Knowledge, Reasons, and Causes Gilbert H. Harman The Journal of Philosophy, Vol. 67, No. 21, Sixty-Seventh Annual Meeting of the American Philosophical Association Eastern Division. (Nov. 5, 1970), pp. 841-855. Stable URL: http://links.jstor.org/sici?sici=0022-362X%2819701105%2967%3A21%3C841%3AKRAC%3E2.0.CO%3B2-P The Journal of Philosophy is currently published by Journal of Philosophy, Inc..
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/jphil.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.
JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact
[email protected].
http://www.jstor.org Sat May 12 00:35:44 2007
THE JOURNAL OF PHILOSOPHY
KNOWLEDGE, REASONS, AND CAUSES "
P
HILOSOPHERS interested in the analysis of knowledge often attempt to divide and conquer. They distinguish three questions: (1) What is it for belief to be based on certain reasons? (2) When do reasons justify belief? and (3) What conditions must be added to the analysis of knowledge as justified true belief if the Gettier counterexamples are to be avoided? These philosophers try to answer each of these questions independently of the others, hoping that satisfaciory answers can be combined in an over-all theory of knowledge.2 I claim that a piecemeal approach cannot work. None of these questions can be answered apart from the others. In this paper I shall try to say how an attempt to analyze what it is for belief to be based on reasons becomes involved with questions about the goodness of reasons and the Gettier counterexamples. Furthermore, I shall suggest that explanation by reasons is not causal explanation. And finally I shall argue that the ordinary concept of knowledge
" T o be presented in an APA Symposium on Knowledge, December 27, 1970; commentators will be Manley Thompson and Edmund Gettier. For Thompson's paper, see this JOURNAL, this issue: 856-869; Gettier's paper is not available at this time. Research was supported in part by the National Science Foundation. 1 Edmund Gettier, "Is Justified T r u e Belief Knowledge?", Analysis, xx111.6, ns 96 (June 1963): 121-123. 2 For example, Keith Lehrer, "Knowledge, Truth, and Evidence," Analysis xxv.5, ns 107 (April 1965): 168-175; "Induction, Reason, and Consistency," British Journal for the Philosophy of Science, XXI, 1 (February 1970): 103-114; Lehrer's contribution to Marshall Swain, ed., Induction, Acceptance, and Rational Belief (Dordrecht: Reidel, 1970); and Keith Lehrer and Thomas Paxson, LXVI,8 (April Jr., "Knowledge: Undefeated Justified T r u e Belief," this JOURNAL, 24, 1969). See also, e.g., E. Sosa, "Propositional Knowledge," Philosophical Studies, xx, 3 (April 1969), esp. pp. 35-36; and Roderick M. Chisholm, Perceiving (Ithaca, N.Y.: Cornell, 1957); Theory of Knowledge (Englewood Cliffs, N.J.: Prentice-Hall, 1966); "On Empirical Evidence," in Lawrence Foster and J. W. Swanson, eds., Experience and Theory (Amherst: Univ. of Massachusetts Press, 1970), pp. 103-134. 841
8 4
THE JOURNAL OF PHILOSOPHY
plays an important role in common-sense psychological explanations and that it ought to be taken more seriously in theoretical psychology. I begin with an example. Albert believes that he will not pass his ethics course. He has excellent reasons for believing this: he failed the midterm examination, he has not been able to understand the lectures for several weeks, and the instructor is known to flunk a high percentage of his students. However, like many others, Albert does not appreciate the force of such reasons. On the other hand, he is influenced by something else. During a class discussion early in the term, he gave an emotional speech in favor of existentialism. Although his instructor said nothing at the time, the man seems to advocate linguistic analysis. ~ l b e r treasons that, since analytic philosophers hate existentialists, the instructor will remember his remarks, hold them against him, and therefore fail him in the course. Even though Albert believes that he will fail and has excellent reasons for believing this, he is not justified in believing it, and he cannot be said to know that he will fail.3 Reasons for believingsomething- must not be confused with reasons for which one believes. Reasons for which one believes are relevant to whether one is 3 I put the point this way, "Albert cannot be said to know that he will fail," because it seems misleading to say that Albert does not know that he will fail. Just as it would be misleading to say of Albert