriety, the questions should be rephrased, “Is it y-ly possible that there be two creatures that are x-ly identical to one another but differ in that one is conscious and the other is not?” This formulation has the additional advantage of rendering transparent how the form of the zombie construct is related to absent, inverted, alien, and dancing qualia thought experiments:8 Zombies and absent qualia: Is it y-ly possible that there be two creatures that are x-ly identical to one another but differ in that one is conscious and the other is not? (cf. Kirk 1974; Block 1978) Inverted qualia: Is it y-ly possible that there be two creatures that are x-ly identical to one another but differ in that one’s consciousness is “inverted” with respect to the other’s? (cf. Shoemaker 1982) Alien qualia: Is it y-ly possible that there be two creatures that are xly identical to one another but differ in that one has consciousness that is entirely different in quality from those had by the other? (cf. Lewis 1980)
218 Chapter 7
Dancing qualia: Is it y-ly possible that there be two creatures that are x-ly identical to one another but differ in that one is always conscious and the other sometimes has the same sort of consciousness as the first and other times has a different sort of consciousness (inverted or alien) or none at all (absent)? (Cf. Chalmers 1996a) Zombies are just one among many thought experiments designed to explore our theories of consciousness. This brief sample of the wrangling over formulations serves as a reminder that, when it comes to zombies, the details matter. After all, zombies are stipulated, not discovered. Imagining zombies is just a vivid way of forcing ourselves to face the consequences of views we already hold. Some of the consequences of a view, or the caveats necessary to maintain it, may not be palatable to all philosophers. But those consequences and caveats don’t come from the notion of a zombie; they are merely highlighted by asking the zombie questions. As Güzeldere writes, “Belief in zombies has become a litmus test for intuitions in recent philosophy of mind. . . . The set of answers one chooses to give to questions of this sort is usually a good indicator of where one stands with respect to a variety of issues regarding consciousness—its ontology, nature, function, evolutionary role, and so on” (1995a: 326–327). But this is precisely what Dennett will have none of. Dennett says the burden on the zombie defender is the following: “One must show that there is a difference between conscious beings and zombies, and one must show that one’s demonstration of this difference doesn’t depend on underestimating in the well-nigh standard way the powers of zombies” (1995: 325). The disagreement between Dennett and zombie defenders seems to concern the natural possibility of functionally identical zombies (Güzeldere 1995a). If that is right, then the above statement of the challenge makes it look as though the burden on the zombie defender is to show that functionally identical zombies are naturally possible—that is, to show that functionalism is false. But Dennett’s assertion overstates the burden. Elsewhere he puts the point differently, fretting that “the philosophers’ concept of a zombie, a strangely attractive notion . . . sums up, in one leaden lump, almost everything that I think is wrong with current thinking about
219 Dennett’s Challenge
consciousness” (1995: 322; emphasis added) and writing, “I have never seen an argument in support of the zombie distinction that doesn’t make a mistake of imagination . . .” (ibid.: 325; emphasis added). With these statements Dennett makes a more modest demand: that someone defend the concept of zombies. Dennett’s challenge is to meet the modest demand, to show that the concept of zombies is not incoherent. To answer this challenge I do not have to show that functionally identical zombies are naturally possible. It will be more than enough if I show that functionally identical zombies can be conceived in a way that is not selfcontradictory, is useful, and is not question-begging. Of course it is going to matter a great deal whether the functionally identical zombies are fine- or coarse-grain duplicates of conscious creatures.9 This, I take it, is part of Dennett’s point in reminding us about his modified, complex zombies: zimboes (1991, 1995). 2 Zombies and Zimboes Given uncompromising remarks such as “It’s hard for me to keep a straight face through all this, but since some very serious philosophers take the zombie problem seriously, I feel obliged to reciprocate” (Dennett 1991: 95) and “I confess that try as I might, I cannot summon up conviction for any other verdict: Zombies are ridiculous!” (Dennett 1994: 540), it would be understandable if one thought Dennett’s position on zombies was clear and simple. But on closer inspection, his view is neither clear nor simple. Dennett argues that all extant discussions of zombies depend on “underestimating in the well-nigh standard way the powers of zombies” (1995: 325). Let us begin with a passage in which he himself appears to make exactly this mistake. First Dennett reminds us of his functionally sophisticated kind of zombie, the zimbo: “In [Consciousness Explained] I introduced the category of a zimbo, by definition a zombie equipped for higher-order reflective informational states” (1995: 322). Zimboes are zombies that are, in some sense, functionally complex. Dennett then goes on to say,
220 Chapter 7
As I pointed out when I introduced the term, zombies behaviourally indistinguishable from us are zimboes, capable of all the higher-order reflections we are capable of, because they are competent, ex hypothesi, to execute all the behaviours that, when we perform them, manifestly depend on our higher-order reflections. Only zimboes could pass a demanding Turing Test, for instance, since the judge can ask as many questions as you like about what it was like answering the previous question, what it is like thinking about how to answer this question, and so forth. (1995: 323)
But zombies that are behaviorally indistinguishable from us are . . . zombies! Zimboes don’t behave any differently (neither better nor worse) than their functionally less sophisticated cousins, the behaviorally identical zombies. Zimboes are zombies, and all zombies are at least behaviorally identical to conscious creatures. As Dennett himself notes, “If, ex hypothesi, zombies are behaviourally indistinguishable from us normal folk, then they are really behaviourally indistinguishable!” (ibid.: 322). The Turing test is an entirely behavioral test, so it could not be used to distinguish between zombies and zimboes. It’s simply not the case that only zimboes (functionally sophisticated zombies) could pass the Turing test—any zombie could pass that test. It looks as though Dennett has underestimated the powers of zombies. Since he raises these points in the context of arguing that the zombie concept is irreparably confused, some might expect Dennett to respond, in Wittgensteinian fashion, that he was showing us what he could not say with an argument, namely, that the concept of zombies is self-contradictory. But there is a better explanation: Dennett believes that merely behaviorally identical zombies are not naturally possible, so any actual zombie must be at least functionally sophisticated—at least a zimbo. Dennett holds that it is a fact about human beings that nothing could be behaviorally identical to us without also having a high degree of functional sophistication, without “higher-order reflective informational states” (1995: 322). For example, he responds to the suggestion that conscious pain is not necessary for injury avoidance (Flanagan and Polger 1995), writing, “In creatures as cognitively complex as us (with our roughly inexhaustible capacity for metareflections and higher-order competitions between policies, meta-
221 Dennett’s Challenge
policies, etc.), the ‘blood-is-about-to-be-lost sensors’ and their kin cannot simply be ‘hooked-up to the right action paths’ as Flanagan and Polger put it” (ibid.: 323).10 For Dennett, behaviorally identical zombies—that are not also functionally sophisticated—are not naturally possible. That is why Dennett claims that only functionally sophisticated zombies (zimboes) could pass the Turing test. But Dennett is certainly not making the Cartesian point that there are fundamental limits on the mechanical powers of organized bits of matter; and he is not arguing that the concept of a behavioral duplicate is ipso facto the concept of a conscious thing (as analytical behaviorism would have it). Rather, Dennett is making the straightforward engineering point that performing sophisticated behaviors requires sophisticated mechanisms. Imagining a behavioral duplicate that is a hollow shell is an exercise in fantasy. If I am correct, Dennett would probably respond that examples of behaviorally identical zombies are possible only if we imagine them to have complex control mechanisms. Dennett is not arguing that consciousness is constituted by behavior. He is arguing that the sorts of mechanisms responsible for consciousness are also required to perform certain behaviors. If a future George Lucas or Steven Spielberg or Rodney Brooks is going to design a perfect behavioral duplicate, it is going to have to be complicated.11 Suppose Dennett is right that the degree of functional complexity that is a prerequisite for behavioral identity is also sufficient for consciousness. In that case there could be no zombies that are merely behaviorally identical to conscious things. Even so, it would not be because the behavioral duplicates are behaviorally identical that they are conscious, but because the duplicates are functionally sophisticated. In virtue of their complex functional organization they are conscious, not in virtue of their behavioral identity simpliciter. Clarifying Dennett’s claims about zombies and zimboes is sufficient to dissolve the prima facie tension in his remarks. For Dennett, behavior is a highly reliable indicator of consciousness. Dennett’s reliance on zimboes and his conviction that behaviorally identical entities must also be functionally sophisticated reinforce the claim that what is at stake in Dennett’s arguments is the natural possibility of functionally identical zombies. Let us agree that behavioral
222 Chapter 7
identity, if we are not to be dabbling with ghosts, or gremlins, or elan vital, requires complex mechanisms. What zimboes have that zombies do not is a sophisticated internal structure—they are functionally complex. But what is in question with zombies is functional identity, not just functional sophistication; so it must be that Dennett intends the zimboes to count as functionally identical to us. Even most functionalists should agree that there is no reason to suppose that those mechanisms necessary for behavioral identity must be functionally identical to our own mechanisms, unless to produce the same overt behavior and utterances is ipso facto to be functionally identical. That is to confuse functional and behavioral relations; behaviors, like functions, are supposed to be multiply realizable. On the other hand, if behavioral identity requires functional identity—as opposed to functional sophistication—we needn’t accept that every mechanism functionally identical to a conscious state is thereby a conscious state. To do so—if behavioral identity requires functional identity—is to accept a simplistic version of behaviorism. But this objection is (or is part of) a response that Dennett dismisses; he compares it to vitalism and derides it as “too puny to weigh against the account of life presented by contemporary biology” (1991: 281–282). But the response is not puny. And the comparison to vitalism should remind us of the debate we considered in the previous chapter; it is a hint to Dennett’s confusion. What is Dennett’s argument for the conclusion that there are limits on nonconscious mechanisms? If—and this is the reading that I have been discouraging—Dennett is maintaining that it is a logical truth that all mechanisms functionally identical to conscious creatures are ipso facto conscious mechanisms, then he has provided no argument for that claim. If—as I have suggested—Dennett is making a claim about the natural limits of nonconscious mechanisms, he has provided no constructive argument for that claim, either; but he does offer an argument. The single reason Dennett provides for holding this view is that the alternative (viz., the possibility that functionally equivalent mechanisms may differ in that some are conscious and some are not) leads to epiphenomenalism and “pernicious nonsense.” If Dennett is right, those consequences
223 Dennett’s Challenge
would indeed worry naturalistic defenders of zombies. But Dennett’s reasoning is faulty. 3 Zombies and Epiphenomenalism What is difficult to discern in the discussion of zombies and zimboes in Consciousness Explained is that Dennett is presenting the zombie defender with a dilemma. He makes the point directly in “The Unimagined Preposterousness of Zombies”: [The point of introducing zimboes] was to make a distinction within the imaginary category of zombies that would have to be granted by believers in zombies, and that could do all the work they imputed to consciousness, thereby showing either that their concept was subtly self-contradictory, since some zombies—zimboes—were conscious after all, or that their concept of consciousness was not tied to anything familiar and hence amounted to an illicit contrast: consciousness as a “player to be named later” or an undeclared wild-card. (1995: 322–23)
So the purpose of introducing a distinction between zombies and zimboes—between behaviorally and functionally identical zombies —is to set up a dilemma: Either all the differences between the conscious and unconscious are functional differences (remember, Dennett holds that there are certain things that only a functionally complex zimbo can do), or else the alleged difference is “not tied to anything familiar” and is thus consciousness “in the systematically mysterious way that supports such doctrines as epiphenomenalism” (1991: 406). The shape of Dennett’s challenge is now clear. Either the concept of a functionally identical zombie is self-contradictory, or else it involves a conception of consciousness as epiphenomenal.12 There can be little doubt about the first horn of the dilemma. Zombies are incoherent on Dennett’s functional conception of consciousness. So to answer Dennett’s challenge, one must address the second horn of the dilemma—one must show that a conception of consciousness that allows for the possibility of functionally identical zombies does not entail epiphenomenalism. Dennett defines philosophers’ notion of epiphenomenalism as follows: “ ‘x is epiphenomenal’ means ‘x is an effect but itself has no
224 Chapter 7
effects in the physical world whatever’ ” (1991: 402). This is what Dennett is referring to as epiphenomenal “in the ridiculous sense.” I call this kind of epiphenomenalism strict metaphysical epiphenomenalism (Polger and Flanagan 2002). In contrast, Dennett writes, Huxley’s sense of epiphenomenalism is that of a “nonfunctional property or byproduct.” He continues, “Huxley used the term in his discussion of the evolution of consciousness and his claim that epiphenomenal properties (like the ‘whistle of the steam engine’) could not be explained by natural selection” (1991: 402). Dennett’s explication of Huxley’s notion leaves room for confusion. There is a concept of epiphenomena as not having been selected for by natural selection. Traits that are epiphenomenal in this sense are free riders or, if they are nevertheless useful, spandrels (Gould and Lewontin 1978). I call this notion etiological epiphenomenalism because it denies that a trait has an etiological function (Polger and Flanagan 2002). Flanagan, for example, argues that dreams are etiological epiphenomena (1995b, 1996, 2000; see also Polger and Flanagan 1999, 2002). But etiological epiphenomenalism is not what Huxley had in mind (even though he was discussing natural selection), and it is not what Dennett refers to using Huxley’s name. Huxley’s notion is that some physical effects of mechanisms are not themselves part of the operation of the mechanism (the steam of the locomotive to the engine, the bell to the clock in a clock tower). The sounds of whistling steam and ringing bells are physical effects (therefore not systematically mysterious), but they are not parts of some particular mechanistic systems. I have called this causal-role epiphenomenalism after Cummins’s (1975) notion of causal-role function, because it denies that a trait has a role in a causal system (Polger and Flanagan 2002). Causal-role epiphenomenalism is what Dennett calls epiphenomenalism “in Huxley’s sense.” Why does Dennett think the mere possibility of functionally identical zombies entails a “systematically mysterious” epiphenomenal concept of consciousness? The argument is contained in a single dense paragraph in Consciousness Explained that pursues two lines of reasoning from the possibility of some kind of zombies to some sort of epiphenomenalism. The first part reasons from the “in principle”
225 Dennett’s Challenge
indistinguishability of zombies to epiphenomenalism “in the ridiculous sense,” strict metaphysical epiphenomenalism. The second part reasons from the functional (but not physical) indistinguishability of zombies, to the claim that consciousness is epiphenomenal “in the Huxley sense,” causal-role epiphenomenalism. Dennett first argues that the “in principle” indistinguishability of zombies entails epiphenomenalism in the ridiculous sense: A philosopher’s zombie, you will recall, is behaviorally indistinguishable from a normal human being, but is not conscious. There is nothing it is like to be a zombie; it just seems that way to observers (including itself, as we saw in the previous chapter). Now this can be given a strong or weak interpretation, depending on how we treat this indistinguishability to observers. If we declare that in principle, a zombie is indistinguishable from a conscious person, then we would be saying that genuine consciousness is epiphenomenal in the ridiculous sense. That is just silly. (1991: 405)
How does the “in principle” modifier operate in this argument? We already know that behaviorally indistinguishable zombies are, for Dennett, also functionally sophisticated. Perhaps it’s not just that the functionally identical zombies are indistinguishable in practice, but rather that they are really functionally indistinguishable—they are functionally identical. So “in principle” is contrasted with “in practice.” If this is what Dennett has in mind, then the first argument parallels the second argument (viz., that the possibility of functionally identical zombies entails that consciousness is epiphenomenal only in the Huxley sense) except that it vies for the stronger conclusion, that consciousness is epiphenomenal in the ridiculous sense. A better interpretation is that “in principle” works to modify the indistinguishability of the zombies. So when Dennett talks about zombies that are “in principle” indistinguishable, he has in mind zombies that are identical in every way to conscious creatures, that is, physically identical zombies. The argument, then, is that if physically identical zombies are naturally possible, then consciousness is epiphenomenal in the ridiculous sense. On this reading, the first argument is a change of target for Dennett, since he is otherwise primarily concerned with functionally identical zombies. In this case, so long as the debate is constrained to broadly naturalistic views (leaving aside considerations of dualism), it is correct to say that the
226 Chapter 7
natural possibility of physically identical zombies entails that consciousness is epiphenomenal in a way that would render it somewhat mysterious.13 Almost none of Dennett’s opponents disputes this point; most naturalists deny the natural possibility of physically identical zombies.14 Dismissing the “in principle” indistinguishable zombies, Dennett sets into an alternative interpretation of the zombie defender’s claim. Dennett’s second argument is that the natural possibility of functionally identical zombies entails that consciousness is epiphenomenal merely in the Huxley sense: “So we could say instead that consciousness might be epiphenomenal in the Huxley sense: although there was some way of distinguishing zombies from real people (who knows, maybe zombies have green brains), the difference doesn’t show up as a functional difference to observers. Equivalently, human bodies with green brains don’t harbor observers, while other human bodies do” (1991: 405). Earlier Dennett writes that if qualia are epiphenomenal in the Huxley sense of epiphenomenalism, then they “are physical effects and have physical effects; they just aren’t functional. Any materialist would be happy to admit that this hypothesis is true” (ibid.: 404). He continues, “That cannot be what epiphenomenalists have in mind, can it? If it is, then qualia as epiphenomena are no challenge to materialism” (ibid.: 405). I agree. Here Dennett is concerned to make the point that the Huxley sort of epiphenomenalism cannot be, specifically, what Frank Jackson is defending in his “Epiphenomenal Qualia” (1982). But the point is perfectly general: If consciousness is epiphenomenal in the Huxley sense it is no problem for materialism, for it is not systematically mysterious. Given this recognition, it is puzzling that Dennett proceeds to argue that the possibility of zombies entails that consciousness is epiphenomenal in the Huxley sense—and therefore that “It is time to recognize the idea of the possibility of zombies for what it is: not a serious philosophical idea but a preposterous and ignoble relic of ancient prejudices. . . . What pernicious nonsense” (1991: 405–406). Two questions arise: Why does Dennett think that the possibility of zombies entails epiphenomenalism in the Huxley sense? And even if
227 Dennett’s Challenge
zombies do have that consequence—I will argue they don’t—what is so pernicious about a possibility that is no threat to materialism? 4 Escaping Dennett’s Zombie Dilemma For Dennett, if two creatures are functionally identical but differ with respect to whether they are conscious, then whatever consciousness is (“a ‘player to be named later’ or an undeclared wild-card”), it does not play a causal role in the system that has it. Thus, it is an epiphenomenon in the Huxley sense. Consider an argument that has the same structure as Dennett’s argument that the possibility of functionally identical zombies entails that consciousness is a causal-role epiphenomenon: We could say that carburetors might be epiphenomenal in the Huxley sense: although there was some way of distinguishing cars with carburetors from “zombie” cars that don’t have carburetors (who knows, maybe zombie cars have electric pumps), the difference doesn’t show up as a functional difference to observers.