either that he realizes he will fail or that he does not realize that he will fail, it would also be wrong to say either that he knows he will fail or that he does not know that he will fail. Perhaps we can account for our inability to say either 'knows' or 'does not know' by distinguishing presupposition from assertion. In the present context, when one says that Albert knows or does not know that he will fail, one asserts that Albert thinks, or does not think, that he will fail. One presupposes (a) that Albert will fail and (b) that, if he should think that he will fail, he would know that he will fail. Since presupposition (b) is false, one should say neither that Albert knows nor that he does not know. Other uses of 'know' have different presuppositions. Sometimes one does not presuppose the truth of what is said to be or not to be known. For example, I might discuss with a representative of the tobacco industry whether scientists know that cigarettes cause lung cancer. I assert and he denies that the scientific evidence establishes a particular conclusion. We both presuppose that, if the evidence establishes or were to establish the conclusion, the scientists know or would know that cigarettes cause lung cancer. This distinction between presupposition and assertion may help to explain the occasional difficulty we find in reaching agreement about whether various examples are examples of knowledge. Someone might point to the fact that we should not say Albert does not know that he will fail. This, it might be suggested, shows that Albert does know after all. We can avoid the latter conclusion if we distinguish presupposition and assertion. With all that said, I shall follow the usual practice and ignore the distinction in this paper.
,
KNOWLEDGE, REASONS, AND CAUSES
843
justified in believing as one does. Reasons for believing something are not relevant unless they are also reasons for which one believes. I want to say what it is for certain reasons to be reasons for which a person believes something and what it is for certain reasons to be reasons that give a person knowledge. A familiar suggestion is that relevant reasons are those a person would offer if asked to justify his belief.4 This suggestion cannot be correct. Albert may offer good reasons, not because he thinks they are any good, but because he thinks they will convince his audience. Such reasons are not reasons for which he believes as he does, nor can they give him knowledge. Here is an example. Albert's adviser asks him to defend his view that he will fail the ethics course. Albert knows that his adviser is an idealist who doesn't believe that grades are determined by whether a teacher likes a student. He knows that his adviser supposes that grades are a good indication of how well a student has mastered the subject. Furthermore, he doubts that his adviser would believe that an analytic philosopher might dislike a student just because he favored existentialism. Therefore, Albert does not tell his adviser what are, in his opinion, the real reasons why he is going to fail. Instead Albert speaks of his poor performance on the midterm examination, his inability to follow the lectures, and the reputation of his instructor as a severe grader. Albert does not think these considerations are of any importance; but he knows that his adviser will find them convincing. Here is a clear case in which reasons put forward to justify a belief are not reasons for which it is believed. Nor is the analysis to be rescued by requiring that Albert be sincere. Being asked to justify his belief might lead Albert to reassess his reasons. This could lead him for the first time to appreciate his good reasons. H e would begin to see the significance of his performance on the midterm examination, his inability to understand the lectures, the instructor's reputation. Although he had believed for some time that he will fail, he would only then come to believe this for the good reasons he then states. Only then could he be said to know that he was going to fail. If he has not yet been asked to justify his belief, he does not believe it for good reasons and he cannot be said to know that it is true. If he were asked to justify his belief, he would at that point come up with good reasons, reasons that would at that point give him knowledge. I t follows that reasons for which a person believes something cannot be identified with 4
Lehrer, "Knowledge, Truth, and Evidence," p. 169.