Of course some cars do not have carburetors, and they in fact have something like electric pumps—they have fuel injectors. From the fact that a car might have a carburetor or might have a (“functionally identical”) fuel injector, it does not follow that carburetors are causal-role epiphenomena.15 On the contrary, carburetors are crucial mechanisms in the cars that have them, just as fuel injectors are crucial mechanisms in the cars that have them.16 It is a mistake to think that because some mechanism is not required, because it is inessential, that it is epiphenomenal. Because there might be two possible mechanisms that could accomplish a function, Dennett’s logic would have us conclude that neither of them is a causal mechanism—that they are both epiphenomenal in the Huxley sense.17 Now that is ridiculous. If I gave a possible explanation of how your car worked—one that, say, posited a superstrong superfast platypus turning the drive shaft— you would not think that I had threatened your previously held beliefs about automobile engines. So why should it give philosophers
228 Chapter 7
even a moment’s pause if someone claims that it is possible (logically, metaphysically, or naturally) for some creature (real or imagined) to accomplish nonconsciously all the things (behaviors, movements, utterances) that we human beings accomplish with consciousness? When it comes to giving mechanistic explanations, possible explanations won’t do the trick. These would be more gappy than even mechanism sketches. It’s not enough that it be possible that a car be abstractly describable in terms of an organization that allows that a platypus is the source of locomotion. Explanations of particular cars must describe the actual workings of their actual parts. Often more than one part could have been used; the right explanations tell us which of those possible parts actually were used. The explanations had best involve, in the case of my car, a combustion engine rather than a platypus and a fuel injector rather than a carburetor. Of course, Dennett does not encourage us to think of alternate mechanisms that could have the same effects, but rather alternate brain colors. The example of green brains is supposed to make us think that it is silly to believe that some factor other than functional identity could make a difference, for example, to whether something is a conscious brain or is a carburetor. Notice that Dennett directs us to a factor that is causally irrelevant. Brain color is not part of any mechanistic explanation of how brains do what they do. Of course brain color could be a side-effect of some relevant factor, but it is clear that Dennett assumes the color itself to be irrelevant. I could make my zombie car story have more of Dennett’s rhetorical panache (“maybe zombie cars have little magic silver boxes”), but the argument would still be a fallacy. By inviting us to think of a difference that we do not assess as important to whether a creature is conscious (surface color of the brain material), Dennett distracts us from considering the relevant alternative—differences in mechanism. Naturalistic theories of mind, such as the identity theory, are free to construe conscious states, or processes, or events as mechanisms. These mechanisms have certain properties, among which may be those that make them conscious. Mechanisms have causal powers. But unless functional equivalence requires complete causal equiva-
229 Dennett’s Challenge
lence, any mechanism is replaceable with a “functionally equivalent” mechanism that has some distinctive causal properties. On this picture, some mechanisms are conscious and others are not. Prima facie, conscious visual experience is part of the mechanism of visual perception in human beings. But that does not mean that some other creatures could not maneuver through the world without visual experiences—or even that some people could not, as the research into blindsight is sometimes thought to suggest (Weiskrantz 1986, 1997). Such alternative mechanisms would neither be epiphenomenal nor render our actual conscious mechanisms epiphenonemal. A 230-horsepower platypus and a fuel injector that masquerades as a carburetor may not be the best examples of alternate mechanisms. Consider instead an example, adapted from Block (1980d), of the plumbing in a house. A functional account or explanation of the plumbing in the house will include the following sorts of information: the water pressure entering the house, the water pressure exiting from various points (faucets, sewer, etc.), the overall capacity of the system in volume and pressure, required water temperature at various points (water heater, faucets, etc.), and so forth. Let us call this functional explanation of the house the Pipe Theory, P for short. P will mention many of the characteristics of the parts of the plumbing system that transport water from one location to another at certain pressures, temperatures, volumes, and so forth. But P will not include such details as the exact routing of the pipes (kitchen before bathroom, bathroom before kitchen), the cross-section shape of the pipes (oval, circular, square), or even whether they are pipes (rather than hoses or aqueducts). And P will not specify whether the pipes, or hoses, or aqueducts are made of aluminum or plastic. (They don’t all have to be the same, either. There could be some aluminum pipes, some plastic hoses, some cement aqueducts. But it will be simpler to assume that the whole house is built one way or another.) Let us say that the house has pipes, and the pipes are made of aluminum. It is entirely consistent with P that the house had plastic hoses. If there were plastic hoses instead then they might have to run a different route, be a different shape, or some such. But those characteristics of the system were not part of P to begin with. So the
230 Chapter 7
aluminum pipes and plastic hoses are both entirely consistent with P; they make no difference to the functional explanation of the plumbing. Does it follow that either the aluminum pipes or the plastic hoses are epiphenomenal to the plumbing of the house? No. P, being a functional explanation, is open to at least standard MR: it may be realized in indefinitely many ways. P may apply to a house that has aluminum pipes, in which case aluminum pipes are what play the causal role of water-carrier in that system. Those aluminum pipes are not causal-role epiphenomena; they are what carries the water. P may also apply to a house that has plastic hoses, in which case plastic hoses play the causal role of water-carrier in that system. P is an abstract description of one or more possible systems. P picks out parts of the system by their causal roles. A part would never even be picked out or specified by P unless it had a causal role. So how could someone think that a part specified by P could be epiphenomenal in the Huxley sense? One problem lies in thinking that it is the roles, rather than the occupants of the roles, that have the causal powers. P may include claims such as “. . . and the water-carrier bears the water to the waterheater. . . .” But water-carrier is a role. It is not a thing that has causal powers; it is a variable, a way of picking out any number of things in terms of their causal powers.18 P is an abstraction, not an object. Only when it is realized does any of its parts have causal powers. Aluminum pipes can have those powers; plastic hoses can have those powers. Neither P in itself nor any part of it has any causal powers. Realizations of P and their parts—aluminum pipes and plastic hoses—have causal powers. Aluminum pipes are not epiphenomenal in the Huxley sense; they are parts of a causal/mechanical system that realizes P. This distinction between role and realizer helps to explain why a physical difference (aluminum pipe or plastic hose) without a functional difference (both are described by P) does not entail that the different physical parts are each causal-role epiphenomena. Exactly the same reasoning can be applied to consciousness. To an identity theorist, the brain is a complicated sort of plumbing. According to identity theory, consciousness is a pipe—it is a realizer that may
231 Dennett’s Challenge
occupy a role. Suppose, for the sake of argument, there is an account that is a functional explanation of human cognitive capacities. This explanation, y, will specify many aspects of human cognition, many functional parts. Some of these functional parts are states, processes, or properties that are, in us, conscious states, processes, or properties. That is, some brain states, or processes, or properties that realize parts of y are in human beings conscious states, processes, or properties. But y could also apply to another kind of creature, a zombie, for which those roles that are occupied in us by conscious states, processes, or properties are, in it, occupied by nonconscious states, processes, or properties. This zombie would be functionally identical to human beings (relative to y) but would lack consciousness. It does not follow that consciousness is epiphenomenal in the Huxley sense, for conscious states, processes, or properties occupy causal roles in our system. Likewise, some other things occupy those roles in the zombie; and whatever occupies those roles is not epiphenomenal in the zombie.19 Why should we care whether consciousness is part of our system? Functional specification is interest-relative. One might care about the nonfunctional (relative to water-carrying) details of one’s plumbing if, for example, one was inclined toward postindustrial interior design, or if one was concerned about how easy it is to access the pipes for repair. (Dennett [1971] points out that we often care about the physical specification of things when they break.) Likewise, the interests of psychological explanation are not the only human interests. We might care about the physical details of our system that are not specified by y for a variety of reasons, such as aesthetic and moral reasons. Conscious states may be replaceable with respect to our bodies carrying on their cognitive duties, but we value having them. Metaphysical questions about consciousness are directly related to what we think—morally, for example—about animals, infants, fetuses, people with brain damage, sleeping people, people on drugs, computers, thermostats, aliens, and rocks or buckets of water whose molecular motion temporarily mimics a finite state machine. This is why zombies are important. We care about bizarre, extreme,
232 Chapter 7
imaginary cases of creatures that are physically different from, but behaviorally and functionally identical to, us precisely because we care about how they reveal our intuitions about the less bizarre, less extreme, and all too real cases of creatures that are both physically different—because of their nature, age or injury—and functionally or behaviorally different from us. Dennett calls zombies “pernicious” because he thinks that the defender of zombies is committed to saying that these important considerations rest on whether or not a subject has some epiphenomenal quality. That would indeed be troubling. But I have argued that the possibility of functionally identical zombies does not depend on consciousness being epiphenomenal, so Dennett’s worry is unfounded. 5 Revisiting Dennett’s Dilemma Dennett is wrong to think that consciousness would have to be epiphenomenal for functionally identical zombies to be nomologically possible. The explanation that I’ve given is just like that used by Block (1980d) against a similar argument from Shoemaker (1975) regarding absent qualia.20 Block took Shoemaker to be arguing that creatures with absent qualia, zombies, are not possible on the grounds that their consciousness would have to be epiphenomenal (and thus, on a causal theory of knowledge, that they would violate the self-evident fact that we have knowledge of our conscious states). Accordingly, Block was arguing that Shoemaker fails to show that absent qualia are impossible, on the grounds that absent qualia do not entail epiphenomenalism. Let us suppose that was the debate. Dennett would be right to point out that Block’s defense depends on the denial of functionalism and thus is question-begging. Likewise, my version of Block’s argument depends on the denial of functionalism with respect to consciousness. But what is problematic for Block is permissible in the present context. My dispute with Dennett differs from Block’s debate with Shoemaker in an important way. Dennett’s challenge does not require me to show that functionally identical zombies are possible (as Block was trying to show that absent qualia are possible). Rather,
233 Dennett’s Challenge
Dennett’s challenge is to show that there is a coherent way of construing zombies that does not entail that consciousness is epiphenomenal. And that I have done: the identity theory is an account of consciousness that allows the possibility of functionally identical zombies while ensuring that consciousness is not epiphenomenal. Remember, Dennett presents the zombie defender with a dilemma: Either the concept of a functionally identical zombie is selfcontradictory, or else it involves a conception of consciousness as epiphenomenal. I have not disputed the first horn of the dilemma. If conscious states are functional states, as Dennett seems to hold, then the notion of functional zombie—a thing functionally identical to you or I but lacking consciousness—is indeed incoherent. It is surely this combination that rightly elicits such strong reactions from Dennett. But, to borrow a phrase, those are not the zombies I am looking for. My strategy has been to attack the second horn of the zombie dilemma. On that horn, Dennett argues that a notion of consciousness that permits functionally identical zombies (that is, a conception on which functionally identical zombie are not incoherent) is epiphenomenalist. I have argued that functionally identical zombies can be coherently conceived of, and that their natural possibility does not entail that consciousness is epiphenomenal in the Huxley sense, that is, that it is a causal-role epiphenomenon. To do so I have adopted a different metaphysical stance than Dennett’s—namely, that of identity theory. This provisional assumption is not questionbegging with respect to Dennett’s zombie challenge; it is mandatory. How a functional state is realized, what the mechanism is (carburetor or fuel injector, engine or platypus) that occupies the definitive role, is important if consciousness is a state, property, process, or event. Against my defense of functionally identical zombies it may be objected that consciousness on Dennett’s view is not an occupant, but a role. In other words, if functionalism is the correct theory of conscious mental states, then functionally identical zombies are impossible. That is surely correct. But impossibility of functional zombies on the assumption of functionalism is irrelevant to the present dispute over zombies.
234 Chapter 7
Dennett argues that consciousness is a functional property. His argument is as follows: First, Dennett provides a sophisticated hypothesis about the mechanisms of our cognitive capacities—the Multiple Drafts model (Dennett 1991). Then he asserts that the Multiple Drafts model is a theory of consciousness: “I hereby declare that YES, my theory is a theory of consciousness. Anyone or anything that has such a virtual machine as its control system is conscious in the fullest sense, and is conscious because it has such a virtual machine” (1991: 281). Finally, Dennett provides a negative argument that the alternative view entails epiphenomenalism, namely, the two-part argument examined above. Dennett’s overall strategy thus depends on whether the negative argument goes through. This is why Dennett is such a vigorous critic of zombies. The challenge he sets for the zombie defender, the dilemma, is in fact crucial to his whole argument for the Multiple Drafts model in Consciousness Explained. Recall the dilemma. Either Dennett’s view is right—in which case functionally identical zombies are incoherent; or else functionally identical zombies are possible—in which case consciousness is epiphenomenal. To the first horn I have agreed. No one doubts that if consciousness is taken to be a functional property, then functionally identical zombies are not just naturally impossible, they are logically impossible. It is with the second horn of the dilemma that Dennett depends on the possibility of functionally identical zombies entailing epiphenomenalism. The claim is that any view that allows for functionally identical zombies is to be rejected because it will entail that consciousness is epiphenomenal in the Huxley sense. That is, it is on the opposing view—the one that does not treat functionally identical zombies as internally incoherent—that the possibility of functionally identical zombies is supposed to entail that consciousness is epiphenomenal. Therefore, for the purposes of this horn of the dilemma the consequences of Dennett’s own functional role view are irrelevant. Dennett’s argument depends on the claim that nonfunctional-role views of consciousness that allow for functionally identical zombies entail that consciousness is epiphenomenal. And that is what I have shown to be false. Specifically, identity theory allows for the possibility of functionally identical zombies without entailing that consciousness is a causal-role epiphenomenon (“the
235 Dennett’s Challenge
Huxley sense”), or that it is a strict metaphysical epiphenomenon (“the ridiculous sense”). Identity theory may have its drawbacks, but epiphenomenalism is not one of them. 6 Functionalism, Mechanism, and the Zombic Hunch Dennett’s zombie challenge is a dilemma. The dilemma depends on the claim that the alternative to Dennett’s view permits the possibility of functionally identical zombies—it does—and thereby entails that consciousness is epiphenomenal—it does not. Dennett remains unconvinced (2000, 2001). I locate the source of our disagreement in a confusion over functions and mechanisms. Dennett does not explicitly argue that all causal relations are functional relations, but it seems to be the idea behind some of his concerns. Consider, for example, Dennett’s formulation of the zombie issue in his “The Zombic Hunch: Extinction of an Intuition?” He writes, “that there is a real difference between a conscious person and a perfect zombie—let’s call that intuition the Zombic Hunch—leading . . . to the thesis of Zombism: that the fundamental flaw in any mechanistic theory of consciousness is that it cannot account for this important difference” (Dennett 2001: 36–37, italics in the original). And he reminds us: “Just remember, by definition, a zombie behaves indistinguishably from a conscious being—in all possible tests, including not only answers to questions [as in the Turing test] but psychophysical tests, neurophysiological tests—all tests that any ‘third-person’ science can devise” (Dennett 2001: 27, bracketed portion in original). Immediately we can see the trouble with Dennett’s formulations. Above I defended the possibility of functionally identical zombies. But I do not accept Dennett’s thesis of Zombism, and I do not believe that functionally identical zombies would be indistinguishable from conscious creatures in all possible scientific tests. This combination is impossible on Dennett’s view. How can it be that I believe in zombies but reject Dennett’s Zombism and the Zombic Hunch? The answer is that I am distinguishing between functional entities and causal mechanisms. From the fact that the difference between conscious and nonconscious creatures is not functional, it does not follow that the difference is
236 Chapter 7
not causal/mechanical. Indeed, if an identity theory is correct then of course we will be able to see, from a neurophysiological or experimental point of view, that there are differences between human beings and zombies. This could fail to be the case only if we thought that all causal properties are functional properties; then it would seem that claiming that there is some nonfunctional causal property is absurd. But that would be to confuse functions with causes, to confuse functionalism and mechanism. If it is not the case that every causal difference is a functional difference, then from the fact that two things are functionally indistinguishable it does not follow that they are not causal/mechanically distinguishable. Indeed they must be so distinguishable. It is a simple consequence of the functionalist abstraction that, for any given functional specification, there will be causal differences between items that equally share that specification, that is, that are functionally identical with respect to that specification. That is the observation behind multiple realizability arguments for functionalism. What leads Dennett to confuse functionalism and mechanism? Dennett’s phobia about zombies offers some hints. In his response to the arguments in the previous sections of this chapter, Dennett writes, Polger says that I make a mistake in declaring in such circumstances that carburetors are causal-role (or Huxley) epiphenomena. As he says, carburetors are crucial functional mechanisms in the cars that have them, and fuel injectors are likewise. But this misses the point of Huxley epiphenomena. If the difference between having one of these and having the other is not functional, then it—the difference—is epiphenomenal in Huxley’s sense. It is not a difference that makes a functional difference, but it is a difference that makes a causal difference. Just look under the hood: one reflects light entirely different from the light reflected by the other (and—to harken to Huxley’s case—if you listen really carefully, one gives off a whistle that the other doesn’t). . . . they are epiphenomenal in Huxley’s sense—since ex hypothesi their differences in causal powers don’t make a functional difference. (2000: 380)
Here Dennett at first seems to be making the distinction between causes/mechanisms and functions, complaining that I want to identify consciousness with mechanisms whose “differences in causal
237 Dennett’s Challenge
powers don’t make a functional difference” and that this thereby has “the cost of removing consciousness from the sphere of human interest (ibid.: 381). I find this claim confusing, because Dennett and I agree that even if consciousness is a causal-role epiphenomenon there is still no problem for naturalism (Dennett 1991: 404, and discussed above in section 3). Epiphenomalism is a problem for naturalism only if it is epiphenomenalism of the strict metaphysical sort. But then Dennett introduces a different idea, suggesting that differences that are not functional differences lie outside the sphere of human interest. I can understand why a naturalistically minded philosopher like Dennett might think that all differences must be causal/mechanical differences. But why must they all be functional, unless of course all causal differences are ipso facto functional differences? That, in my view, would be to conflate functionalism and mechanism. Dennett also thinks that an identity theory would not be able to answer certain moral questions about consciousness: “Why should it be immoral to dismantle a conscious person without permission (and without anesthesia) but not immoral to dismantle a zombie without permission?” (2000: 381). The correct response is threefold. First, I don’t know whether moral status attaches to consciousness in the way that Dennett suggests, or not. Perhaps it is just as immoral to dismantle certain kinds of nonconscious systems as it is to dismantle conscious systems. Certainly I think we have moral obligations to some things (e.g., the environment, perhaps society as a whole) that are not conscious beings. Second, insofar as moral status attaches to consciousness, it does so regardless of our theory of the metaphysics of consciousness. Dennett’s idea that moral status should depend on functional differences seems to be parasitic on his thinking that the difference between conscious and nonconscious creatures must be functional. Functional differences are important because they account for differences in consciousness. But the proposal on the table is that Dennett holds a mistaken theory about consciousness, and that he is therefore mistaken about functional differences being the morally salient differences. Even assuming that consciousness is what is morally important, it is no objection to the identity theory that it entails that the valuable characteristic (consciousness) is distributed
238 Chapter 7
differently than it is according to functionalism—unless there is independent reason to think that all morally salient differences are functional differences. And I do not think that we are forced to accept the latter thesis. Third, the driving force behind Dennett’s worries, I take it, is that we ought to be able to tell which things are conscious and which are not. It would be incoherent, or mysterious, or wrong to think that consciousness is important, for example, morally important, if we cannot tell which things are conscious and which are not. If there is a real difference between conscious creatures and nonconscious duplicates then it must be a difference that is at least in principle detectable. But moreover, Dennett seems to think that we’d have to be able to tell which things are conscious and which are not from a completely external perspective. He suggests the case of two types of creatures, conscious creatures (type A) and nonconscious creatures (type-B): There is no difficulty telling types A and B apart; half the beings have green brains that sworl to the left and make heavy use of acetylcholine, and half have red brains that sworl to the right and make heavy use of serotonin, etc. The trouble is that the labels, “A” and “B,” have been removed from our samples and mixed up. Our only task is to examine the samples and determine which are type A and which are type B. What do we look for? What is it about any non-functional causal difference you care to describe that could motivate us to decide that it is the difference that goes with consciousness rather than with unconsciousness? Notice that I am not playing verificationism here. I am not demanding “criteria”; I am asking for the minimum: something, anything, that would give somebody the slightest good reason for preferring the hypothesis that causal property k goes with consciousness, not unconsciousness. It will not do, of course, to see which set of non-functional causal properties most closely matches us, because we are not at this point entitled to any assumptions about how widespread consciousness might be among normal H. sapiens. (Dennett 2000: 382)
Now we can see Dennett’s conundrum. He wants to be able to solve the problem of other minds without the foothold of prior knowledge about some conscious creatures. He does not even grant that we know whether we or other human beings are conscious. I suppose that if one sets up the question in this way it will seem to be an insoluble problem. But it’s not a special problem for identity theory. This
239 Dennett’s Challenge
sort of problem lurks for functionalism and any other theory of mind. The hypothesis that functional property j goes with consciousness (not unconsciousness) will be equally inscrutable in this sort of setup. The problem arises for dualistic theories as well. If we don’t know which kinds of things have minds, then we don’t know whether it’s the ones with the res cogitans or without. I don’t think that it is a requirement on metaphysical accounts of mind that they be able to sort the conscious things from the nonconscious things in an a priori manner. I began with the assumption that I am conscious, and that other organisms like me (at least) are also conscious. If Dennett insists on wondering about the consciousness of other human beings, then he is engaged in a form of general skepticism. I do not see that we must set out to understand other minds by losing our own. I cannot for the life of me see how it could be more respectable to believe that there is a serious question about whether ordinary human beings are conscious than to believe in the possibility of harmless functional zombies. In part, I suppose that Dennett has failed to distinguish two ways that zombies figure in debates in philosophy of mind. There are some philosophers who believe, or seem to believe, in Dennett’s version of Zombism: Chalmers (1996a) is a clear example, along with Levine (2001) and maybe McGinn (1991). Perhaps Nagel (1974) and Kripke (1972/1980) are others. For those thinkers, the Zombic Hunch is just that—it is a hunch or intuition from which they begin their considerations. The considerations then take the form of conceivability arguments: Since (behaviorally or functionally or physically) identical zombies are possible . . . something or other follows about the metaphysics of consciousness. But that is not the form of my argument. My argument is not a conceivability argument. The distinction between my use of zombies and, for example, Chalmers’s, is important, and it is one that Dennett should accept. I do not begin with the possibility of zombies and then wonder whether or where consciousness fits into the picture. To the contrary, I begin with an idea of how consciousness fits into a naturalistic picture of the world—including, crucially, the causal powers of conscious states—and then I see what follows. Since I distinguish between functions and other causal/mechanical relations, I admit
240 Chapter 7
that other mechanisms could be “functionally identical” to the mechanisms that are conscious mental states in us. Identity theories locate consciousness at a certain order of abstraction, typically among neurophysiological states, events, processes, or properties. Two things could be identical at some other order of abstraction, for example, “functionally” in any sense relevant to Generic Functionalism, but fail to be identical with respect to the neurophysiology relevant for consciousness. Functionally identical zombies are possible. So what? Presumably there is some order of abstraction at which my current car, or any other car, is functionally identical to the Jaguar convertible that I covet. There are nevertheless differences between them that make a difference to me. Likewise, in some interesting sense, a bird’s wing, an insect’s wing, and an airplane wing are functionally alike. But what we care about when we get on an airplane is not that commonality but rather the differences. These differences may be mechanical even if they are not functional.
8 Minds, Brains, and Persons
According to the view I have been advocating, our conscious mental life is a biological phenomenon. Mental states are brain states. The mind–brain identities are species-specific, and they occur at many levels of mereological organization. I argued that identity theory can accommodate the varieties of multiple realizability worth wanting, and that it is compatible with the apparent contingency of the mind–brain relation. Conscious mental states, being identical to brain states, are robustly causal. They are just the sorts of items that figure in mechanistic explanations of mental capacities. All this is not to deny that identity theory may have some unusual features. In closing, I want to acknowledge a counterintuitive aspect of the identity theory. I’ll use that characteristic to point out two broad avenues for further development of the theory, one empirical and the other philosophical. 1 Sensations in Petri Dishes Hilary Putnam voices the following worry: If, for example, we say that the modules for visual “appearances” are in the visual cortex, which is where they have been hypothesized to be, then we run up against the fact that parts of the visual cortex (say, the parts that service one side of the visual field) can be dissociated from the “speech areas.” Are we to say that in such a case—a case of “blind sight,” or a case of “split brain”—that there are visual sense data (“appearances”) of which the person is not aware? And what would happen if our technology advanced to the point
242 Chapter 8
at which we could remove the “module” involved in the visual recognition of, say, chairs from the brain and keep it alive and functioning in a vat (in response to stimuli provided by a computer)? Would one then have “chair sense data” without any person to experience those sense data? If there can be “sense data” (or “appearances”) in a small group of neurons, why should be we not speak of the “sense data” of a thermostat? This way madness lies. (1999: 30–31)
This is a somewhat complicated passage to interpret, embedded as it is in Putnam’s late-1990s project of rejecting any “identity” theory of mind, which for him includes both identity theory and functionalism. But the basic idea behind Putnam’s example is clear enough: If sensations are brain states, then it seems we must face the possibility that the brain states can be removed from the brain and kept in a vat, a petri dish, Hinkfuss’s pail, or some other suitable receptacle. It seems that I might not be able to rule out such a possibility. Even so, I don’t think it presents any serious problems for my view. First, it is not at all clear that functionalism differs on this matter. Putnam doesn’t think that it does. Certainly the sort of teleofunctionalism that takes brains to be functional devices like hearts and livers must allow the same consequence. If hearts and kidneys can be removed and kept in vats, as surely they can, then so too can brains and parts of brains. Far from an absurd or embarrassing consequence of identity theory, the idea that some mental states can be disconnected from others by disconnecting some brain states from others is perfectly respectable. It seems to be, for example, a general principle of clinical neurology—perhaps the only principle. The surgeon cuts through to the brain and removes or oblates the offending brain tissue. Moreover, insofar as we regard the possibility of a sensation in a petri dish as bizarre, let us not forget the strangeness of the scenario that we are being asked to imagine. Brains are very complicated. Even those who think that brains are informationally modular do not usually believe that there is complete spatial localization of the underlying mechanisms. Of course, there is localization of very simple capacities, for example, sensitivity to orientation of stimuli cast onto a certain area of the retina (Hurvich 1981). But anything
243 Minds, Brains, and Persons
as complex as a sensation is not likely to be so “clean cut.” I argued in chapter 2 that we do not know how to identify or individuate sensations and brain states. But we know enough about brains and brain tissue to suppose with confidence that brain states do not have crisp or even determinate boundaries. And if, as seems likely, brain states are also temporally distributed—spread across some milliseconds or seconds of time—then we have an even more unruly thing that we are to imagine in this petri dish. It seems to me that part of the bizarreness of imagining a sensation in a petri dish is imagining that we could somehow extract this unruly, vague, four-dimensional object from its surrounds, and—without destroying it—store it in a dish. That is a strange idea, I agree. It is strange in just the way that it would be odd to take seriously the idea of a holding a tempest in a teapot, or extracting intact just the veins from a slab of marble. This just goes to show that if you imagine a crazy scenario you will imagine some crazy consequences. 2 Sensations in Brains But there is another reason, a more serious reason, for Putnam’s concerns. As long as, even in principle, it makes sense to talk about a sensation that has been “dissociated” from the rest of the brain, then it will be appropriate to inquire about whose sensation it is. If there is a sensation of yellow in a petri dish, then who is having the yellow sensation?1 Not, we assume, the dish. The simplest answer is that the sensation is its own subject. If there is a brain state in a petri dish that is identical to a sensation, then the brain state itself is the subject that is having the sensation. It is, presumably, an extremely simple subject, far simpler than you or I. But it is a conscious entity even so. This, of course, is the madness that Putnam warned of. But remember that we are talking about a very complicated unruly thing in this dish. We must cure ourselves of Leibniz’s image that we might find in our heads a familiar device of wheels, levers, and pulleys. In this sense, brains are quite unlike cars and brain states quite unlike carburetors. We are not imagining a gear in a petri dish, we are imagining a state or process occurring in bit of living tissue probably more
244 Chapter 8
complicated than the entire nervous system of most small invertebrates. Is it really so odd that such a thing could be a sensation? Consider this image: “What we now know is that each of us is an assemblage of trillions of cells, of thousands of different sorts. Most of the cells that compose your body are descendants of the egg and sperm cell whose union started you (there are also millions of hitchhikers from thousands of different lineages stowed away in your body), and, to put it vividly and bluntly, not a single one of the cells that compose you knows who you are, or cares” (Dennett 2001: 28). Dennett’s picture of a human being, of a human person, does not differ from my own in this respect. We both view human beings as bundles, to use Hume’s word, of cells. Brains are composed of cells; they are mereological wholes. Brains have parts. But the parts of a mechanistic explanation of brains will not be nice wedges or blocks of tissue. The parts, I have been suggesting, are probably complicated, vague, unruly “assemblages” of cells. If such an assemblage should become “dissociated,” or set out on its own, it would constitute a mental thing in its own right. It would be a simple mental thing, to be sure. But it could be a sensation. The question of how some state could come to be “dissociated” or disconnected from a complex brain like ours is interesting and substantial. It is the flip side of the question of how mental states are, normally in human beings, connected to one another. For a philosopher of my leaning, this comes to the question of how sensations are bundled together. Putting my point in this old-fashioned way, I might seem to have suddenly abandoned my sensibilities. But it seems to me that this question (the question about the bundling of sensations, which are identified with brain states or properties) is just a philosophical gloss of the project that neuroscientists of many sorts are engaged in. It is just—just!—the question of how the brain works. If the identity theory is right, then we can study neurological disconnections and disorders to gain understanding of psychological dissociations and deficits.2 We can, that is, pursue neuroscience as a methodology for understanding the conscious mind. If sensations can become dissociated, they can probably become dissociated without the radical extraction of the petri dish scenario.
245 Minds, Brains, and Persons
I imagine that there are numerous ways neural connections can fail without physical detachment. There could be, as Putnam fears, sensations in my body that are not felt as sensations by me. As Putnam hints, this is one way of interpreting blindsight and split-brain cases. This is to say that what constitutes me as a psychologically unified individual according to the usual platitudes may not be what constitutes me as an individual in another sense (Wilson 1999). My body may, in some cases, fail to be a unified psychological individual. The bacteria in my gut are part of me in one sense, but not in others; they are crucial to the proper operation of my digestive system, but they do not contribute genetic material to my offspring, and it doesn’t look like they contribute to my conscious states. Perhaps we could say that the bacteria in my gut are part of my “digestive individual” but not part of my “genetic individual.” There may be a pain in my foot that is part of my body but that has become “disconnected” from the rest of the bundle that is my mind. There is nothing any more mysterious about this disconnection than about the normal connection; it is entirely mechanical. I might even desire the disconnection, for example, if the pain were quite severe. I take it that some “pain killers” have just such an effect (Melzack and Wall 1965, 1988). 3 Sensations and Persons It seems comfortable to say that identity theory is a neo-Humean theory: The mind is a bundle of conscious mental states. Of course, it is Humean materialism. Humean materialists do not face Hume’s problem; we should not think of sensations as distinct and unrelated as Hume did.3 Mental states, on the naturalistic picture, have real relations between them. They are bundled into mechanisms by their causal connections. But if I look inside myself I find, as Hume notes, only particular sensations. I am suggesting that we return to the old idea that the mind and the self are one and combine it with the new idea that the mind–self is a biological phenomenon. This proposal may be useful in thinking about traditional problems of persistence of and knowledge of the self.
246 Chapter 8
The identity theory complements the theories of personal identity that Owen Flanagan (e.g., 1992, 1996, 2000) and Daniel Dennett (e.g., 1991) have been advocating for some years now. On this view, there is no single unified self that inhabits our bodies. The self, rather, is a “center of narrative gravity” (Dennett 1991; Flanagan 1992): Framed in this way, the seemingly radical idea that “the thoughts themselves are the thinkers” makes utter sense. As the stream flows and accrues a history, various models, including an ever changing model of the self, become part of the dynamic dispositional structure of the brain. We are brainy creatures who meet experience thinking. (Flanagan 1992: 188)
We are, in short, conscious organisms.
Notes
Introduction 1. Searle adds a notorious twist when he says that mental states are “caused by” brain processes, rather than the more standard claim that they are supervenient on or realized by brain processes. Interpreting Searle is a tricky business. But for my present purposes, he may be understood as claiming merely that mental events are caused by physical (brain) events. This is an ordinary (though not of course uncontroversial) endorsement of mind–body (psychophysical) causation. It is a further question whether to take Searle to be asserting that mental phenomena are somehow byproducts of biological processes, like bile (e.g., Heil 1992: 125–35). 2. James describes the experience of a baby this way in his Principles of Psychology (1890/1950: 488), and he seems to extend the description to adults in Some Problems of Philosophy (1911/1999: 32). 3. About 1 to 2 percent, according to Somorin (1973). See also Graham (1978). Special thanks to Todd Roach for these crucial data. 4. On the picture being suggested here, naturalism seems to have two components: methodological metanaturalism and substantial theoretical naturalism. The natural method is a metaprinciple. It recommends that we take all the information we can get and try to reach reflective equilibrium. As such it is nearly trivial; it merely recommends inference to the best explanation to answer questions about the world, in particular about minds and consciousness. But as soon as we start to say what information is submitted to the method (e.g., the specific claims of physics and biology) and what is not (e.g., the specific claims of fables and science-fiction stories), we begin to craft a nontrivial thesis. On this picture, the existence of ghosts does not impugn the nearly trivial metaprinciple, but it shows that some of the contributing disciplines (e.g., physics and biology as we know them) are incorrect. (I assume that current physics and biology are incompatible with the existence of ghosts of the usual superstition and horror-movie sorts.) The idea is that metanaturalism is trivial or nearly trivial, but that it is a substantial and nontrivial claim that current naturalistic theorizing (i.e., in physics and biology, and by extension current naturalism in philosophy of mind) are metanatural. This way of thinking about naturalism is similar to some interpretations of the principle of natural selection (PNS) in biology that take
248 Notes
PNS itself to be a trivial theorem of statistics but take it as nontrivial that some particular processes are instances of PNS (see Brandon 1990). 5. Because the natural method presupposes that physics and biology are natural sciences, it implies that logic and mathematics are naturalistic. In particular, if there are abstract and mathematical objects, they do not violate the No Ghosts Rule. (If abstract and mathematical objects exist, they are not made or composed of a substance that is not material or physical.) Why? Physics, biology, and other natural sciences appear to make use of notions of logic and mathematics, including, for example, sets and counterfactual conditionals. Physics and biology are naturalistic, and they presuppose logic and mathematics. It follows that logic and mathematics are compatible with naturalism. 6. Nevertheless, it is worthwhile to mention the sorts of reasons for doubting Brentano’s thesis. Some considerations are: (a) the shortage of positive arguments for Brentano’s thesis; (b) the hubris of taking distinctively human kinds of mentality as paradigmatic; (c) the possibility of nonintentional conscious mental states; (d) the ubiquity of intentional or quasi-intentional systems that are not minds; (e) the problems of explaining introspective discrimination of one’s own intentional states; and (f ) the puzzles that externalism about content creates (e.g., for mental causation) when mental states are individuated intentionally. To develop each of these lines of thought thoroughly is a project for another day. 7. Even this standard story, that philosophers stopped using the term “conscious,” is questionable. For example, Hilary Putnam, in his functionalist papers of the early 1960s, repeatedly considers the question of whether machines can be conscious (see, e.g., those reprinted in Putnam 1975c). 8. One is Berent Enç (1986: 425 n. 54); and J. J. C. Smart (1959) might hold the view, concerning which see Lycan (1987: 20).
Chapter 1 1. While I prefer the term “naturalism,” unless otherwise noted I use it interchangeably with “physicalism” and “materialism.” Some philosophers take physicalism to be the doctrine of materialism about the mind (Flanagan 1991), where materialism is the general view that what there is and all there is material. Ned Block (1980c) uses “physicalism” to refer to identity theory. David Armstrong (1968) calls his theory of mind “materialism.” Hilary Putnam continues to call both functionalism and type-identity theory “identity theories” (1988, 1999), as does Cynthia MacDonald (1989). Marian David (1997) cautions about the inclusive use of the category “identity theory.” I use the name “identity theory” to refer to the mind–brain type-identity theory, according to which brain state types are physical types rather than, e.g., functional types. 2. An exception is Thomas Nagel, who writes, “Physicalism is a position that we cannot understand because we do not at present have any conception of how it might be true” (1974: 176). Most nonphysicalist theories nevertheless operate within the framework of mechanistic philosophy of mind. (This tacit agreement may be problematic. Gilbert Ryle (1949) argued that materialism was no alternative to dualism because it accepted the Cartesian construction of the mind–body problem. Ryle’s
249 Notes
point and similar remarks from Ludwig Wittgenstein and Wilfrid Sellars inspire an amorphous alliance of currently fashionable “nonrepresentational” critiques of contemporary philosophy of mind from, e.g., McDowell 1994 and Putnam 1994b, 1999.) 3. Behaviorism is of course mechanistic, but it rejects treating the mind itself as a mechanism. Descartes treated the mind as a mechanism on analogy with material mechanisms; but his theory is undermined by the difficulty of explaining how an immaterial “mechanism” could mediate bodily actions. 4. A question about which I will have little to say. I will often talk of properties or states depending on the convenience and style. It is fair to understand me always as referring to states, processes, events, or processes—whatever combination of which turns out to be the best way to talk about conscious experience. 5. See Akiba (2000). 6. Christopher Hill (1991) explores simplicity arguments in favor of identity theory. 7. Although the concept of multiple realizability is credited to Putnam in his functionalist essays, especially “The Nature of Mental States” (1967), the expression seems to originate in Lewis (1970). Thanks to a number of people for helping search out the origins of the term, especially David Sanford, Bill Lycan, and Dan Ryder. 8. See also Richardson (1979), Enç (1983, 1986), and Churchland (1986); but also Endicott (1989, 1993) and Clapp (2001). 9. SETI is the acronym for the Search for Extraterrestrial Intelligence, the NASA project directed at discovering signs of intelligent life elsewhere in the universe. 10. Ned Block introduces what he calls the Disney Principle: “In Walt Disney movies, teacups think and talk, but in the real world, anything that can do those things needs more structure than a teacup” (1997: 120). The Disney Principle is plausible and does not seem to require any particular theory about the nature of mind; but it does not draw a line between standard MR and radical MR. Similarly, Bob Richardson notes, “Any functionalist view accepts some restriction on potential realizations. There is at least the minimal demand that the system which realizes the type be of sufficient complexity” (1979: 535). 11. But we can focus our attention on the theory rather than the exploits, as when we use science-fiction novels and movies to illustrate or teach philosophical theories. We may ask: What sort of theory of mind is implicit in the portrayal of the android Data, on Star Trek? Is Data portrayed consistently? And so forth. 12. Here I am indebted to Chris Gauker, who urged me to elaborate my reasoning on this point. 13. If properties are abundant, then the Kim–Adams reply may seem too easy. Yet the properties shared must ground the regularities of psychological explanation and thus be explanatorily useful, as Shapiro (2000) also argues. This is not much of a constraint, but it is a start. (If properties are sparse, then the functionalist must show that functional properties will make the cut.) We’re in no position to say just what those properties are—they are whatever properties our brains have in virtue of which
250 Notes
we have conscious mental states. Bickle (2003), for example, believes that the relevant properties are those studied by molecular neurobiology. But what is disputed by functionalists and identity theorists is the structure of the properties that are relevant (functional, type-physical, type-biological, etc.) rather than the specific catalog of properties. 14. But see Brian Keeley’s “Making Sense of the Senses” (2002). Keeley also raises the prospect that some human beings have sensory modalities that have not been previously recognized, for example, a vomeronasal sense. 15. Also, there is reason to think that dolphins, like human beings and unlike bats, engage in sophisticated intelligent social behaviors. This fact has not escaped the notice of philosophers (see MacIntyre 1999, ch. 3, passim). 16. See Thompson (1995) and Watkins (1999). 17. For convenience I shall follow the practice of pretending that the “species” are roughly phylogenetic species. But the species-specificity needed by identity theory may turn out to be more general than biological species, or more specific still. This is related to the issue of “grain,” discussed by Bechtel and Mundale (1999), Bechtel and McCauley (1999), and Graham and Horgan (2002). We will come to this in a moment. 18. But for further considerations see Endicott (1993), Bolender (1995), Jaworski (2002). 19. Ned Block offers: “For example, the central nervous system is often supposed to be quite plastic (though recent results appear to challenge this long-held view), especially in the young. If the brain is injured, knocking out some capacities, the capacities often reappear with different neural realizations. And there are many states and capacities that are known to be implemented differently in different people, e.g., the ability to read” (1997: 109). For more on plasticity see Buonomano and Merzenich (1998), Das (1997), and Ramachandran (1998). 20. Fuel injectors are multiply realized, too, of course; but carburetors are not fuel injectors. One difference is that the pressure that sprays the fuel into the combustion chamber is provided by a pump in the case of fuel injectors, rather than atmospheric pressure as in carburetors. 21. Talk about granularity, which we will see again with respect to Graham and Horgan (2002), is perennial in philosophy of mind. For example, concerning the possibility of mind–brain identity, Flanagan writes, “A lot depends on how strict and fine-grained we demand the type-identities to be, and that depends largely on considerations relative to particular interests” (1992: 47). Other examples include Tye (1995, e.g., 172–173) and Heil (1992, e.g., 168). 22. Note that Graham and Horgan’s grain problem is different from Wilfrid Sellars’s. Sellars (1956) contended that it was a problem for materialism that physical objects have a granularity whereas sensations are homogenous and without grain. 23. For reservations about the Bechtel and Mundale examples, but not their conclusion, see Sungsu Kim (2002).
251 Notes
24. But Bechtel and McCauley (1999) make the stronger argument that a version of identity theory is supported by the evidence, particularly metascientific evidence about the practices of neuroscience. 25. It is an open question just how to characterize these generalizations if, as seems likely, there are no laws of nature in the classical sense. The claim that I am endorsing and that Davidson denies is that there are generalizations about mental states that are of the same sort as generalizations about other natural phenomena. Although philosophers of mind are sometimes chastised for cavalierly and naively invoking laws or lawlike regularities, it is not our special burden to say what will do the explanatory work that laws did in classical theories. Or, perhaps more to the point, it has yet to be shown that this will be a special or unique problem for philosophy of mind. 26. I do not suppose that there is only one way in which states may be physically alike or different. Physical and psychological similarity may each extend in multiple ways. 27. See John Bickle’s Psychoneural Reduction: The New Wave (1998) for a detailed response to this application of multiple realizability. See also Bickle (2003). 28. Of course what I would like to call these is “alien” qualia, but Ned Block (1978) uses that term to talk about qualia different from our own that are had, specifically, by creatures just like us in some way. William Lycan (1996) coins the term “Strange Qualia” to refer to nonintentional qualia; then he goes on to talk about “New Strange Qualia.” So we are running out of convenient terms to conscript into the business of referring to qualia significantly different from our own. 29. Functionalists identify psychological kinds with functional kinds, which probably explains why Dennett sometimes resists calling himself a functionalist. He sometimes seems to think that psychological kinds are correlated with functional kinds, and that brute correlations are to be preferred over identities. 30. The multiple realizability argument is supposed to show that identity theory is wrong and therefore to pave the way for functionalism. Sometimes it is said that multiple realizability exposes the fact that identity theory holds that psychological kinds are essentially biological kinds. And, as we all know, belief in essences is very wicked. Functionalism, however, is no different as far as commitment to essentialism. Whereas identity theory holds that psychological kinds are essentially physical kinds, viz., biological kinds, functionalism holds that psychological kinds are essentially functional kinds. Why should functional essences be preferred to biological essences, at least on the grounds of any scruples about essences? Moreover, by being committed to the identification of psychological kinds and functional kinds, the functionalist attributes a common functional property to all realizations of a psychological kind. If it is possible that all instances of a psychological kind share a functional property, then why is it not possible that all instances of a psychological kind share a physical property? (That all instances of psychological kinds have properties in common is just what the Kim–Adams reply maintains.) Of course, there might be reasons for thinking that psychological states have functional rather than physical properties in common. But those arguments stem from the particular claims of functionalism, not from the claim of multiple realizability alone. 31. See also Stoljar (2000) and Braddon-Mitchell (2003).