844
THE JOURNAL OF PHILOSOPHY
reasons he would sincerely offer if asked to justify his belief. Nor can reasons that give a person knowledge be identified with reasons he would thus sincerely offer if asked to. But there is a more basic objection to the suggestion that relevant reasons are those a person would offer if asked. T h e suggestion rests on an identification of reasons with conscious reasons. A man's conscious reasons are those he can tell us about. T o equate reasons for which he believes something with reasons he can tell us about is to assume that reasons for which he believes something are conscious reasons. This is a mistake. T h e reasons for which people believe things are rarely conscious. People often believe things for good reasons, which give them knowledge, even though they cannot say what those reasons are. We can easily imagine Albert unaware of his real reasons for believing that he will fail. We might suppose that (contrary to the suppositions I have been making so far) Albert really does appreciate the significance of his failure to pass the midterm, his inability to understand the lectures, the reputation of the instructor. This leads him to realize that he will fail. But, not liking to think that he will fail because of his own incompetence, Albert prefers to think that he will fail because the instructor dislikes him. Therefore, Albert is not consciously aware of what the real reasons are for which he believes he will fail. He may even know that he will fail. T h e reasons for which he believes that he will fail can give him knowledge even though he cannot correctly identify those reasons. T h e same point can be made without appeal to a psychoanalytically flavored example. In most cases a person is unable to state his reasons in any sort of detail. At best he can give only the vaguest indication of the reasons that convince him. It is only in rare cases that we can tell a person's reasons from what he can say about them. Indeed, it is doubtful that a person can ever fully identify his reasons. This will not seem obvious if one thinks only of deductive reasons: reasons that deductively entail one's conclusion. Deductive "reasoning" can be fully specified; hence the idea that "deductive reasons" can be fully specified. Not so for inductive reasons. I t is doubtful that anyone has ever fully specified an actual piece of inductive reasoning, since it is unlikely that anyone could specify the total relevant evidence in any actual case. T h e difficulty is not simply that there is so much relevant evidence, but also that one cannot be sure whether various things should or should not be included in the evidence. One cannot always be sure what has influenced one's conclusion.
KNOWLEDGE, REASONS, AND CAUSES
845
Reflection on examples suggests that coming to believe something for certain reasons has something to do with why one comes to believe it. T h e reasons for which Albert believes as he does seem to depend on the explanation of his coming to believe as he does. These reasons can involve either his theory that the teacher dislikes him or his realization that he is not doing well in the course. What distinguishes these possibilities cannot be Albert's thinking that certain reasons are his reasons. In either case we may suppose that he thinks of the former reasons as his reasons. T h e difference seems to lie in why he believes as he does. In imagining that he believes as he does for the reasons involving his realization that he is philosophically incompetent, we ascribe to his belief a different explanation from that we ascribe to it when we say that he really believes as he does for the reasons involving his theory that the instructor has it in for existentialists. More generally, it is difficult to see how to imagine a difference in the reasons for which people believe as they do without imagining a difference in the explanation why they believe as they do. T h e same point holds for one's own self. When a person wonders whether a consideration represents one of his reasons, he wonders whether that consideration influenced his conclusion. But that is to wonder whether it has anything to do with why he believes as he does, with the explanation of his belief. In order to say what sort of explanation this is, we must distinguish two cases: that in which one originally comes to believe something and that in which one continues to believe something although one's reasons change. When a person first comes to believe something, why he believes it appears to be a function of how he came to believe it. T h e differences we imagine in imagining the two possibilities concerning Albert are differences in how Albert comes to believe that he will fail. But it is not in general true that the reasons for which one believes as one does are simply a matter of how one came to believe what one believes. When the reasons why one believes something change, the reasons for which one believes as one does also change. In that case, the explanation why one believes as one does has changed and is no longer simply a matter of how one came to believe as one does. Furthermore, there is a clear connection between reasons and reasoning. T o specify a man's reasons is to specify reasoning that leads to his conclusion. It is not enough to specify premises from which the conclusion may be derived. One must also specify how the conclusion is obtained from those premises. Two people, starting from the same premises and reaching the same conclusion can go by different routes. As a result they may believe the conclusion for