252 Notes
Chapter 2 1. Another argument based on multiple realizability, due to Jerry Fodor (1974), aims at a stronger conclusion. But this argument plays a different dialectical role than the standard multiple realizability arguments. See chapter 6. 2. Fred Feldman (1973) makes a related point. For an entirely different approach that also finds Kripke guilty of question-begging, see Olav Gjelsvik’s (1987) ingenious argument that Kripke is not, on his own view about puzzle cases of rigid designation, entitled to his antimaterialist conclusions. 3. Advocates of this response are too numerous to survey, including (at least implicitly) every metaphysical functionalist; see chapter 3. For discussion of many available responses to Kripke, see Boyd (1980) and Lycan (1974a, 1987). Here I do not question the standard presentation according to which functional terms are typically used to refer nonrigidly; but it is worth noting that functional terms, like any others, can be used rigidly. Similarly, granting rigidity, one could nevertheless challenge the necessity of identity (Gibbard 1975). 4. Others may include Janet Levin (1985, 1991), and Frank Jackson (unpublished) now that he is a materialist. 5. Christopher Hill (1981) has offered, if I understand correctly, what is essentially an existence proof for a materialist explanation of the apparent contingency. But Hill does not provide a particular model, as I do. 6. Here, especially, I want to be clear that I may not be in complete agreement with Sidelle’s analysis. He argues that although scientific identities are empirically discovered, the conditions of identity must be known a priori. On that view, it is proper to say that knowledge of identity conditions must come before knowledge of identity. My use of Sidelle’s notions of candidacy and of identity conditions does not depend on his analysis of them. In particular, Sidelle (1989) introduces these ideas and then argues that they are a priori, consistent with his nominalism. I do not believe that my argument rests on this question of a prioricity—which is in fact Sidelle’s central concern in the paper from which I am drawing. 7. The identity conditions for a (kind of ) thing specify what features it must have to be a particular (kind of ) thing. Identity conditions ipso facto specify when a (kind of ) thing is distinct from other (kinds of ) things; conditions of identity and conditions of individuation are two sides of the same coin. Hence my uses of both “conditions of identity” and “conditions of individuation.” 8. Even then you might have some lingering doubts if you did not know that you (in fact) had learned the complete identity conditions for a thing or kind. 9. I am arguing that there is a model that can explain away the apparent contingency that remains in cases in which the criteria of identity that we use are the correct but generally specified identity conditions. If, as is likely, we do not use the conditions of identity for things as the criteria for identifying or individuating them, then this only helps my cause. If our criteria of identity are wrong or—if there is a difference—not the conditions of identity, then so much greater will be the sense of contingency.
253 Notes
10. You might think—perhaps you would also think this of games—that chairs do not have well-defined conditions of identity. This is beside the point for my purposes. I am concerned with failures that depend on not knowing (or not knowing enough about) conditions of identity. If identity conditions are vague or uncertain, that might be yet another source of apparent contingency. But this is different from the apparent contingency that results from not knowing how to individuate the items at hand. 11. Indeed Kripke’s epistemic model is plausibly just a special case of my candidacy model. I have emphasized that one might be in ignorance of the identity conditions for a thing, and so not know that it was a candidate for identification. But another way that you could fail to know whether something is a candidate for identification would be to confuse it (and its identity conditions) with something else superficially similar (and its respective identity conditions). So the candidacy model can explain why Kripke’s epistemic model works, as well. 12. Of course the availability of background theories—like molecular chemistry— that put particular facts in context is relevant (see Polger and Flanagan 1999). Whether background theories always or sometimes determine identity conditions is an interesting question that will not be pursued here. 13. Shoemaker (1996) remarks that C-fiber firings never were good candidates for identification with pains. That the particular example is false only helps to explain why it seems that it could fail be the case—it isn’t! 14. For a thorough discussion of the introspectionist program, see Güzeldere (1995b), and his introduction to Block, Flanagan, and Güzeldere (1997). 15. Whether it is best to describe such a case as hearing the sounds differently or as judging differently the same sounds one always heard is curious, but immaterial to my argument. All that matters for my purposes is that one can come to make finer discriminations in sounds than one previously made. 16. See Güven Güzeldere’s introduction to Block, Flanagan, and Güzeldere (1997). 17. So I am not only claiming that we do not know that the identity conditions for sensations and brain processes are different. I am also suggesting that at least some projects in philosophy and neuroscience are premised on the belief that some brain processes (whatever they are) will be found that have the same identity conditions as some cognitive processes (whatever they are.) Discussions of so-called isomorphism (e.g., Palmer 1999) or the neuron doctrine (Stoljar and Gold 1999) attempt to assess these projects. See also Bechtel and Mundale (1999) and Bechtel and McCauley (1999). 18. See Byrne and Hilbert (1997a,b) for an excellent collection of recent work on color. See also Clark (1993) and Thompson (1995). 19. Concerning pain, see Gustafson (1995, 1998), Hardcastle (1999a), Sufka and Lynch (2000), Sufka (2000), Sufka and Price (2002), Aydede (2000, 2001), and Aydede, Güzeldere, and Nakamura (forthcoming). Concerning dreams, see especially Flanagan (1995b, 1996, 2000).
254 Notes
20. I draw your attention to Shoemaker’s (1996, ch. 8) discussion of the firstpersonal aspect of Kripke’s argument. The candidacy model does not depend on whether the examples are considered from the first- or third-person point of view. 21. Some people do not know the difference between a martini, a Gibson, and a Manhattan. They believe that any drink served in an inverted conical glass—that is, in a martini glass—is a martini of some sort. It’s not that they cannot tell the difference, by ordinary criteria of sight and taste, between the cocktails. It’s just that they don’t know the identity conditions for martinis. A person who does not know the identity conditions for martinis might say things such as, “But a martini doesn’t have to be made with gin, it may be made with vodka . . . and it might have a small onion in it.” It seems to such a person that martinis are only contingently identical to a kind of drink made with gin. This is exactly the sort of mistake that the candidacy model is able to explain. 22. For related discussion, see Blumenfeld (1975) and Brand (1979). 23. For an elaboration of this style of reply, see George Bealer (1994). Bealer argues that Kripke has underestimated the resources available to the scientific essentialist who wants to maintain an identity theory. The candidacy model differs from the objections that he entertains on behalf of the identity theorist. My proposal is weaker than that considered (and rejected) by Bealer, because at this point I aim only trying to explain the appearance of contingency, not to establish the identity claim. Moreover, since my model is different from the two versions of Kripke’s epistemic model considered by Bealer, I believe that it can meet the repaired modal argument that Bealer favors. This is in part because my model explains the missing scientific essentialist intuition that opens the door for Bealer’s move. Considering the example of creatures that have (instead of C-fibers) Ct-fibers composed largely of X, Y, Z, and W, Bealer writes: “The point is that it would not be counterintuitive to say that they are in pain. Note the contrast. It would be counterintuitive to say that samples of XYZ on twin earth are samples of water” (1994: 199). On my view this intuition about pain is missing because, whereas we know how to identify chemical stuffs, we do not know how to identify sensations or brain processes. If that is right, then if one knew enough about sensations and brains one might well have the relevant scientific essentialist intuition to get the argument going. (Moreover, considerations of the previous chapter suggest that the intuition that creatures with Ct-fibers could be in pain may be an artifact of a mismatch in granularity between pain and Ct-fibers.) 24. Presumably Boyd’s is a tactic that many functionalists would endorse. The functionalist is committed to the realization relation being necessary. It is that aspect of functionalism that has paved the way for an industry of science-fiction examples about unlikely but functionally identical systems. Like the candidacy model, Boyd’s strategy claims that we are wrong in our intuition that a C-fiber firing could fail to be a pain, and it attributes this mistake to something that we cannot know about Cfiber firings just “by looking.” Boyd says that what we do not know “by looking” is the functional role that the C-fiber firing realizes; my model says what we do not know is how to individuate brain states, properties, processes, or events. Each of these strategies claims that the appearance of contingency is simply mistaken—not, as in Kripke’s model, a displaced true assessment of contingency. My candidacy model is preferable to Boyd’s rebuttal on this count because it explains both directions of the apparent contingency univocally. And, of course, Boyd’s functionalist explanation takes “pain” to be a nonrigid designator and is therefore not available to an identity theorist of my stripe.
255 Notes
25. Jaegwon Kim (1998) has recently endorsed functional specification as necessary to a plausible “reductive” program, as well. It is worth noting that functionalists have typically argued that functional specification is contrary to “reduction.” Notable exceptions are Armstrong (1968) and Lewis (1969), who are usually counted as identity theorists; they endorse functional specification, but take themselves to be “reductionists.” 26. Identity theory accepts that the mind–brain identities are brute, that it will be a brute fact that there is something it is like to be in thus-and-such brain state. But— unlike the view of Chalmers (1996a)—identity theory keeps the bruteness among the furniture of the natural world. So there are no new nonphysical properties or states, and mental states are causally potent. 27. Note the similarity of Levine’s explanatory desiderata and Sidelle’s candidacy criteria. Both use the language of “compatibility.” Levine writes, “An explanation is satisfying if it becomes intelligible why this event occurred, and not some other, why the object in question manifests this property, and not some other. In order for this explanatory feat to be accomplished, the alternative to be ruled-out must be incompatible with the explanans. . . . That is, we must be able to see why if the explanans is true then the alternative events could not have occurred” (1991: 38). Compare this to Sidelle’s remarks about candidacy, quoted above: “to be a candidate for identification with our subject of inquiry . . . something must have identity conditions which are compatible with those for our subject” (1992: 273). 28. This explanation differs from another strategy that has recently been considered by Levine (1997) and Van Gulick (1998). Van Gulick responds to David Chalmers’s (1996a) use of the explanatory gap by arguing that it is the inadequacy of our present concepts that causes the apparent gaps. If only we had the correct concepts of the mental and the physical, we would be able to close the explanatory gap. Van Gulick is also one of the most explicit advocates of the structure of experience approach. Perhaps, then, my call for understanding the identity conditions of sensations and brain processes is just a different way of urging the conceptual tidying-up that Van Gulick has had in mind all along? I do not think so. Van Gulick seems to be responding to Chalmers’s antimaterialist argument by using elements of the arguments made by Levine (1983) and McGinn (1991) to the effect that current materialism is explanatorily but not metaphysically inadequate. This leaves just the explanatory gap remaining, against which he then invokes the structure of experience reply. In contrast, my candidacy model uses conditions of identity against both the explanatory gap and the metaphysical divide. A final consideration against understanding Van Gulick or Levine as having identity conditions in mind as the solution to the explanatory gap is that they are both sympathetic to functionalism. Functionalists deny that sensations and brain processes have compatible identity conditions, i.e., that they are candidates for identification. This is part and parcel of multiple realizability. So it is unlikely that either Levine or Van Gulick was thinking that conditions of identity are the missing key to making mind–brain identity claims intelligible. 29. Levine concedes that the conceptual analysis may be a priori. He writes, “I think we have to recognize an a priori element in our justification. That is, what justifies us in basing the identification of water with H2O on the causal responsibility of H2O for the typical behavior of water is the fact that our very concept of water is of a substance that plays such-and-such a causal role” (1997: 550). He even notes that his view on this matter is related to views that have been expressed by Sidelle. (See Levine 1997: 554, n. 13, in reference to Sidelle 1989.)
256 Notes
Chapter 3 1. Here I am recognizing, for the purposes of this taxonomy, that some theorists take the class of mental states to be constituted by intentional states, rather than by conscious states as I have supposed. Some discussion of this difference will follow. 2. See, for example, Owen Flanagan’s entry on “behaviourism” in The Oxford Companion to Philosophy (1995a). 3. If there ever were any analytic behaviorists. The standard examples, Ludwig Wittgenstein and Gilbert Ryle, each denied that he held the doctrine. And David Armstrong’s (1968) amusing retort is not entirely satisfying as an explanation of either’s views. 4. Nor am I, by any stretch of the imagination, trying to include every theory that travels under the label “functionalism,” or even every so-named doctrine related to minds, brains, or psychology. I have already said that I will leave out some epistemological variations; but there are even more. Frank Jackson and Philip Pettit’s moral functionalism borrows directly from the doctrine in philosophy of mind (1995, 1996), but is not included in my discussion. Likewise for Michael Lynch’s recent (2001a) “functionalist” account of truth. Others may be related only in spirit—such as the views of nineteenth-century “Chicago Functionalists” (e.g., Angell 1907) in psychology, who were influenced by early American pragmatists ( James 1890/1981; Dewey 1896; see Roback 1952). Doctrines of functionalism in biology are surely related, but I will only scratch the surface. (On some views, e.g., Millikan 1993, we should ultimately view issues in philosophy of mind as a subset of those in biology.) Some kinds of “functionalism” may be less obviously related, such as those uses according to which a conceptual scheme is the “function” from language to the world (see Lynch 1998). If there is a “functionalism” in anthropology, or literary theory, or economics, it is not my concern. Still, some of what I have to say about functionalism in philosophy of mind could probably be said about any doctrine of functionalism. I leave the saying of it to someone else. 5. Fred Dretske is a clear example. He denies that he is a functionalist, because he uses “functionalism” to encompass only one of the variations that I consider. Yet it seems clear that Dretske’s theory is broadly functionalist in spirit; and I will argue in the next chapters that it is an exemplar of one style of functionalism. 6. Successfully so, to my mind. If there is a problem about materialism, I take it that Feigl, Smart, and Armstrong showed that it is not due to any fundamental conflict between mental and physical concepts or language. 7. Though on some versions of the story, Smart and Fodor also invented functionalism, or some version of it (see Block 1978, 1980c). 8. Please remember that, with this schema and with those I offer in subsequent sections, my goal is to express the basic idea behind each family of functionalist approaches. I hope that these formulations are easily recognizable by friend and foe alike as capturing the gist of what has been debated. But I do not claim that these schemas capture every nuance that any particular functionalist recognizes—much less every functionalist.
257 Notes
9. Although it was not his intent, I take Lycan (1990, ch. 2, sec. 5) as having provided a blueprint for my rehabilitated identity theory. The suggested way of distinguishing functionalism from identity theory is entirely derivative of Lycan’s discussion (1987: 18–21). The important distinction is between reference-fixing and sense-fixing (analytical or definitional) uses of functional specifications. I cast this distinction in terms of de re and de dicto functional specifications. This unusual way of talking (viz., of specifications or theories being de re or de dicto) introduces complications that might be avoided, but I believe that it can help us to see (in later sections of this chapter) how different versions of theoretical and explanatory functionalism relate to metaphysical functionalism, and where the metaphysical work is being done. It seems to me that it is being done not by functional specification itself but by whether or not such specification is treated as definitional. 10. It’s possible that something would change about the number nine (at least, its instances would change). But let’s not get into the philosophy of mathematics. The point of this stock example is clear enough. 11. The example needs some extra qualification because of the possibility of succession and the temporal nature of the status “being president.” For more on “being president” see Rey (1997: 176). 12. On the traditional way of understanding these questions, they are ordinary epistemological questions: Can we know about other minds? Can we gain self-knowledge by introspection? But a more interesting question for functionalism is that of when our theories and explanations posit other minds and introspective access. For this reason I take these to be theoretical or explanatory questions. 13. Such a theory is sometimes applied to public linguistic units—though not always in the same way or by the same people who advocate it for mental states. I do not discuss this part of the theory here because I am concentrating specifically on functionalism about the mental, and because many theorists take their public linguistic functionalism to be dependent on functionalism about the intentionality of mind. Where do questions about representations fit in? The answer depends on one’s approach to representation, and in particular on whether one takes representation to be built on intentionality or vice versa. For a thorough discussion see Fodor (1985 in 1990), Cummins (1983, 1989), and Dretske (1988, 1995). 14. But the narrow/wide issue arises also within intentional functionalism insofar as it is bound up with representational theories. It seems, for example, to be what is under discussion in Fodor’s (1985) and Field’s (1978) discussions of “monadic” or “orthographic accident” accounts of mental representation. 15. This seems partly to do with Dennett’s Rylean dislike for doctrines and “-isms.” Although he doesn’t reject other categories, he is somewhat evasive about whether he is a functionalist. (Perhaps, like Lewis, Dennett doesn’t know if he’s a functionalist.) Georges Rey (1997) groups Dennett with Ryle and Wittgenstein, attributing to them a version of functionalism that he calls “superficialism.” 16. For Millikan and Dretske, moreover, the state must be a sign or representation for the system itself. 17. It is common to reserve this term for the ahistorical and narrow content varieties only; but that, from my point of view, is to jump ahead to our next taxonomic principle, that of the kinds of functions invoked.
258 Notes
18. But there is great dispute over whether this is supposed to be propensity normativity (Armstrong and Lewis) or purposive normativity (Millikan and Dretske). 19. Acknowledging that there are issues between the two versions, Fodor (1985) nevertheless groups them together as versions of functional-role semantics. Cummins (1983) draws an additional distinction between functional-role semantics (which he seems to take in the narrow sense) and what he calls “conceptual-role semantics.” On conceptual-role semantics, the content of a state is determined by its role in mediating a system’s cognitive (rather than computational or causal) processes, such as inference. For my purposes, this would still count as a version of functionalist semantics. 20. In characterizing the views, I have invariably stepped on some toes; although there is general agreement that the intentional and semantic questions can be distinguished, there is no general agreement about how to characterize the difference. Nothing in my argument rests on how I draw the distinction between intentional and semantic functionalism. 21. Millikan’s (1984, 1993) view is quite similar in this respect. 22. Fodor rehearses three reasons for doubting that function-role semantics will work. First, it is likely that empirical constraints will underdetermine semantic content. Second, it does not yield a naturalistic account of the relation between propositions and the world. Third, it raises problems about the individuation of intentional states. (See Fodor 1985, in 1990: 24–28.) 23. Cummins, I take it, is rehearsing one argument that links semantic, intentional, and metaphysical functionalism. This is why I mention this passage regarding semantic functionalism. The argument is: If I am a thinking thing “in virtue of the computational structure of my brain” then my thoughts must be what they are “in virtue of their place in that structure.” Thoughts are individuated according to their semantic content. Therefore, the semantic content of my thoughts must be fixed in virtue of their place in the computational structure of my brain. So metaphysical functionalism entails semantic functionalism. This, I take it, is roughly the line of argument that Fodor (1986) is critiquing, as well. 24. In a troubling remark, Cummins confesses that what he is calling functional-role semantics is a position that no one actually holds: “it must be confessed that what I have been calling functional-role theories are really straw men; no one actually holds them” (1989: 122–23). Cummins must be interpreting the theory more narrowly than I have. 25. For present purposes I provide only a sketch of the technique. There are many clear discussions of Ramsification, including Block (1978), Kim (1996), and Rey (1997); for more on its intricacies, see Lewis (1970) and Shoemaker (1981b). 26. Lewis solves the problem by brute force: “Form the conjunction of these platitudes; or better, form a cluster of them—a disjunction of all the conjunctions of most of them. (That way it will not matter if a few are wrong)” (1972, in Block 1980a: 212). This is not very satisfying. Georges Rey (1997) proposes that rather than defining all terms together (the “holistic” version) we define smaller clusters; he calls this the “molecular” approach.
259 Notes
Rey has in mind the “homuncular” approach defended by Fodor (1968), Dennett (1978), and Lycan (1981, 1987). 27. Block (1978), Van Gulick (1982), Shoemaker (1981b), and Rey (1997) each discuss the pros and cons of different choices of theory to Ramsify. Rey (1997) regards the main options as reflecting different “meta-approaches” to functionalism. 28. Almost no one expressly uses Ramsification; it is merely a device for explaining functionalism and exhibiting some of its properties. Although many refer to it and almost every general text explains the procedure (Block 1978; Kim 1996; Rey 1997), few philosophers make use of the technique or its formalisms. David Lewis and Sydney Shoemaker are prominent exceptions. 29. But the latter nomenclature could cause some confusion. I think that those who use the expression “analytical functionalism” usually have in mind the tradition of analytical behaviorism and the traditional philosophical use of the term “analysis.” On that interpretation, analytical functionalism is so called because it offers an analysis of the meanings of mentalistic terms. But some—e.g., Shoemaker (1981b)— restrict the use of the label “analytical functionalism” to the variety that takes only the analytic and conceptual truths about psychology as the base theory to be Ramsified. Shoemaker calls the Lewis version (which is often cited as the canonical version of analytical functionalism) “common sense functionalism,” arguing that it seems to be “an unpromising view” (1981b, in 1984: 104). 30. Marian David (1997) considers a related problem with the quantificational form of Ramsified theories. The trouble is that Ramsification existentially quantifies over predicates and leaves the entities with (typically) universal quantification; but the functionalist needs the predicates to be universally quantified (for the metaphysical claim) and the objects only existentially quantified (to allow for multiple realizability). David is not optimistic that there is any simple remedy, noting, “it seems clear that there is no simple way of repairing this problem through alterations of the embedded theory (the functional role): no amount of finagling ‘R’ will make a claim of the form ‘($x)(y)(xRy)’ come out as a claim of the form (x)($y)(xRy)’” (1997: 145). 31. And insofar as cognitive psychology is metaphysically functionalist, it is likely to be so as a consequence of philosophical functionalism in its standard formulation, as represented by the influence of artificial intelligence, for example. 32. We will return in a moment to the question of whether we ought to accept this hypothesis about language. 33. But it is important to point out that not all functional explanations make use of what in the next chapter I will call, following current fashion, teleological notions of function. No doubt this will be the source of some confusion. 34. The etiological notion is the most favored way of thinking of evolutionarily tied functions these days, but it is not the only contender. Different notions of function will be discussed in the next chapter; the one that most goes against what I have been saying about the historically retrospective etiological explanation is Bigelow and Pargetter’s (1987) forward-looking account.
260 Notes
35. Millikan follows Pettit (1986) in using this terminology. Frank Jackson and Philip Pettit now advocate a related idea that they call “program explanation” (1990; see also Kim 1998: 72–77). 36. Grantham and Nichols (1999) distinguish two aspects of evolutionary psychology, explanatory and predictive. I am concerned with the explanatory aspect. 37. See Caplan (1978). 38. For a detailed critique of evolutionary psychology, see Bob Richardson’s Maladapted Psychology (in preparation). See also Grantham and Nichols (1999), Buller (1999) and Richardson (2001). 39. Cummins refers to other kinds of laws, including psycho-physical “instantiation” laws (1983: 6–7). I do not mean to imply that Cummins thinks there is no role for subsumption under causal law; I am merely trying to exhibit different styles of functionalist explanation. Cummins (1983) distinguishes transition theories (theories that explain changes) from property theories (theories that explain the instantiation of a property.) His insistence that psychology needs property explanations is probably stronger than it needs to be. The point is that undue focus on transition theories, and their appeal to causal laws, should not exclude alternative explanations, and alternative kinds of laws (1983: 7–9). In addition, as Cummins notes, some of the other styles of explanation “can be given deductive-nomological format, but the result is completely uninformative” (ibid.: 7). (Likewise, deductive-nomological explanation can be reconstructed in Cummins’s analytical paradigm. See ch. 6 for related points about functional and causal explanation.) 40. Amundson and Lauder (1994) provide a clear and influential discussion of the complementary use of both notions of function in the practices of biologists.
Chapter 4 1. There are no entries for “realization” or “instantiation” in the Blackwell Companion to Philosophy of Mind (Guttenplan 1994). The entry in the Oxford Companion to Philosophy (Honderich 1995) is three sentences. 2. One might hope that some one notion of function will win out over all others, thus leaving us with only one functional realization relation. But this seems not just optimistic but to mischaracterize the work that various notions of function do in philosophy of science, philosophy of mind, philosophy of language, and metaphysics, among others. The situation seems to call for recognizing the different explanatory uses of various notions of function (see, e.g., Amundson and Lauder 1994), rather than to seek a single kind of über-function. But the latter alternative will not be entirely ignored; I will return to critique it in chapter 6. 3. The same complaint applies to Block’s recent remark: “I keep speaking of realization. What is it? As Kim notes, we can think of realization this way. Suppose we have a family of interconnected macro-properties (e.g., mental properties or economic properties) of a given system (say a person). Suppose that corresponding to each of these macro properties there is a micro property in this system, and that the family of interconnected micro properties provides a mechanism for explaining the connections among members of the macro family. Then the micro properties realize
261 Notes
the macro properties” (1997: 108). Contra Block, Kim would object to this claim, since he denies that microproperties realize macroproperties (1998). 4. See also Cummins (1983) and Lycan (1987) for passing remarks. 5. See also Endicott (in preparation). For a more narrowly focused account, see Chalmers (1994). 6. Prima facie, the view of realization that I am proposing can do the work that Horgan and Tienson require of the relation. There is nothing in what they say that cannot be accommodated, though the transitivity requirement may need to be qualified. 7. I’m not sure whether Poland is best interpreted as holding that there are both one general and many specific realization relations, or whether, like me, he is speaking generally about realization but is ultimately committed only to the many specific realization relations. 8. Shoemaker thinks that this will help avoid problems raised for functionalism by Bealer (1997). Those issues we shall have to set aside for another day. 9. Of course, Gillett and Wilson needn’t welcome this rapprochement. If they are correct, then functional realization as I understand it is mistaken, for it fails to account for many cases of realization. The functionalist may try to turn the tables and adopt Gillett’s or Wilson’s account of realization and then formulate functionalism using whatever notion of function goes with it. The resulting position will resemble that advocated by Shoemaker and Fodor, according to which all causal properties are functional properties. I consider this view in chapter 6. 10. Heil (1992) insists that realization is a relation between token property instances. Gillett (2002) insists that realization is a relation between properties, but is less decisive on whether realization is a relation between property types or tokens. Poland (1994) allows for realization of properties or objects. To some extent I think this is cosmetic and that the various claims are translatable. I tend to speak of realizing function roles, properties, or states, choosing my terms largely for stylistic grounds. Nevertheless I should note the potential for controversy; and I will argue in the next chapter that in at least one case the malleability of realization and implementation locutions has been substantially misleading. 11. But as a matter of fact, I am skeptical that there is any one such general relation. I doubt that there is one relation that accounts for all instances of material composition. So I doubt that any relation covers material composition and also cases of functional realization of mental states, as well as whatever sort of realization is appropriate to political or sociological entities and properties. 12. See chapter 7 for discussion of kinds of epiphenomenalism. 13. The research of Weiskrantz (e.g., 1986, 1997) on so-called blindsight and Libet (e.g., 1985) on action potentials prior to voluntary movement has especially enthused philosophers. Milner’s and Goodale’s (1995) work on visual and motor pathways has also generated some interest. See Van Gulick (1990, 1994), Dennett (1991), Flanagan (1992), and Block (1995), among many others.
262 Notes
14. Horgan offers the following “partial explication” of the quausation relation (1989: 50): For any two events c and e and any two properties F and G, c qua F causes e qua G iff: (i) c causes e; (ii) c instantiates F; (iii) e instantiates G; and (iv) the fact that c instantiates F is explanatorily relevant to the fact that e occurs and instantiates G. Horgan (1989) then defends an interpretation of clause (iv) concerning explanatory relevance. 15. I will not for the moment worry about whether we should endorse a yet stronger causal claim, for example, that mental states have causal powers that are distinct from the causal powers of their realizers. This, I take it, would be a form of emergence. See Kim (1999) and Shoemaker (2002). Certainly emergence is not a widely accepted feature of functionalism; it would be far too strong as a basic requirement. 16. I am assuming that causation does not occur exclusively among mircrophysical entities. I suppose that bricks are causally efficacious in the strong sense, but I do not argue for this claim herein. If the functionalist can show that mental states can be causally potent in whatever way that bricks are, I will be satisfied for the moment. Jackson and Pettit (1990) seem to hold that common macroproperties are causally relevant but not causally efficacious, presumably including the property of being a brick. It’s unclear whether this should take the sting out of the deflation of the causal powers of the mental, or whether it merely compounds our worries about causation. 17. Louise Antony (1999b: 7) also discusses the concern that mental properties should be objective in a way that contrasts with conventionality and arbitrariness. 18. Dennett’s (1971, 1987) “intentional stance” may present one prominent exception. But even the intentional stance is not entirely open to interpretation: On Dennett’s view, there is a fact of the matter about whether it is useful to adopt the intentional stance for any given system.
Chapter 5 1. In asking what makes a particular state a conscious mental state I am not asking a “hard” question in Chalmers’s (1995, 1996a) sense. The “hard” problem is to explain why thus-and-such functional state kind is a mental state kind. The metaphysical functionalist maintains that the functional state kind and the mental state kind are one—they are identical; likewise, the identity theorist maintains that mental state kinds are identical to physico-biological state kinds. Functionalists and identity theorists can agree that Chalmers’s “hard” question is answered by an identity. (Of course it still makes sense to inquire about how things of a kind can manifest the combinations of properties in the way that they do, and that is a hard enough problem when it comes to minds and brains.) What functionalists and identity theorists disagree about is what constitutes mental state kinds. Functionalists identify mental state kinds with functional state kinds. In contrast, identity theory identifies mental kinds with brain state kinds, physico-biological kinds.
263 Notes
2. Even then describability will be a poor candidate. Whatever realizes a kind is an instance of a kind. What realizes a kind of description should be an instance of the description. But instances of descriptions are not, in general, describable by the descriptions of which they are instances. 3. Thus we salvage a great deal of the discourse about realization—including Lycan’s (1987) formulation of machine functionalism, now twice quoted. Similar charity is called for if we are to make sense of the common idea that realization is the relation between role and occupant, e.g., “if a first-order property exemplifies a functional role with respect to some system S, the property is said to play that role in S. Equivalently, the property is said to occupy the role in S, and to realize the role in S” (Hill 1991: 47). Something that occupies a functional role has a function. If the occupant–role relation is realization, then the occupant (a particular of some sort) is an instance of the kind picked out by the role. Either functional roles are themselves kinds of functions or else functional roles are relations and role-talk is shorthand for talk about the kinds of functions picked out by roles. The latter is to be preferred. 4. The relation between programs and functions is the relation between predicates and properties. A machine program is a predicate for the property of having a function, and having a function is a matter of having certain other properties, usually relational properties. 5. What kind of function? Probably causal, since it’s Ramsified. But we don’t know— it depends on what the mysterious theory H said. 6. For a thorough discussion, see Lycan (1974b). My discussion is heavily indebted to Lycan’s, and my terminology is derivative of his. 7. I follow Lycan in assuming that the alternative to the abstract functionalism is physicalist functionalism for the purposes of my discussion. But it is often pointed out that functionalism is at least compatible with nonphysicalist theories, such as dualism. If you prefer, follow Lycan’s suggestion and substitute “structural” for “physical” in the argument. (It is generally acknowledged that one could defend a dualistic version of metaphysical functionalism, but that there is little reason for doing so.) 8. In fact even mathematical functions are subject to far more constraints than are usually recognized in the free-wheeling world of functional ascription. But see Hilary Putnam (1988) and David Chalmers (1994, 1996b) regarding whether every rock implements every finite state abstract automaton. 9. A related point concerns the practice, common in the literature, of speaking about strong and weak functional equivalence. The idea seems to be that if two systems have the same external relations—inputs and outputs, say—then they are functionally equivalent in the weak sense; whereas if they share internal organization as well, then they are more strongly functionally equivalent. But this is absurd. It seems to make sense only when functional equivalence is left undefined. As soon as functional equivalence is given substance—in terms of a machine table or other functional specification—then the systems either are or are not functionally equivalent with respect to that specification. What varies is not the strength of the equivalence but the robustness of the functional specifications. 10. Brian Cantwell Smith has argued that computer programs do not operate as they are usually taken to; for example, they almost never literally manipulate symbols (Smith 1999).
264 Notes
11. In fact the situation is a bit more nuanced. One possibility is that we think of Turing machine programs strictly, as though they must literally govern tapes, tape heads, etc. with some serious limits. Although I do not want to underestimate the power of such a system, I submit that we know that minds are not literally Turing machines of that sort. (Specifically, we know that brains do not contain tapes, tape heads, and so forth.) If we view physical Turing machines in this narrow way, then they are definitely a special and limited case of causal-role functionalism. But we may just as well say that Turing machines should not be understood so literally and narrowly. In that case, rather than thinking of physical machine functionalism as a special case of causal-role functionalism, we might instead think of Turing machine tables as simply a special notation for recording causal-role explanations. As far as I know, every functionalist theory, in particular, Ramsified psychology, could be represented by a large enough machine table or long enough program. But in that case Turing machines do not provide substance to the explanatory framework, and we will still need to select the correct explanation (e.g., psychological theory) from which to construct the “program.” 12. In fairness, it is not required that we know which psychological theory to which mental states are relativized; it would be enough—however frustrating—if we could be confident about the existence of some privileged psychological system. 13. Lewis (1972) himself offers a different solution, about which see chapter 3, note 26. 14. One might restrict the psychological theory to only those analytic truths that are biologically abstract, but that would raise serious questions about the completeness of the theory and whether it yields unanchored (Rey 1997) or unmixed (Shoemaker 1981b) kinds. 15. Block (1978) provides an extensive list of such cases. Most subsequent examples have forms that can be found among Block’s objections. 16. For a more detailed and eloquent discussion, see Lycan (1987: ch. 2, sec. 5). 17. It is widely noted that the exclusion argument can be made in many different ways, and indeed Kim himself has offered several variations (Horgan 1997; Kim 1993, 1996, 1997, 1998; Bontly 2002; Gillett and Rives 2001; and see the essays about Kim’s work in Tomberlin 1997). 18. John Bickle (1996, 1998, 2003) is perhaps the most vocal proponent of the deflationary position. 19. See not only Horgan (1989, 2001), but also Bontly (2002) and Gillett and Rives (2001). 20. Although this proposal did not receive attention until much later, it was at least raised as a possibility. Fodor suggests in Psychological Explanation that, “it may be that only those effects of a biological structure that are adaptive, etc., count as its functions” (1968: xx). And Dennett makes some related remarks in Content and Consciousness (1969). 21. I coin a new term in order to reserve “teleological functionalism,” “teleofunctionalism,” and the like for the view, discussed in the next section, that is actually
265 Notes
held by the philosophers who identify themselves by that term. I borrow this use of the modifier “direct” from Davies (1994). 22. Millikan’s account of biological kinds is not without its detractors. Sober writes, “Biological states are not individuated in terms of their biological functions, since biologically distinct states can have identical biological functions. I expect no less of psychology” (1985: 190). This is ruled out if the biological functions are etiological as Millikan argues. 23. They also follow a variety of Brentano’s thesis and are thereby committed to direct teleofunctionalism as a metaphysical theory. 24. See Davies (2001) for a useful discussion of the options for combining accounts of function. Griffiths and Davies each offer more nuanced formulations than the simple hybrids that I consider in this chapter, but I do not see that their more complex variations avoid the general difficulties for the hybrids discussed here. This is not a problem for those accounts in their own terms, since Griffiths and Davies are interested mainly in functional explanation and thus do not take their combination functions as the basis for a distinctive ontology, as the functionalist in philosophy of mind does. Yet Griffiths’s and Davies’s views are surely where one should turn if the hybrid view is to be pursued. 25. For example, species themselves (Hull 1976). For a metaphysically minded view of biological individuals see Jack Wilson’s Biological Individuality (1999). 26. But for reservations on this count, see Enç (2002). 27. Bigelow and Pargetter (1987) find causal-role functions lacking on similar grounds; and they regard Cummins’s account as a form of eliminativism about functions, for it denies that functions are purposes or norms. 28. About which see Paul Griffiths (1993) and Peter Godfrey-Smith (1994), both reprinted in Allen, Bekoff, and Lauder (1998). 29. Hybrid teleofunctionalism is not to be confused with mixed (Shoemaker 1981b), or anchored (Rey 1997) versions of functionalism that are, say, functionalist about intentional states but not about qualia. Hybrid teleofunctionalism, as considered herein, aims to be functionalist in Van Gulick’s and Shoemaker’s strong sense. 30. Compare Armstrong’s (1968) and Lewis’s (1972) talk of “aptness.” 31. (H) might also be interpreted as the claim that mental states are causal-role functional states of systems that each have some teleological function or other, whatever it may be. But that claim is too open-ended to be what the hybrid teleofunctionalist has in mind. 32. (H¢) cannot account for malfunction, either. 33. Peter Godfrey-Smith calls teleological theories of meaning or representation success-linked. He writes, “The third type of success-linked theory of meaning I will look at is the indicator/success hybrid. An example is the theory of Dretske’s Explaining Behavior (1988). This theory makes use of both indication (law-based correlation) and function” (1996: 183).
266 Notes
34. Putnam also now thinks it was a mistake, but for different reasons. He believes that “no psychological theory individuates or ‘implicitly defines’ its states” in the way that functionalism requires (1994a: 480). I agree. But even if a psychological theory did individuate states thus, the realization relation for types named by that theory would be different from the realization relation for the types expressed by a machine table “theory.”
Chapter 6 1. Cited above (Shoemaker 1981b, in 1984: 274). For the moment we will suppose that this somehow eliminates the explanation-relativity that is explicit in Cummins’s (1975) notion. But in fact CTP-functionalism begins with a theory of functional properties and proceeds by transforming it. 2. Lycan admits that it is hard to see how atomic particles have any degree of teleological function (1987: 45). He wants a notion of function that comes in degrees; and he thinks that teleological functions, applied to his hierarchical “homuncular” view of nature, satisfy that requirement. I have doubts, but I will not pursue the matter here. 3. See Hill (1991) for an argument that identity theory and functionalism are one and the same, and that the result favors an identity theory. 4. It might be thought that Lycan’s rejection of Two-Levelism will help him here. But biological abstractness (multiple realizability) has turned out to be a question of orders rather than levels. No matter how many levels there are, and no matter how many of them count as biological to some agree, the functionalist must show that mental properties are higher-order properties. 5. Concerning the autonomy of the mental see, e.g., Richardson (1979), Horgan (1993), the essays by Horgan, Antony and Levine, and others in Tomberlin (1997), and Antony (1999a,b). It might seem that Antony no longer holds the autonomy thesis; she writes, “there are also non-reductive materialists—I count myself among them—who hold that mental properties are only irreducible in the ontological sense. We claim that there must be microphysical explanations of all non-basic properties and laws” (1999a: 37–38). But she still requires that a real mental property “figure essentially in non-analytically-guaranteed regularities” (ibid.: 42). See also Pereboom and Kornblith (1991), Pereboom (2002), and Braddon-Mitchell (2003). My strategy differs from theirs in that I urge that the status of mental states be divorced from questions of “reduction” and autonomy altogether, and I endorse the identity theory, which they do not. They seem to think that multiple realizability blocks identification, even if identity would not entail “reduction.” 6. We can see now that there are two arguments for functionalism that are based on multiple realizability: the generality argument discussed in chapter 1, and the autonomy argument currently under consideration. Moreover, both require Generic Functionalism. In the latter case, biological abstractness is necessary to secure the constructive portion of the autonomy argument, and if the resulting states are not causally efficacious they will not figure essentially in any autonomous regularities. 7. These two ways of formulating the pair come to the same thing for present purposes.
267 Notes
8. See, e.g., Salmon (1989). 9. For this reason the questions that motivate new wave “reductionism” may seem orthogonal to metaphysical questions about the mind. Typically this is of little concern to the proponents of new wave “reductionism” because they are interested only in explanatory claims to begin with, and they see preoccupation with metaphysical concerns as a sign that philosophy of mind is off course. 10. On that count I think that the advocates of “reduction” are correct; they are often eager to take this next step, and they criticize their autonomy-minded opponents for failing to take up the charge. If the crucial issue is “reduction” versus autonomy, then “reductionists” are right to take the battle to the empirical trenches. 11. For more on mechanism, see also Bechtel and Richardson (1993) and Glennan (1996). 12. I borrow the terminology of “ideally complete” explanation from Brandon (1990). 13. There is still a question about how the different sorts of explanations are related, but each can be correct without excluding the other. See, e.g., Skipper (1999). 14. In fact Putnam does not deny that we can make predictions about pegs and mental states from the microphysical or neurobiological perspective, but he denies that such predictions are explanatory. What he is surely correct about is that such an explanation would not be a psychological explanation of the usual sort. Perhaps we are now in a position to shed light on the motivations for autonomy: Could it be that Putnam wants unification explanations rather than causal/mechanical explanations? His remarks lend some credence to this reading (Putnam 1975b: 296–298). (Notice that Putnam makes a stronger claim about psychology than he does in his final assessment of the square peg example, where he settles for saying that the microphysical explanation is simply not a good explanation.) 15. The pluralistic explanatory strategy that I am urging bears a resemblance to the “compatibility” theses defended by Horgan (1993, 1997, 2001), Pereboom and Kornblith (1991), and Pereboom (2002). But each of them takes the explanatory compatibility to block “reduction” and thereby support an “antireductionist” or “nonreductionist” account of minds. They fail to appreciate that rejecting the strong autonomy claim allows us to avoid the “reduction” versus autonomy debate altogether. 16. My suggestion is in spirit with that offered by Brandon (1996) as a diagnosis of the debate between “reductionism” and holism in biology. 17. Whether Shoemaker or Lycan would accept this distinction between functionalism and mechanism I do not know; and, if they would, we can only guess as to whether they would continue to defend functionalism or join the mechanists. As a sheer matter of speculation, I suppose that Lycan is a determined functionalist and means to claim (in his wilder moments) that the world is functions all the way down in some teleological sense more robust than that of mere mechanistic role functions. Shoemaker, on the other hand, may be less committed to functionalism than he is to the causal theory of properties. 18. I owe the corkscrew example to Shapiro (2000).
268 Notes
19. If the functional explanation is not a causal/mechanical explanation, then it is compatible with the identity theorist’s mechanistic explanation. But in that case functionalism will not carry ontological commitment to functional kinds as causal kinds. So I take it that the functionalist intends to provide the basis for causal/mechanical explanation. 20. Another way of putting the point is to say that “reduction” is not compulsory. 21. Richardson (1979, 1982) defends what he calls the de facto autonomy of psychology, while criticizing the claims of Putnam and Fodor for a stronger (de jure, he says) autonomy thesis. Similarly, Horgan (1993) defends a moderate variety of autonomy of psychology. And Antony (1999a,b) seems to be softening her position. Yet each of these authors positions autonomy vis-à-vis “reduction,” and that is precisely the contrast that I am resisting.