846
THE JOURNAL OF PHILOSOPHY
different reasons. If the first man's reasoning is sound but the second's is not, the first can believe for good reasons while the second believes for bad reasons. T h e first can know while the second fails to know, even though they both start from the same premises and reach the same conclusion. This suggests that a person's reasons are a function of his reasoning. T o say that a man believes something for certain reasans is to say that he believes it as the result of certain reasoning. A familiar objection is that one is rarely aware of the relevant reasoning. But awareness is not important. I claim that one believes something for certain reasons only if one believes it as the result of certain reasoning. I do not claim that one is aware that one believes something for certain reasons only if one is fully aware that one believes it as the result of relevant reasoning. After Freud and Wittgenstein it is no longer plausible to equate mental processes with conscious processes. Reasoning is in some sense a mental process. I t is rarely a conscious process. Even when one is aware of one's reasoning, one is rarely able to tell exactly what considerations have influenced one's conclusion. So far I have argued that to cite the reasons for which someone believes something is to explain why he believes as he does; I have claimed that when a person comes to believe something, the relevant sort of explanation tells how he came to believe as he does; and I have now proposed that the relevant sort of explanation explains his belief as the result of reasoning. What sort of explanation is this? One naturally thinks of causal explanation; but familiar theories of causation do not seem compatible with the idea that reasoning causes the resulting belief. Reasoning is a process, the last stage of which is belief in its conclusion. If reasoning caused the belief, the effect would be part of the cause. T h a t would violate the Humean principle that the cause must be distinct from the effect, although it is not obvious that the Humean principle ought to be accepted. Explanation by reasons might still be causal in a Humean sense, if reasoning is a causal process. But let me continue for a while to discuss the idea that the reasoning itself is the cause. Supposing that reasoning causes the belief in its conclusion also conflicts with a strict covering-law theory of causal explanation. On that theory, a cause is one of the initial conditions. A description of the effect must follow logically from a description of the initial conditions plus a statement of the appropriate laws, without following
KNOWLEDGE, REASONS, AND CAUSES
847
logically from the description of the initial conditions alone.= But the relevant description of one's reasoning entails that one believes its conclusion. So one's reasoning cannot be cited as the causal explanation of one's belief if the covering-law theory of causal explanation is accepted. On the other hand, if we were to suppose that beliefs are caused by reasoning, we could account for certain facts about reasons for which people believe things. For example, a man can come to believe things for several sets of reasons, each sufficient in itself to have led him to that conclusion. Some of these sets may comprise bad reasons for his conclusion. He can still be warranted in believing as he does as long as some of his reasons are good reasons. Such good reasons can give him knowledge, although he also believes as he does for bad reasons. One can know even if one has inferred one's conclusion from false premises-if one has also inferred it from true premises. Such cases make sense if reasons are causes, for then they are examples of causal overdetermination or of multiple causation. Ordinary usage is unclear. Suppose that at first Albert thinks he will fail because of the teacher's hatred for existentialists. Then he learns that the teacher assigns all final grades mechanically, on the basis of examination performance. Albert continues to believe that he will fail, although the reasons for which he believes change. We would cite his new reasons as the explanation why he now believes as he does; and we would grant that these new reasons can give him knowledge. But if now asked what caused his belief, we would cite what originally led to his belief (if we would cite anything). If asked what the cause of his belief is now or what now causes his belief, we might cite his new reasons, although I suspect that this use of 'cause' is deviant. Other things equal, if a person believes a conclusion for certain reasons and becomes doubtful about those reasons, he becomes doubtful about the conclusion. "Other things equal" is meant to rule out the possibility just described in which one acquires new reasons as one comes to doubt the old. T h e phrase must also be used to rule out overdetermination and the analogue of multiple causation. In case there are several sets of reasons for which someone 5 Carl G. Hempel, "Aspects of Scientific Explanation," Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, (New York: Free Press, 1965), esp. pp. 337-338. Hempel does not require that causal explanations take this strict form. He points out that relevant antecedent conditions and laws are often left indefinite (pp. 347-354). His own account of explanation by reasons is on pp. 463487. I am obviously indebted to that account, although I think my final account is not compatible with it.