Chapter 7 1. I will focus on in particular on Dennett’s arguments in Consciousness Explained and since (1991, 1994, 1995, 2000, 2001). 2. Güzeldere actually considers behavioral, functional, and physiological zombies. I shall consider zombies that are physically identical to any arbitrary degree of specificity. The purpose of stipulating zombies in this way is to allow that physically identical zombies can be a construct distinct from functionally identical zombies even if biology is functional. 3. David Chalmers (1996a) introduces a similar framework of distinctions. 4. Güzeldere (1995a) suggests that an equivalent, and more convenient, way of talking about the ways that zombies could be identical is by considering them as distinct kinds of zombies: behavioral zombies, functional zombies, and physical zombies. I will sometimes use this way of talking. So when I ask, say, whether behaviorally identical zombies are naturally possible, this is shorthand for question (Q1), “Is it naturally possible that there be zombies that are behaviorally identical to human beings?” 5. This is what Terence Horgan (1987) calls physical possibility, which he distinguishes from a slightly weaker nomological possibility. 6. It is sometimes joked that what is metaphysically possible is what Saul Kripke says is metaphysically possible. David Lewis (1980) says that he does not know how to prove that something is possible. Appeals to possible worlds explain to what statements of metaphysical possibility refer, but they do little to tell us which worlds are metaphysically possible. Metaphysical possibility has an important part in conceivability arguments for zombies, e.g., Kirk (1974) and Chalmers (1996a). See also Balog (1999), Levine (2001), Block and Stalnaker (1999), Chalmers and Jackson (2001), and Braddon-Mitchell (2003). 7. These are the sort that Flanagan and I (1995) defended against Moody’s (1994) thought experiment, which sparked Dennett’s distress. 8. These questions are all phrased in terms of interpersonal comparisons. If you think that the interpersonal cases are always or sometimes ill defined, you can still ask all of these questions in their intrapersonal form. To do so, simply consider the
269 Notes
two creatures in the above formulations as two creature-stages; that is, “Is it y-ly possible that a creature at time t2 be x-ly identical to the same creature at t1 but differ in that . . . ?” It is a short step from the intrapersonal form to the first-person form, “Is it y-ly possible that I at time t2 be x-ly identical to myself at t1 but differ in that . . . ?” 9. Distinguishing finer and coarser varieties of functional specification in this way does not in itself involve Two-Levelism (Lycan 1987). 10. But the matter is not simple. As Dennett tells Michael Gazzaniga, regarding a sophisticated robot, Cog, designed to model some human capacities: “Achieving human-level hand-eye coordination is a central goal, but before that can be addressed, we have to ensure that Cog won’t poke its eyes out with inadvertent motions of its arms! So a pain system, and innately ‘hard-wired’ (actually softwarecontrolled, of course) avoidance of such mischief is a high priority” (quoted in Gazzaniga 1997). 11. This requirement, notice, is much stronger than Block’s “Disney Principle” (1997). 12. To be precise: epiphenomenal or otherwise “systematically mysterious.” But the only sense of ungrounded mystery that Dennett provides is epiphenomenalism. 13. Since Dennett’s opponent in chapter 12, section 5 of Consciousness Explained is Frank Jackson (1982), his argument is in part against nonphysicalist claims about consciousness and qualia. Eliminating dualism from consideration is not allowable if we were trying to settle the matter between Jackson and Dennett. But my goal is to defend the notion of zombies on naturalist terms. 14. But not all. Dretske’s (1995) view of consciousness permits physically identical zombies. This is because Dretske’s preferred notion of function is etiological rather than causal; it allows that two things could be physically identical at a time but have different histories, and thus different etiological functions. I don’t know how Dennett feels about the kinds of zombies that Dretske’s view permits. 15. Dan Ryder has cautioned me that it could be objected that carburetors and fuel injectors are not functionally equivalent on the grounds that fuel injectors have performance characteristics that carburetors do not. The response to this is as follows: First, although one can construct a fuel injector to have better (e.g., more fuel efficient) operation than any carburetor, one does not have to. I assume that it is possible to construct a fuel injector with the same performance characteristics as a carburetor. I see no reason that this duplication could not be so complete as to also match the counterfactual cases (such as those outside the range of any carburetor). Specifically, the fuel injector would be designed to break under certain conditions, viz., those under which the carburetor would break. (Of course they would break in different ways. Would that show that there were not functionally equivalent, after all? It sounds like we’re sneaking more and more of the causal details into the functional story. To avoid this, it is important to designate ahead of time the systems and the behaviors relative to which functions are specified.) 16. Dennett is not the only philosopher who falls victim to this sort of thinking. David Chalmers’s (1995, 1996a) argument for the incompleteness of physical theory vis á vis consciousness seems to involve the reasoning that because it is always possible to provide an explanation for mental phenomena in functional or mechanistic terms
270 Notes
that do not mention or entail facts about consciousness, consciousness must not be explainable functionally or mechanistically. 17. Bruce Mangan (1998) calls this the Fallacy of Functional Exclusion. Mangan and I independently arrived at similar ideas about a mistake being made in the debate about the function and efficacy of consciousness and came to similar diagnoses of the origin of the problem. I developed the point as having to do with the structure of the debate over functionalism and the invocation of possible explanations, whereas Mangan concentrates on exposing the fallacy, locating the problem in philosophers’ failure to think of consciousness in biological terms. 18. If P is applied to a system of pipes, then the water-carrier is a bound variable, with the value pipe. 19. According this story, however, either psychology is not entirely functional (Y is not a psychology) or else consciousness is not essential to psychology. 20. See also Shoemaker’s (1981a) reply, and Hardcastle’s (1995) discussion of the exchange.
Chapter 8 1. I am going to provide an answer to that question. But I want to be clear that Putnam, at least, does not care what the answer is. He regards it as absurd that the question has arisen in this way; any view that has to answer this question is on the wrong path. 2. See also Bechtel and Mundale (1999) and Bechtel and McCauley (1999). 3. John Perry (2001) also notes some Humean aspects of the version of identity theory that he defends on the basis of what he calls “antecedent physicalism.”
References
Adams, F. 1979. Properties, functionalism, and the identity theory. Eidos 1(2): 153–179. Adams, F. 1985. Comparison shopping in the philosophy of mind. Critica 17(50): 45–70. Akiba, K. 2000. Identity is simple. American Philosophical Quarterly 37(4): 389–404. Allen, C., M. Bekoff, and G. Lauder (eds.). 1998. Nature’s Purposes: Analyses of Function and Design in Biology. Cambridge, Mass.: The MIT Press. Amundson, R., and G. Lauder. 1994. Function without purpose: The Uses of causal role function in evolutionary biology. Biology and Philosophy 9: 443–469. Reprinted in Allen, Bekoff, and Lauder (1998). Angell, J. 1907. The province of functional psychology. Psychological Review 14(2): 1–91. Antony, L. 1999a. Making room for the metal. Philosophical Studies 95: 37–44. Antony, L. 1999b. Multiple realizability, projectibility, and the reality of mental properties. Philosophical Topics 26: 1–24. Antony, L., and J. Levine. 1997. Reduction with autonomy. In Tomberlin (1997). Aristotle. 1941. The Basic Works of Aristotle. R. McKeon (ed.). New York: Random House. Armstrong, D. 1968. A Materialist Theory of Mind. London: Routledge and Kegan Paul. Aydede, M. 2000. An Analysis of pleasure vis-à-vis pain. Philosophy and Phenomenological Research 61(3): 537–570. Aydede, M. 2001. Naturalism, introspection, and direct realism about pain. Consciousness and Emotion 2(1): 29–73.
272 References
Aydede, M., G. Güzeldere, and Y. Nakamura. Forthcoming. The Puzzle of Pain: Philosophical and Scientific Essays. Cambridge, Mass.: The MIT Press. Balog, K. 1999. Conceivability, possibility, and the mind-body problem. Philosophical Review 108(4): 497–528. Bealer, G. 1994. Mental properties. Journal of Philosophy 91(4): 185–208. Bealer, G. 1997. Self-consciousness. Philosophical Review 106(1): 69–117. Bechtel, W., and R. McCauley. 1999. Heuristic identity theory (or back to the future): The mind-body problem against the background of research strategies in cognitive neuroscience. In Proceedings of the 21st Annual Meeting of the Cognitive Science Society, 67–72. Mahwah, N.J.: Lawrence Erlbaum. Bechtel, W., and J. Mundale. 1999. Multiple realizability revisited: Linking cognitive and neural states. Philosophy of Science 66: 175–207. Bechtel, W., and R. Richardson. 1993. Discovering Complexity: Decomposition and Localization as Strategies in Scientific Research. Princeton, N.J.: Princeton University Press. Berkeley, G. 1998 (1713). Three Dialogues between Hylas and Philonous. Jonathan Dancy (ed.). New York: Oxford University Press. Bickle, J. 1996. New wave psychophysical reductionism and the methodological caveats. Philosophy and Phenomenological Research 56(1): 57–78. Bickle, J. 1998. Psychoneural Reduction: The New Wave. Cambridge, Mass.: The MIT Press. Bickle, J. 2003. Philosophy of Neuroscience: A Ruthlessly Reductive Approach. Dordrecht: Kluwer. Bigelow, J., and R. Pargetter. 1987. Functions. Journal of Philosophy 84(4): 181–196. Block, N. 1978. Troubles with functionalism. In C. W. Savage (ed.), Minnesota Studies in the Philosophy of Science, vol. 9. Minneapolis, Minn.: University of Minnesota Press. (Reprinted in Block 1980a.) Block, N. (ed.) 1980a. Readings in Philosophy of Psychology, volume 1. Cambridge, Mass.: Harvard University Press. Block, N. (ed.) 1980b. Readings in Philosophy of Psychology, volume 2. Cambridge, Mass.: Harvard University Press. Block, N. 1980c. Introduction: What is functionalism? In Block (1980b). Block, N. 1980d. Are absent qualia impossible? Philosophical Review 89(2): 257–274. Block, N. 1995. On a confusion about the function of consciousness. Behavioral and Brain Sciences 18: 227–247. Block, N. 1997. Anti-reductionism slaps back. In Tomberlin (1997).
273 References
Block, N., and R. Stalnaker. 1999. Conceptual analysis, dualism, and the explanatory gap. Philosophical Review 108(1): 1–46. Block, N., O. Flanagan, and G. Güzeldere (eds.). 1997. The Nature of Consciousness: Philosophical Debates. Cambridge, Mass.: The MIT Press. Blumenfeld, J. B. 1975. Kripke’s refutation of materialism. Australian Journal of Philosophy 53: 151–156. Bolender, J. 1995. Is multiple realizability compatible with antireductionism? Southern Journal of Philosophy 33: 129–142. Bontly, T. 2002. The supervenience argument generalizes. Philosophical Studies 109: 75–96. Boyd, R. 1980. Materialism without reductionism: What physicalism does not entail. In Block (1980a). Braddon-Mitchell, D. 2003. Qualia and analytical conditionals. Journal of Philosophy 100(3): 111–135. Brand, M. 1979. On Tye’s “Brand on event identity.” Philosophical Studies 36: 61–68. Brandon, R. 1990. Adaptation and Environment. Princeton, N.J.: Princeton University Press. Brandon, R. 1996. Reductionism versus holism versus mechanism. In R. Brandon, Concepts and Methods in Evolutionary Biology. New York: Cambridge University Press. Brentano, F. 1973 (1874). Psychology from an Empirical Standpoint. Rancurello, Terell, and McAlister (trans.). London: Routledge and Kegan Paul. Broad, C. D. 1925. The Mind and Its Place in Nature. New York: Harcourt, Brace. Brodmann, K. 1909. Vergleichende Lokalisationslehre der Grosshirnrinde in ihren Prinzipien dargestellt auf Grund des Zellenbaues. Republished in translation as Brodmann’s “Localisation in the cerebral cortex,” L. Garey (ed. and trans.) London: Imperial College Press, 1999. Buller, D. 1999. DeFreuding evolutionary psychology: Adaptation and human motivation. In Hardcastle (1999). Buonomano, D., and M. Merzenich. 1998. Cortical plasticity: From synapses to maps. Annual Review of Neuroscience 21: 149–186. Byrne, A., and D. Hilbert (eds.). 1997a. Readings on Color, volume 1: The Philosophy of Color. Cambridge, Mass.: The MIT Press. Byrne, A., and D. Hilbert (eds.). 1997b. Readings on Color, volume 2: The Science of Color. Cambridge, Mass.: The MIT Press. Caplan, A. 1978. The Sociobiology Debate. New York: Harper and Row.
274 References
Carnap, R. 1935. Philosophy and Logical Syntax. London: Kegan Paul, Trench, Trubner. Chalmers, D. 1994. On Implementing a computation. Minds and Machines 4(4): 391–402. Chalmers, D. 1995. Facing up to the problem of consciousness. Journal of Consciousness Studies 2(3): 200–219. Chalmers, D. 1996a. The Conscious Mind: In Search of a Fundamental Theory. New York: Oxford University Press. Chalmers, D. 1996b. Does a rock implement every finite-state automaton? Synthese 108: 309–333. Chalmers, D. (ed.). 2002. Philosophy of Mind: Classical and Contemporary Readings. New York: Oxford University Press. Chalmers, D., and F. Jackson. 2001. Conceptual analysis and reductive explanation. Philosophical Review 110(3): 315–361. Churchland, P. M. 1979. Scientific Realism and the Plasticity of Mind. New York: Cambridge University Press. Churchland, P. M. 1982. Is “thinker” a natural kind? Dialogue 21(2): 223–238. Churchland, P. M. 1985. Reduction, qualia, and the direct introspection of brain states. Journal of Philosophy 82(1): 1–22. Churchland, P. M. 1988. Matter and Consciousness, revised edition. Cambridge, Mass.: The MIT Press. Churchland, P. S. 1983. Consciousness: The transmutation of a concept. Pacific Philosophical Quarterly 64: 80–93. Churchland, P. S. 1986. Neurophilosophy: Toward a Unified Science of Mind–Brain. Cambridge, Mass.: The MIT Press. Churchland, P. S., and T. Sejnowski. 1988. Perspectives on cognitive neuroscience. Science 242(4879): 741–745. Clapp, L. 2001. Disjunctive properties: Multiple realizations. Journal of Philosophy 98: 111–136. Clark, A. 1993. Sensory Qualities. Oxford: Clarendon/Oxford University Press. Cohen, M., and S. Bookheimer. 1994. Functional magnetic resonance imaging. Trends in Neurosciences 17(7): 268–277. Craver, C. 2001. Role functions, mechanisms, and hierarchy. Philosophy of Science 68: 53–74. Craver, C., and L. Darden. 2001. Discovering mechanisms in neuroscience: The case of spatial memory. In Theory and Method in Neuroscience, P. Machamer, R. Gush, and P. McLaughlin (eds.). Pittsburgh, Pennsylvania: University of Pittsburgh Press.
275 References
Craver, C., and W. Bechtel. Forthcoming. Mechanisms and mechanistic explanation. In S. Sarkar (ed.), Dictionary of the Philosophy of Biology. Cummins, R. 1975. Functional analysis. Journal of Philosophy 72(20): 741–765. Cummins, R. 1983. The Nature of Psychological Explanation. Cambridge, Mass.: The MIT Press. Cummins, R. 1989. Meaning and Mental Representation. Cambridge, Mass.: The MIT Press. Das, A. 1997. Plasticity in adult sensory cortex: A review. Network: Computation in Neural Systems 8: R33–R76. David, M. 1997. Kim’s functionalism. In Tomberlin (1997). Davidson, D. 1970. Mental events. In Foster and Swanson (eds.), Experience and Theory. Amherst, Mass.: University of Massachusetts Press. Davidson, D. 1987. Knowing one’s own mind. Proceedings and Addresses of the American Philosophical Association 60: 441–458. Davies, P. 1994. Troubles for direct proper functions. Noûs 28(3): 363–381. Davies, P. 2000. The nature of natural norms: Why selected functions are systemic capacity functions. Noûs 34: 85–107. Davies, P. 2001. Norms of Nature: Naturalism and the Nature of Functions. Cambridge, Mass.: The MIT Press. Dennett, D. 1969. Content and Consciousness. London: Routledge and Kegan Paul. Dennett, D. 1971. Intentional systems. Journal of Philosophy 68: 87–106. Dennett, D. 1978/1981. Brainstorms. Cambridge, Mass.: Bradford Books/The MIT Press. Dennett, D. 1984. Elbow Room: Varieties of Free Will Worth Wanting. Cambridge, Mass.: The MIT Press. Dennett, D. 1987. The Intentional Stance. Cambridge, Mass.: The MIT Press. Dennett, D. 1988. Quining qualia. In Marcel and Bisiach (1988). Dennett, D. 1991. Consciousness Explained. Boston: Little, Brown. Dennett, D. 1994. Get real. Philosophical Topics 22(1, 2): 505–568. Dennett, D. 1995. The unimagined preposterousness of zombies. Journal of Consciousness Studies 2(4): 322–326. Reprinted in Dennett (1998). Dennett, D. 1998. Brainchildren. Cambridge, Mass.: The MIT Press.
276 References
Dennett, D. 2000. With a little help from my friends. In Ross, Brook, and Thompson (2000). Dennett, D. 2001. The zombic hunch: Extinction of an intuition. Royal Institute of Philosophy Supplement 48: 27–43. Deutsch, C., S. Oriss, A. DiFiore, and W. McIlvane. 2000. Measuring brain activity correlates of behavior: A methodological overview. Experimental Analysis of Human Behavior 18: 36–42. Dewey, J. 1896. The reflex arc concept in psychology. Psychological Review 3: 357–370. Dretske, F. 1986. Misrepresentation. In Belief: Form, Content, and Function, R. Bogdan (ed.). Reprinted in Lycan (1990). Dretske, F. 1988. Explaining Behavior. Cambridge, Mass.: The MIT Press. Dretske, F. 1995. Naturalizing the Mind. Cambridge, Mass.: The MIT Press. Enç, B. 1983. In defense of the identity theory. Journal of Philosophy 80: 279–298. Enç, B. 1986. Essentialism with individual essences: Causation, kinds, supervenience, and restricted identities. In Midwest Studies in Philosophy 11: 403–426. Enç, B. 2002. Indeterminacy of function attributions. In A. Ariew, R. Cummins, and M. Perlman (eds.), Functions: New Essays in the Philosophy of Psychology and Biology. Oxford: Oxford University Press. Endicott, R. 1989. On physical multiple realizability. Pacific Philosophical Quarterly 70: 212–224. Endicott, R. 1993. Species-specific properties and more narrow reductive strategies. Erkenntnis 38: 303–321. Endicott, R. In preparation. The Book of Realization: Analyzing a Cluster of Concepts for the Philosophy of Mind, Solving a Computational Conundrum for Cognitive Science. Feigl, H. 1958. The “mental” and the “physical.” In H. Feigl, M. Scriven, and G. Maxwell (eds.), Minnesota Studies in the Philosophy of Science 2:370–497. Feldman, F. 1973. Kripke’s argument against materialism. Philosophical Studies 24(6): 416–419. Reprinted in Block (1980a). Fetzer, J. (ed.). 2002. Evolving Consciousness. Amsterdam: John Benjamins. Field, H. 1978. Mental representation. Erkenntnis 13(1): 9–61. Reprinted in Block (1980a). Flanagan, O. 1991. The Science of the Mind. Cambridge, Mass.: The MIT Press. Flanagan, O. 1992. Consciousness Reconsidered. Cambridge, Mass.: The MIT Press. Flanagan, O. 1995a. Behaviourism. In Honderich (1995).
277 References
Flanagan, O. 1995b. Deconstructing dreams: The spandrels of sleep. Journal of Philosophy 92: 5–27. Reprinted with modifications in Flanagan (1996). Flanagan, O. 1996. Self Expressions: Mind, Morals, and the Meaning of Life. New York: Oxford University Press. Flanagan, O. 2000. Dreaming Souls. New York: Oxford University Press. Flanagan, O., and T. Polger. 1995. Zombies and the function of consciousness. Journal of Consciousness Studies 2(4): 313–321. Fodor, J. 1968. Psychological Explanation. New York: Random House. Fodor, J. 1974. Special sciences, or the disunity of science as a working hypothesis. Synthese 28: 97–115. Reprinted in Block (1980a). Fodor, J. 1978. Propositional attitudes. Monist 61(4): 501–523. Reprinted in Block (1980b). Fodor, J. 1983. The Modularity of Mind. Cambridge, Mass.: The MIT Press. Fodor, J. 1985. Fodor’s guide to mental representation: The intelligent auntie’s vademecum. Mind 94: 76–100. Reprinted in Fodor (1990). Fodor, J. 1986. Banish DisContent. In Language, Mind, and Logic, J. Butterfield (ed.). New York: Cambridge University Press. Reprinted in Lycan (1990). Fodor, J. 1990. A Theory of Content and Other Essays. Cambridge, Mass.: The MIT Press. Fodor, J. 1997. Special sciences: Still autonomous after all these years. In Tomberlin (1997). Fox, J. 1984. The brain’s dynamic way of keeping in touch. Science 225: 820–821. Gazzaniga, M. 1997. Conversations in the Cognitive Neurosciences. Cambridge, Mass.: The MIT Press. Gibbard, A. 1975. Contingent identity. Journal of Philosophical Logic 4: 187–221. Gillett, C. 2002. The dimensions of realization: A critique of the standard view. Analysis 64(4): 316–323. Gillett, C., and B. Rives. 2001. Does the argument from realization generalize? Responses to Kim. Southern Journal of Philosophy 39: 79–98. Gillett, C. Unpublished. Moving beyond the subset model of realization. Author’s manuscript. Gillett, C., and B. Loewer (eds.). 2001. Physicalism and Its Discontents. Cambridge: Cambridge University Press. Gjelsvik, O. 1987. A Kripkean objection to Kripke’s argument against identity theories. Inquiry 30: 435–450.
278 References
Glennan, S. 1996. Mechanisms and the nature of causation. Erkenntnis 44: 49–71. Godfrey-Smith, P. 1994. A modern history theory of functions. Noûs 28(3): 344–362. Godfrey-Smith, P. 1996. Complexity and the Function of Mind in Nature. New York: Cambridge University Press. Gould, S. J., and R. Lewontin. 1978. The spandrels of San Marco and the Panglossian paradigm: A critique of the adaptationist program. Proceedings of the Royal Society, London 205: 581–598. Graham, D. 1978. Caffeine: Its identity, dietary sources, intake, and biological effects. Nutritional Review 36: 97–102. Graham, G. 1993. Philosophy of Mind: An Introduction. Oxford: Blackwell. Graham, G., and T. Horgan. 2002. Sensations and grain processes. In Fetzer (2002). Graham, G., and L. Stephens. 1994. Philosophical Psychopathology. Cambridge, Mass.: The MIT Press. Grantham, T., and S. Nichols. 1999. Evolutionary psychology: Ultimate explanations and Panglossian predictions. In Hardcastle (1999b). Griffiths, P. 1993. Functional analysis and proper function. British Journal for Philosophy of Science 44: 409–422. Gunderson, K. (ed.). 1975. Language, Mind, and Knowledge, volume VII: Minnesota Studies in the Philosophy of Science. Minneapolis, Minn.: University of Minnesota Press. Gustafson, D. 1995. Belief in pain. Consciousness and Cognition 4: 323–345. Gustafson, D. 1998. Pain, qualia, and the explanatory gap. Philosophical Psychology 11: 371–387. Guttenplan, S. (ed.). 1994. A Companion to the Philosophy of Mind. Oxford: Blackwell. Güzeldere, G. 1995a. Varieties of zombiehood. Journal of Consciousness Studies 2(4): 326–333. Güzeldere, G. 1995b. Consciousness: What it is, how to study it, what to learn from its history. Journal of Consciousness Studies 2(1): 30–51. Hardcastle, V. 1995. Locating Consciousness. Amsterdam: John Benjamins. Hardcastle, V. 1999a. The Myth of Pain. Cambridge, Mass.: The MIT Press. Hardcastle, V. (ed.). 1999b. Where Biology Meets Psychology: Philosophical Essays. Cambridge, Mass.: The MIT Press. Hardin, C. 1988. Color for Philosophers. Indianapolis, Ind.: Hackett. Hardin, C. 1991. Reply to Levine. Philosophical Psychology 4(1): 41–50.
279 References
Heil, J. 1992. The Nature of True Minds. New York: Cambridge University Press. Heil, J. 1999. Multiple realizability. American Philosophical Quarterly 36(3): 189–208. Hempel, C., and P. Oppenheim. 1948. Studies in the logic of explanation. Philosophy of Science 15: 135–175. Hill, C. 1981. Why Cartesian intuitions are compatible with the identity thesis. Philosophy and Phenomenological Research 42(2): 255–265. Hill, C. 1991. Sensations: A Defense of Type Materialism. Cambridge: Cambridge University Press. Hogben, L. 1930. The Nature of Living Matter. London: Kegan Paul, Trench, Trubner. Honderich, T. 1995. The Oxford Companion to Philosophy. New York: Oxford University Press. Horgan, T. 1987. Supervenient qualia. Philosophical Review 96: 491–520. Horgan, T. 1989. Mental quausation. Philosophical Perspectives 3: 47–76. Horgan, T. 1993. Nonreductive materialism and the explanatory autonomy of psychology. In S. Wagner and R. Warner (eds.), Naturalism: A Critical Appraisal. Notre Dame, Ind.: University of Notre Dame Press. Horgan, T. 1997. Kim on mental causation and causal exclusion. In Tomberlin (1997). Horgan, T. 2001. Causal compatibilism and the exclusion problem. Theoria 16: 95–116. Horgan, T., and J. Tienson. 1996. Connectionism and the Philosophy of Psychology. Cambridge, Mass.: The MIT Press. Horgan, T., and J. Tienson. 2002. The intentionality of phenomenology and the phenomenology of intentionality. In Chalmers (2002). Huettel, S. A., and G. McCarthy. 2000. Evidence for a refractory period in the hemodynamic response to visual stimuli as measured by MRI. NeuroImage 11(5): 547–553. Hull, D. 1976. Are species really individuals? Systematic Zoology 25: 174–191. Hurvich, L. 1981. Color Vision. Sunderland, Mass.: Sinauer. Jackson, F. 1982. Epiphenomenal qualia. Philosophical Quarterly 32(127): 127–136. Jackson, F. Unpublished. Mind and illusion. Author’s manuscript. Jackson, F., and P. Pettit. 1990. Program explanation: A general perspective. Analysis 50: 107–117. Jackson, F., and P. Pettit. 1995. Moral functionalism and moral motivation. Philosophical Quarterly 45(178): 20–40.
280 References
Jackson, F., and P. Pettit. 1996. Moral functionalism, supervenience, and reductionism. Philosophical Quarterly 46(182): 82–86. James, W. 1890/1950. The Principles of Psychology. New York: Dover Publications. James, W. 1911/1979. Some Problems of Philosophy. Cambridge, Mass.: Harvard University Press. Jaworski, W. 2002. Multiple-realizability, explanation, and the disjunctive move. Philosophical Studies 108: 289–308. Keeley, B. 2002. Making sense of the senses: Individuating modalities in humans and other animals. Journal of Philosophy 99(1): 1–24. Kim, J. 1972. Phenomenal properties, psychophysical laws, and identity theory. Monist 56(2): 177–192. Excerpted in Block (1980a) under the title “Physicalism and the multiple realizability of mental states.” Kim, J. 1989. The myth of nonreductive materialism. Proceedings and Addresses of the American Philosophical Association 63(3): 31–47. Reprinted in Kim (1993). Kim, J. 1993. Supervenience and Mind. New York: Cambridge University Press. Kim, J. 1996. Philosophy of Mind. Boulder, Colo.: Westview. Kim, J. 1997. The mind–body problem: Taking stock after forty years. In Tomberlin (1997). Kim, J. 1998. Mind in a Physical World: An Essay on the Mind–Body Problem and Mental Causation. Cambridge, Mass.: The MIT Press. Kim, J. 1999. Making sense of emergence. Philosophical Studies 95: 3–36. Kim, S. 2002. Testing multiple realizability: A discussion of Bechtel and Mundale. Philosophy of Science 69(4): 606–610. Kirk, R. 1974. Zombies v. Materialists. Proceedings of the Aristotelian Society 48: 135–152. Kripke, S. 1971. Identity and necessity. In M. Munitz (1971). Kripke, S. 1972/1980. Naming and Necessity. Cambridge, Mass.: Harvard University Press. Levin, J. 1985. Functionalism and the argument from conceivability. Canadian Journal of Philosophy 11 (supplementary volume). Levin, J. 1991. Analytic functionalism and the reduction of phenomenal states. Philosophical Studies 61(3): 211–238. Levine, J. 1983. Materialism and qualia: The explanatory gap. Pacific Philosophical Quarterly 64: 354–361. Levine, J. 1991. Cool red. Philosophical Psychology 4(1): 27–40.
281 References
Levine, J. 1997. On leaving out what it’s like. In Block, Flanagan, and Güzeldere (1997). Levine, J. 2001. Purple Haze: The Puzzle of Consciousness. New York: Oxford University Press. Lewis, D. 1966. An argument for the identity theory. Journal of Philosophy 63: 17–25. Lewis, D. 1969. Review of art, mind, and religion. Journal of Philosophy 66: 23–35. Excerpted in Block (1980a) as “Review of Putnam.” Lewis, D. 1970. How to define theoretical terms. Journal of Philosophy 68: 203–211. Reprinted in Lewis (1983). Lewis, D. 1972. Psychophysical and theoretical identifications. Australasian Journal of Philosophy 50: 249–258. Reprinted in Lewis (1999). Lewis, D. 1980. Mad pain and Martian pain. In Block (1980a). Lewis, D. 1983. Philosophical Papers, volume I. New York: Oxford University Press. Lewis, D. 1994. Reduction of mind. In Guttenplan (1994). Lewis, D. 1999. Papers in Metaphysics and Epistemology. New York: Cambridge University Press. Libet, B. 1985. Unconscious cerebral initiative and the role of conscious will in voluntary action. Behavioral and Brain Sciences 8: 529–566. Loar, B. 1981. Mind and Meaning. New York: Cambridge University Press. Locke, J. 1975 (1690). Essay Concerning Human Understanding. Peter H. Nidditch (ed.). Oxford: Clarendon Press. Lycan, W. 1974a. Kripke and the materialists. Journal of Philosophy 71(18): 677–689. Lycan, W. 1974b. Mental states and Putnam’s functionalist hypothesis. Australasian Journal of Philosophy 52: 48–62. Lycan, W. 1981. Form, function, and feel. Journal of Philosophy 78: 24–50. Lycan, W. 1987. Consciousness. Cambridge, Mass.: The MIT Press. Lycan, W. (ed.). 1990. Mind and Cognition: A Reader. Cambridge, Mass.: Blackwell. Lycan, W. 1996. Consciousness and Experience. Cambridge, Mass.: The MIT Press. Lynch, M. 1998. Truth in Context: An Essay on Pluralism and Objectivity. Cambridge, Mass.: The MIT Press. Lynch, M. 2001a. A functionalist theory of truth. In Lynch (2001b). Lynch, M. 2001b. The Nature of Truth: Classic and Contemporary Perspectives. Cambridge, Mass.: The MIT Press.
282 References
MacDonald, C. 1989. Mind–body Identity Theories. London: Routledge. MacIntyre, A. 1999. Dependent Rational Animals: Why Human Beings Need the Virtues. Chicago: Open Court. Machamer, P., L. Darden, and C. Craver. 2000. Thinking about mechanisms. Philosophy of Science 67: 1–25. Mangan, B. 1998. Consciousness, biological systems, and the fallacy of functional exclusion. Paper presented at Toward a Science of Consciousness conference (Tucson III), Tucson, Arizona. Marcel, T., and E. Bisiach. 1988. Consciousness in Contemporary Science. New York: Oxford University Press. Marr, D. 1982. Vision: A computational investigation into the human representation and processing of visual information. San Francisco: W. H. Freeman. McCarthy, G., A. Puce, J. Gore, and T. Allison. 1997. Face-specific processing in the human fusiform gyrus. Journal of Cognitive Neuroscience 9: 605–610. McDowell, J. 1994. Mind and World. Cambridge, Mass.: Harvard University Press. McGinn, C. 1991. The Problem of Consciousness. Oxford: Basil Blackwell. Melzack, R., and P. Wall. 1965. Pain mechanisms: A new theory. Science 150: 971–979. Melzack, R., and P. Wall. 1988. The Challenge of Pain. London: Penguin Books. Millikan, R. 1984. Language, Thought, and Other Biological Categories. Cambridge, Mass.: The MIT Press. Millikan, R. 1989. In defense of proper functions. Philosophy of Science 56: 288–302. Reprinted in Millikan (1993): 13–29. Millikan, R. 1993. White Queen Psychology and Other Essays for Alice. Cambridge, Mass.: The MIT Press. Milner, B., and M. A. Goodale. 1995. The Visual Brain in Action. New York: Oxford University Press. Moody, T. 1994. Conversations with zombies. Journal of Consciousness Studies 1(2): 196–200. Munitz, M. 1971. Identity and Individuation. New York: New York University Press. Nagel, E. 1961. The Structure of Science: Problems in the Logic of Scientific Explanation. New York: Harcourt, Brace, and World. Nagel, T. 1974. What is it like to be a bat? Philosophical Review 83(4): 435–450. Neander, K. 1991. Functions as selected effects: The conceptual analyst’s defense. Philosophy of Science 58: 168–184.
283 References
Palmer, S. 1999. Color, consciousness, and the isomorphism constraint Behavioral and Brain Sciences 22(6): 1–21. Pereboom, D. 2002. Robust nonreductive materialism. Journal of Philosophy 99(10): 499–531. Pereboom, D., and H. Kornblith. 1991. The metaphysics of irreducibility. Philosophical Studies 63: 125–145. Perry, J. 2001. Knowledge, Possibility, and Consciousness. Cambridge, Mass.: The MIT Press. Pettit, P. 1986. Broad-minded explanation in psychology. In P. Pettit, and J. McDowell (eds.), Subject, Thought, and Content. Oxford: Oxford University Press. Pitt, D. In press. The phenomenology of cognition, or What is it like to think that P ? Philosophy and Phenomenological Research. Poland, J. 1994. Physicalism: The Philosophical Foundations. New York: Oxford University Press. Polger, T. 2000. Zombies explained. In Ross, Brook, and Thompson (2002). Polger, T. 2001. Review of Brainchildren: Essays on Designing Minds. Philosophical Psychology 14(2): 245–250. Polger, T. 2002. Putnam’s intuition. Philosophical Studies 109(2): 143–170. Polger, T., and O. Flanagan. 1999. Natural answers to natural questions. In Hardcastle (1999b). Polger, T., and O. Flanagan. 2001. A decade of teleofunctionalism: Lycan’s Consciousness and Consciousness and Experience. Minds and Machines 11(1): 113–126. Polger, T., and O. Flanagan. 2002. Consciousness, adaptation, and epiphenomenalism. In Fetzer (2002). Putnam, H. 1960. Minds and machines. In Hook (ed.), Dimensions of Mind. New York: New York University Press. Reprinted in Putnam (1975c). Putnam, H. 1967. Psychological predicates. In W. Capitan and D. Merrill (eds.), Art, Mind, and Religon. Pittsburgh, Penn.: University of Pittsburgh Press. Reprinted in Putnam (1975c), Block (1980a), and elsewhere as “The nature of mental states.” Putnam, H. 1975a. The Meaning of “Meaning.” In Gunderson (1975). Putnam, H. 1975b. Philosophy and our mental life. In Putnam (1975c). Putnam, H. 1975c. Mind, Language and Reality: Philosophical Papers, volume 2. New York: Cambridge University Press. Putnam, H. 1988. Representation and Reality. Cambridge, Mass.: The MIT Press.
284 References
Putnam, H. 1994a. Putnam, Hilary. In Guttenplan (1994). Putnam, H. 1994b. Sense, nonsense, and the senses: An inquiry into the powers of the human mind. Journal of Philosophy 91(9): 445–517. Reprinted in Putnam (1999). Putnam, H. 1999. The Threefold Cord: Mind, Body, and World. New York: Columbia University Press. Ramachandran, V. 1998. Consciousness and body image: Lessons from phantom limbs, Capgras syndrome, and pain asymbolia. Philosophical Transactions of the Royal Society, London B, 353: 1851–1859. Reader’s Digest (ed.). 1981. Complete Car Care Manual. Pleasantville, N.Y.: The Reader’s Digest Association. Rey, G. 1997. Contemporary Philosophy of Mind. Boston: Blackwell. Richardson, R. 1979. Functionalism and reductionism. Philosophy of Science 46: 533–558. Richardson, R. 1982. How not to reduce a functional psychology. Philosophy of Science 49(1): 125–137. Richardson, R. 2001. Evolution without history: Critical reflections on evolutionary psychology. In H. R. Holcomb III (ed.), Conceptual Challenges in Evolutionary Psychology. Dordrecht: Kluwer. Richardson, R. In preparation. Maladapted Psychology. Roback, A. 1952. History of American Psychology. New York: Library Publishers. Rorty, R. 1979. Philosophy and the Mirror of Nature. Princeton, N.J.: Princeton University Press. Ross, D., A. Brook, and D. Thompson (eds.). 2000. Dennett’s Philosophy: A Comprehensive Assessment. Cambridge, Mass.: The MIT Press. Ryle, G. 1949. The Concept of Mind. Chicago: University of Chicago Press. Salmon, W. 1989. Four decades of scientific explanation. In Kitcher and Salmon (eds.), Minnesota Studies in the Philosophy of Science XIII: Scientific Explanation. Minneapolis, Minn.: University of Minnesota Press. Reprinted as Four Decades of Scientific Explanation (Minneapolis, Minn.: University of Minnesota Press, 1989). Searle, J. 1980. Minds, brains, and programs. Behavioral and Brain Sciences 3(3): 417–424. Searle, J. 1992. The Rediscovery of the Mind. Cambridge, Mass.: The MIT Press. Sellars, W. 1956. Empiricism and the philosophy of mind. In H. Feigl and M. Scriven (eds.), Minnesota Studies in the Philosophy of Science I. Minneapolis, Minn.: University of Minnesota Press. Shapiro, L. 2000. Multiple realizations. Journal of Philosophy 97: 635–654.
285 References
Shapiro, L. Forthcoming. The Mind Incarnate. Cambridge, Mass.: The MIT Press. Shoemaker, S. 1975. Functionalism and qualia. Philosophical Studies 27: 291–315. Reprinted in Block (1980a). Shoemaker, S. 1980. Causality and properties. In P. van Inwagen (ed.), Time and Cause. Dordrecht: D. Reidel. Reprinted in Shoemaker (1984). Shoemaker, S. 1981a. Absent qualia are impossible—A reply to Block. Philosophical Review 90(4): 581–599. Shoemaker, S. 1981b. Some varieties of functionalism. Philosophical Topics 12(1): 83–118. Reprinted in Shoemaker (1984). Shoemaker, S. 1982. The inverted spectrum. Journal of Philosophy 79(7): 357–381. Reprinted in Block, Flanagan, and Güzeldere (1997). Shoemaker, S. 1984. Identity, Cause, and Mind. New York: Cambridge University Press. Shoemaker, S. 1996. The First-Person Perspective and Other Essays. New York: Cambridge University Press. Shoemaker, S. 2000. Realization and mental causation. In The Proceedings of the Twentieth World Congress of Philosophy, volume 9: Philosophy of Mind, B. Elevitch (ed.). Bowling Green, Ohio: Philosophy Documentation Center. Revised as Shoemaker (2001). Shoemaker, S. 2001. Realization and mental causation. In Gillett and Loewer (2001). Shoemaker, S. 2002. Kim on emergence. Philosophical Studies 108: 53–63. Shoemaker, S. Forthcoming. Realization, micro-realization, and coincidence. Philosophy and Phenomenological Research. Sidelle, A. 1989. Necessity, Essence, and Individuation. Ithaca, N.Y.: Cornell University Press. Sidelle, A. 1992. Identity and identity-like. Philosophical Topics 20(1): 269–292. Siewert, C. 1998. The Significance of Consciousness. Princeton, N.J.: Princeton University Press. Skinner, B. 1953. The Science of Human Behavior. New York: Macmillan. Skipper, Jr., R. 1999. Selection and the extent of explanatory unification. Philosophy of Science 66: S196–S209. Smart, J. J. C. 1959. Sensations and brain processes. Philosophical Review 68: 141–156. Reprinted in V. Chapell (ed.), The Philosophy of Mind (Englewood Cliffs, N.J.: Prentice Hall, 1962). Smith, B. C. 1999. Requiem for computationalism. Presidential address to the annual meeting of the Society for Philosophy and Psychology, Stanford University. Smuts, J. 1926. Holism and Evolution. London: Macmillan.
286 References
Sober, E. 1985. Panglossian functionalism and the philosophy of mind. Synthese 64: 165–193. Reprinted in Lycan (1990). Sober, E. 1999. The multiple realizability argument against reductionism. Philosophy of Science 66: 542–564. Somorin, O. 1973. Spectrophotometric determination of caffeine in Nigerian kola nuts. Journal of Food Science 38: 911–912. Stoljar, D. 2000. Physicalism and the necessary a posteriori. Journal of Philosophy 97(1): 33–54. Stoljar, D., and I. Gold. 1999. A neuron doctrine in philosophy of neuroscience. Behavioral and Brain Sciences 22: 809–869. Strawson, G. 1994. Mental Reality. Cambridge, Mass.: The MIT Press. Sufka, K. 2000. Chronic pain explained. Brain and Mind 1: 155–179. Sufka, K., and M. Lynch. 2000. Sensations and pain processes. Philosophical Psychology 13: 299–311. Sufka, K., and D. Price. 2002. Gate control theory reconsidered. Brain and Mind 3: 277–290. Thompson, E. 1995. Colour: A Study in Cognitive Science and the Philosophy of Perception. Boston, Mass.: Blackwell. Tomberlin, J. (ed). 1997. Philosophical Perspectives 11: Mind, Causation, and World. Boston, Mass.: Blackwell. Tye, M. 1995. Ten Problems of Consciousness: A Representational Theory of the Phenomenal Mind. Cambridge, Mass.: The MIT Press. Tye, M. 2000. Consciousness, Color, and Content. Cambridge, Mass.: The MIT Press. Van Gulick, R. 1980. Functionalism, information, and content. Nature and System 2: 139–162. Reprinted in Lycan (1990). Van Gulick, R. 1982. Functionalism as a theory of mind. Philosophy Research Archives 8: 185–204. Van Gulick, R. 1988a. A functionalist plea for self-consciousness. Philosophical Review 97(2): 149–181. Van Gulick, R. 1988b. Consciousness, intrinsic intentionality, and self-understanding machines. In Marcel and Bisiach (1988). Van Gulick, R. 1989. What difference does consciousness make? Philosophical Topics 17: 211–230. Van Gulick, R. 1993. Understanding the phenomenal mind: Are we all just armadillos? In Block, Flanagan, and Güzeldere (1997).
287 References
Van Gulick, R. 1994. Deficit studies and the function of phenomenal consciousness. In G. Graham and L. S. Stephens (eds.), Philosophical Psychopathology. Cambridge, Mass.: The MIT Press. Van Gulick, R. 1998. Reduction, supervenience, and phenomenal consciousness. Presented at Towards a Science of Consciousness III, Tucson, Arizona. Wagner, S., and R. Warner (eds.). 1993. Naturalism: A Critical Appraisal. Notre Dame, Ind: University of Notre Dame Press. Watkins, M. 1999. Do animals see colors? An anthropocentrist’s guide to animals, the color blind, and far away places. Philosophical Studies 94(3): 189–209. Watson, J. 1913. Psychology as the behaviorist views it. Psychological Review 20: 158–177. Weiskrantz, L. 1986. Blindsight: A Case Study and Implications. New York: Oxford University Press. Weiskrantz, L. 1997. Consciousness Lost and Found: A Neuropsychological Exploration. New York: Oxford University Press. Wilson, E. O. 1975. Sociobiology: The New Synthesis. Cambridge, Mass.: Harvard University Press. Wilson, J. 1999. Biological Individuality. New York: Cambridge University Press. Wilson, R. 2001. Two views of realization. Philosophical Studies 104: 1–30. Wright, L. 1973. Functions. Philosophical Review 82(2): 139–168.
Index
Adams, Fred, 6, 10 Amundson, Ron, 171 Anomalous monism, 29 Antecedent physicalism, 32 Antireduction. See Reduction; Autonomy Antony, Louise, 64, 161, 191, 195 Argument from Nuh-uh, 7–10 Aristotle, 100, 164 Armstrong, David, 26, 44, 78, 97, 106 Artificial intelligence, 133 Strong AI, 81 Weak AI, 105, 108 Autonomy, 159, 190–198, 201–211. See also Explanation argument from, 192–196, 201–209 colloquial use, 209 Bealer, George, 6 Bechtel, William, 5, 16, 21–25, 28, 32–33, 51–52, 61 Behavior, 216–217, 220, 225 Behaviorism, 74–75 Berkeley, George, 57 Bickle, John, 5–6, 16, 32, 195 Bigelow, John, 169–170 Binding problem, 244 Biopsychology. See Psychology, biopsychology Blindsight, 229, 241, 245 Block, Ned, xxii, 28, 131, 152, 158, 187, 189, 214, 229–232 Boyd, Richard, 61 Brain imaging studies, 52, 54–56, 108. See also Brain maps
Brain maps, 17–18, 23–25, 53–54 Brain states. See Identity conditions, of brain states Brentano’s thesis, xxi–xxiii, 87–88, 149, 167, 176 Broad, C. D., 44 Brodmann, Korbinian, 23–25 Bundles. See Sensations, bundles; Persons and personal identity Carnap, Rudolf, xvii Causal exclusion argument. See Exclusion argument Causal roles, 64, 94, 98, 116, 120, 150, 182, 230. See also Function, causal role; Properties, causal theory of Causation. See also Mental causation; Exclusion argument macrocausation, xx, 161, 163 quausation, 128 strong efficacy, 128–129, 148, 150, 167, 241 weak efficacy, 127–128, 168 Center of narrative gravity, 246 Chalmers, David, 2, 37–38, 63, 127, 131, 214, 218, 239 Churchland, Patricia, xv, 54, 127, 195 Churchland, Paul, 15, 30, 75, 85 Clark, Austen, xv Computationalism, 67, 80–81, 93, 105, 117–118, 120, 127, 140, 148–149, 151, 158. See also Artificial intelligence Conscious inessentialism, 227 Craver, Carl, 196–204
290 Index
Cummins, Robert, 92–93, 101, 104–107, 113, 146–51, 162, 183–186, 196, 199–200, 224 Darden, Lindley, 196–204 David, Marian, 28 Davidson, Donald, 29, 170 Davies, Paul, 168, 171 Deductive-nomological (D-N) explanation, 63, 165, 205, 211–240, 244, 246 Dennett, Daniel, xv, xxii, 35 intentional stance, 88, 108 Multiple Drafts model, xxii, 234 physical stance, 231 on zombies (see Zombies) Descartes, René, xv Disjunctive view. See Identity, speciesspecific or local Dretske, Fred, xxii, 90–93, 128–129, 140, 165–172, 176 Dualism, 178 Cartesian, xviii, 79, 239 property, 37 Eliminativism, 126–127, 192, 196 Enç, Berent, 15, 30, 44 Epiphenomenalism, 126, 129, 160–161, 175, 192, 205, 222–230, 233–234 causal-role, 127, 225–230, 234, 237 etiological, 224–227, 230–236 spandrels, 224 strict-metaphysical, 127, 224, 235–237 Essence and essentialism, 39, 42–43, 60–63, 82, 96, 98, 119–120, 167, 189, 191, 195, 201 nominal essence, 82, 96, 99, 157 real essence, 33, 82, 96, 157, 179 Evolution, 100–102, 107, 164–168, 172 Evolutionary psychology. See Psychology, evolutionary psychology Examples aliens and Martians, xiii, 12, 14, 15, 19, 21, 26–27, 131, 155, 170 Amos the mouse, 169–170 ashtray, 48 automobile, 52, 153, 198, 243 (see also Examples, carburetors and fuel injectors) bricks and windows, xx, 161–162 caffeine, xv Cal Ripken, Jr. and Lou Gehrig, 83 carburetors and fuel injectors, 19–20,
198, 200, 210, 227–229, 233, 236, 240, 243 corkscrew, 208 cricket, 48, 56–57, 67 friendly physicist, 202–203, 207–208 Hinkfuss’s pail, 131, 153, 231–232, 242 instant hibernation, 167 liquid enjoyed by George Washington, 45–46 lumps of clay, xx, 47–48, 66 mousetrap, 197 octopus, 8, 22, 26–27, 155 pain and c-fibers, 7, 42, 51, 57–58, 61–63, 75, 79, 82, 96, 154, 157, 220, 245 petri dish, xxvi, 242–243 platypus, 227–229, 233 plumbing and pipes, 229–231 being president, 84 schoolchild’s art project, 47–48 square peg and round hole, 203–204, 207–208 thermostat, 15, 131, 231–232, 242 urn, 47, 66 water and H2O, 40–48, 59–60, 64, 67, 132, 231–232 widgets and thingamajigs, 48, 61 Exclusion argument, 160–164, 184, 207–208 Explanation, 104, 108, 196, 203–205, 209 mechanical, 209–211 pluralism, 200–211 vs. prediction, 36 psychological, 150, 154–156, 191, 203, 231 (see also Psychology) unification, 203–208 Explanatory gap, xxv, 63–70 Explanatory hooks, 65 Feigl, Herbert, 78 Field, Hartry, 81, 143–144 Flanagan, Owen, xv–xvi, xx, xxii, 1, 28, 58, 64–65, 165, 168, 172, 174, 207, 220, 224, 246 fMRI (functional magnetic resonance imagining), 52–56, 108 Fodor, Jerry, 28, 71, 80–81, 86–87, 90–93, 107, 115, 133, 144, 150, 182, 191, 194–195 Folk psychology. See Psychology, folk or common sense Function abstract, 73–74, 140–149, 215, 234
291 Index
biological or natural, 73–74, 113, 165, 190 (see also Function, teleological) causal role, 73–74, 95, 101, 104–105, 107, 113, 122, 145–164, 168, 171–176, 182–186, 190, 196, 200, 206, 210, 215, 224 computational, 140–152 (see also Function, abstract) etiological, 101, 113, 165–175, 206, 210, 215 (see also Function, teleological) hybrid, 168, 172–177, 190 indicator, 165, 176 propensity, 169–171 proper function (see Function, etiological) teleological, 74, 164–176, 187, 206, 210 (see also Function, etiological; Function, propensity; Function, hybrid; Function, biological or natural) Functional analysis, 104–107, 150, 196, 199 Functional duplicates, 170, 213–217, 225–229, 233 Functional equivalence, 143–145, 148–149, 182 Functionalism. See also Generic Functionalism anatomy of, 72–74, 109, 137 computational, 72–73, 140–152 CTP-functionalism, 183–185 direct-teleological, 165–167, 170–175, 178, 210 explanatory, xxv, 73–75, 85, 99–106, 109, 159, 182, 193, 200–202, 208, 211–213, 230 hybrid, 172–177, 210 intentional, xxv, 73–75, 85–89, 91–92, 105, 149 machine (see Machine functionalism; Functionalism, computational) and mechanism (see Mechanism, confused with functionalism) metaphysical, xxv–xxvi, 72–85, 92, 97–100, 105, 108–110, 124–127, 133, 139–142, 144, 150, 157, 159, 167, 171, 176–177, 181–182, 191, 211, 234 methodological, xxv, 73–75, 85, 106–108, 158–159 minimal (see CTP-functionalism) psychofunctionalism, 108, 156, 157,
158, 186 semantic, xxv, 73–75, 86–95, 140, 149 standard formulation, 81–85 strong sense, 79, 85, 93, 114, 144, 152, 155, 157, 159, 171, 183 teleofunctionalism, 164–172, 182, 242 teleological, 165–166, 172 (see also Functionalism, teleofunctionalism; Functionalism, direct-teleological; Functionalism, hybrid) theoretical, xxv, 73–75, 85–86, 92–100, 154–155, 158–159 varieties of, 71–78, 109, 137, 181 weak sense, 79, 85, 93, 144, 152 161 Functional specification, 199 de dicto, 82–85, 98, 108 de re, 82–85, 98 meaning fixing, 82–84, 98, 154–157 nonconventional, 130, 151, 154, 177 reference fixing, 82–84, 98, 157–159 Functions all the way down, xxvi, 182–191, 206 Generality (of psychology) argument, 161–164. See also Multiple realizability Generic Functionalism, 110, 125–139, 146, 152–155, 158, 160, 163, 165, 168, 171, 178–179, 183–186, 240 biological abstractness, 126, 133–134, 155–159, 163–169, 177–178, 184–186, 190–193 causal efficacy, 126–129, 148, 169, 177–178 objectivity, 126–131, 148, 151, 154–156, 164–165, 169–172, 178, 184–185 synchrony, 126–133, 168–171, 178 Gillett, Carl, 116–118, 121–124 Godfrey-Smith, Peter, 104, 108, 151, 153, 164, 166, 173–174 Gould, Stephen J., 224 Graham, George, 22–23, 66 Grain and granularity, 20–23, 26, 61, 69, 188, 219 grain project, 22–23, 66 Grantham, Todd, 102–103 Griffiths, Paul, 168 Güzeldere, Güven, 214–215, 218 Hardin, C. L., 65 Heil, John, 5–6, 116, 118, 121, 124 Hempel, Carl, 63, 195
292 Index
Heuristic identity theory, 32 Hill, Christopher, 14–15 Homology, 23–24, 240 Homunctional organization, 173, 187, 198, 201 Horgan, Terence, xxii, 16, 22–23, 66, 116–120, 128, 147, 161, 216 Hume, David, 174, 244, 245 Identity a posteriori or scientific, 37–38, 40, 43, 45, 51, 67 apparently contingent, xxiv, 40–44, 48, 51, 59, 66, 157, 241 brute, 36–37, 65 illusion of contingency (see Illusion of contingency; Identity, apparently contingent) necessary, xxiii, 36, 40–43, 45, 51, 67, 70 species-specific or local, 14–15, 19–20, 26–29, 30–31, 33, 69, 155–157, 241 Identity conditions, 45–51, 60, 79, 82, 171 of brain states, 50–59, 68–69 candidacy, 45–48 compatibility, 48–50, 60 of sensations, 50–51, 68–69 Identity Theorists’ Hall of Fame, 44 Illusion of contingency, 39–70. See also Identity, apparently contingent candidacy model, 49–51, 58–63, 66–68 epistemic model, 41–44, 59, 62, 68 Implementation, 111, 113, 117. See also Realization Individual differences. See Multiple realizability, individual and intraindividual differences Instantiation. See Realization Intentionality, xxii, 73, 87, 89, 91, 121, 149, 167 Intraindividual differences. See Multiple realizability, individual and intraindividual differences Introspection, 86, 133 Jackson, Frank, 37–38, 63, 128, 129, 132, 226 James, William, xiv Kim, Jaegwon, 2, 5–6, 10, 15, 28, 30, 33, 78, 95–96, 107, 110, 114, 116, 121–124, 142–143, 150–151, 160–162, 188, 208
Kim–Adams reply. See Multiple realizability, Kim–Adams reply Kirk, Robert, 217 Kripke, Saul, xxiv–xxv, 38–70, 156, 181, 216, 239 Language of thought, 85, 89–90 Lauder, George, 171 Laws of nature, 37, 102, 104, 158, 182, 194, 203, 207, 216 Leibniz’s Mill, 243 Levels and orders, 161–164, 187–188, 196–198, 200–201, 205, 208, 241 Levine, Joseph, xxv, 63–66, 161, 191, 195, 239 Lewis, David, 15–16, 26–28, 32, 71, 93–95, 98–99, 106, 154–156, 217 Lewontin, Richard, 224 Loar, Brian, xxii Locke, John, 47 Lycan, William, xxii, 23, 42–44, 72, 80, 84, 89–90, 96, 140, 162, 165–169, 172–174, 179, 182, 186–190, 201, 206–207 Machamer, Peter, 196–204 Machine functionalism, 72–73, 80, 94, 120, 131, 140–141, 149, 152. See also Functionalism, computational Machine program, 141–142, 151 Machine table, 94, 141–145. See also Machine program Marr, David, 117, 158 Materialism. See Naturalism McCauley, Robert, 5, 21, 32 McGinn, Colin, 33, 239 Mechanism and mechanistic explanation, xxvi, 52, 64, 102, 159, 181–182, 196–213, 222, 226–233, 241–245 confused with functionalism, xxvi, 211–214, 222, 235–240 schema, 198–199 sketches, 198–200, 228 Mental causation, xxiii, 3, 35, 70, 87, 127, 158, 161, 163, 173, 241. See also Causation Mereology, 121–122, 161–162, 197–198, 205–206, 241 Millikan, Ruth, 87, 90–93, 101–107, 113, 165–169 Moral value, 231–232, 237–238 Multiple realizability, xxiv, 1–39, 64, 69, 110, 117, 126, 133–136, 146, 157,
293 Index
166, 168–169, 177, 184, 186, 189–195, 222, 230, 236, 241 empathetic, 12–17, 69, 155 and identity theory, 30–33 individual and intraindividual differences, 16–17 Kim–Adams reply, 10–12, 16–17, 20–21, 23, 26–29, 33 radical MR, 6–11, 15, 17, 29–31, 134, 136, 151 SETI MR, 6, 11–12, 18–19, 21, 30–31 standard MR, 6–11, 15, 17, 21, 24, 29–31, 134–136, 151, 230 weak MR, 6, 11, 19, 21 Mundale, Jennifer, 5, 16, 21–22, 24–25, 51–52, 61 Nagel, Ernst, 195 Nagel, Thomas, xix–xx, 14, 239 Natural method, xv–xx. See also Naturalism Naturalism, i–xix, 36, 95, 124, 183, 185, 192 Naturalized metaphysics. See Naturalism Neander, Karen, 101, 107, 166 Nichols, Shaun, 102–103 No Ghosts Rule, xviii–xix Normativity, 90–91, 102–103, 152, 165, 167, 173–174, 210 biological, 165, 169, 172 Oppenheim, Paul, 63, 195 Orders. See Levels and orders Other minds. See Problem of other minds Pargetter, Robert, 169–170 Perry, John, 32 Persons and personal identity, xxvii, 48, 231–232, 241–246 Pettit, Philip, 128–129, 132 Physicalism. See Naturalism Pitt, David, xxii Place, Ullin T., 44 Plasticity, 17–18. See also Multiple realizability, individual and intraindividual differences Platitudes, 97–98, 154–155, 157, 245. See also Psychology, folk or common sense; Ramsey–Lewis method Poland, Jeffrey, 116–120, 124–125 Polger’s brain, 55–56 Pragmatism, xvii, 174, 201
Problem of inputs and outputs, 152, 187 chauvinism, 152, 186, 187, 189 liberalism, 152, 186, 187 Problem of other minds, 33–35, 235–239 Properties, causal theory of (CTP), 182–186, 206. See also Mechanism, confused with functionalism Propositional attitudes, 86, 96 Psychofunctionalism. See Functionalism, psychofunctionalism Psychology. See also Explanation, psychological; Generality (of psychology) argument analytic, 96, 99, 106, 154, 157, 186 autonomy of, xxvi, 190–196 (see also Autonomy) biopsychology, 101–104 evolutionary psychology, 101–104 folk or common sense, 96–99, 106, 154, 157, 186 scientific, 86, 96–99, 103–107, 154, 157, 186, 191 Psychophysiology, xv Psychosemantics, 86, 90 Putnam, Hilary, xxiii–xxiv, 3, 15, 22, 24, 28, 40, 61, 69, 72, 79–80, 88–89, 96, 106, 131, 140, 145, 156, 158, 177–182, 185–86, 191, 195, 203–208, 241–245 Putnam’s intuition, 3–4. See also Multiple realizability Qualia absent, 189, 214, 218, 232 alien, 214, 217, 218 dancing, 214, 217, 218 inverted, 189, 214, 217, 218 Ramsey–Lewis method, 94–100, 106–107, 151, 182–183 Ramsification. See Ramsey–Lewis method Realization, xxiv–xxv, 310–113, 116, 139–179 and describability, 143, 147–148, 228 dimensioned view, 124 flat view, 122–124, 144 functional, 36, 113, 178–179, 230 general metaphysical notion, 116–125 as the hardware/software relation, 112, 117 as having a function, 113–114
294 Index
Realization (cont.) as mapping, 148–149 (see also Functional equivalence) multiple (see Multiple realizability) not a single relation, 112–114, 178 orthodox view, 112–113, 140 realization theories, 119, 124 subset model, 122–124 transitivity of, 117, 147 Reduction, xxi, xxvi, 37–38, 126, 193–196, 201–205, 208, 209–210. See also Autonomy and identity, 193–194 and multiple realizability, 5, 30–32 new wave, 32, 195–96 Reflective equilibrium. See Natural method Representation, xxii, 93, 147, 167, 176 Rey, Georges, 75, 85, 96, 99–100 Rigid designators, 39–43, 61 Rorty, Richard, xvii Ryle, Gilbert, 78, 99 Salmon, Wesley, 202–207 Searle, John, xiii–xiv, 81, 105, 108, 131 Sejnowski, Terence, 54 Self. See Persons Semantics, functional role, 90, 93. See also Functionalism, semantic Sensations. See also Identity conditions, of sensations bundles, 243–246 eccentric, 33–35, 69 Shapiro, Lawrence, 5, 11, 208 Shoemaker, Sydney, xxii, 6, 79, 85, 91–99, 106, 114–116, 121–124, 144, 152–159, 182–186, 189, 205–206, 214, 217, 232 Sidelle, Alan, 45–48 Siewert, Charles, xxii Skepticism, xv, 36, 238–239 Skinner, B. F., 75 Smart, J. J. C., 2, 44, 78–79, 94, 97, 157 Sober, Elliot, 5, 81, 205 Sociobiology, 103 Species-specificity. See Identity, speciesspecific or local Structure of experience reply, 65, 67 Supervenience, 78, 114, 116, 121–122 Swampman, 170 Teleofunctionalism. See Functionalism, teleofunctionalism; Functionalism,
teleological; Functionalism, directteleological; Functionalism, hybrid Teleology, 100, 102, 164, 173 Thales, 40 Tienson, John, xxii, 116–120, 147 Topic neutral, 94–95 Turing machine, 141–143, 145, 149, 151–152, 206. See also Machine table Turing test, 131, 220, 235 Two-Levelism, 187–189 Tye, Michael, xxii, 82, 89, 132, 140 Van Gulick, Robert, 65, 75, 79, 85, 91–92, 108, 134, 144, 148, 152, 155, 159, 168, 172–173, 185, 189 Watson, John, 75 What-it-is-like, xix–xx Wilson, E. O., 103 Wilson, Robert, 116, 118, 121, 124 Wright, Larry, 101, 107 Zimboes. See Zombies, zimboes Zombies, xxvi, 35, 170, 213–24 and conceivability, 239 Dennett’s challenge, xxvi, 213–214, 218–219, 223, 233 Dennett’s challenge as a dilemma, 223, 227, 234–235 varieties of, 214–217, 225 zimboes, 219–223, 225 Zombie Scorecard, 215–217