848
THE JOURNAL OF PHILOSOPHY
believes something, he must become doubtful about all sets before becoming doubtful about his belief. T h e relevance of any of the sets is this: if he became doubtful of all the other sets, his belief would rest crucially on that set so that, if he should then become doubtful of it, he will become doubtful of his belief, other things equal. Any such set of reasons supports someone's belief in the way that a subset of its legs may support a table. Reasons can even support a man's belief and give him knowledge in cases in which we would be reluctant to describe them as reasons why he believes as he does or as reasons for which he believes as he does. Keith Lehrer describes the following example. Larry's wife, Mabel, has been accused of embezzling a large sum of money from the bank in which she works as a teller. Larry loves Mabel dearly and thinks he knows her as well as anyone can know anyone else. He cannot believe that Mabel could have embezzled money. I n order to save her from jail he sets out to prove Mabel's innocence by finding the real embezzler. He is successful and is able to uncover evidence sufficient to prove that the bank manager took the money. T h e evidence Larry collects supports his belief in Mabel's innocence, and it can give him knowledge where before he had only firm conviction. However, it need not add any force to his conviction, which might already be so strong that nothing could shake it. We would not say that Larry's belief in Mabel's innocence is even partially caused by the new reasons he presents to establish her innocence. Lehrer points out that we would be equally reluctant to say that these new reasons are among the reasons why or for which Larry believes as he does. Reasons for which something is believed must at least partially explain why it is believed. Reasons that give knowledge need not partially explain the relevant belief, as long as they potentially explain it-that is, as long as they offer the sort of support roughly described above. If Larry were to stop loving Mabel and were to come to doubt that he knew her well at all, he would still believe her innocent of the embezzlement; and his belief woud then be explained by his reasons for believing the bank manager guilty. I have been asking whether the relevant sort of explanation is causal. I noted that to take reasoning to be a cause of belief in its conclusion conflicts with Humean accounts of causation, including the covering-law theory. Ordinary language is less clear here. Sometimes we do say that reasoning causes one to believe its conclusions. On the other hand, new reasoning for an old conclusion can afterward explain one's belief although it is not clear that we would say it caused one's belief.
KNOWLEDGE, REASONS, AND CAUSES
849
To explain someone's belief, we need only cite the reasoning that led him to his conclusion. As noted above, this would be causal explanation if it assumed that reasoning is a causal process. But we make no such assumption when we offer such explanation. That a man accepts the premises does not determine that he will accept the conclusion. That those premises provide a good reason for accepting the conclusion does not ensure that he will think u p the reasoning that shows this. It may be true that one step of reasoning is in fact part of the cause of the next step. But we make no such causal assumption when we cite a person's reasoning as the explanation of his belief. Our explanation can be true even if the causal assumption is not.6 Explaining why someone believes something is like explaining why a nondeterministic automaton is in a particular state. Various moves are possible at any moment. I t is not assumed that it is determined which move one will make. I t is not even assumed that various moves are more or less probable. It is assumed only that they are possible. We can explain how one got to wherever it was one got without implying that one had to get there. Putnam and others compare people to automata in order to shed light on psychological states and the mind-body problem." Automata are sets of states functionally related to each other and to input and output. These states may be realized in various ways. In one machine states will be instantiated in tubes and wires; in another machine the same states will be instantiated in transistors and printed circuits. A physical mechanism realizes or instantiates a given automaton if and only if there are (properly behaved) states of that physical mechanism that stand in the appropriate functional relationships. Just as states of an automaton are to be defined in terms of functional relationships rather than in terms of the exact nature of their physical realization, so too for psychological states. T h e same psychological states may be differently realized in different people or even in the same person at different times. These states are defined in terms of their functional relationships to other psychological states, as in reasoning; to input, 6 Despite an apparent disagreement on this one point, I owe a tremendous amount to Donald Davidson's writings and to many conversations I have had with him. I list some of the relevant articles: "Actions, Reasons and Causes," this JOURNAL, LX, 23 (Nov. 7, 1963): 685-700); "Causal Relations," ibid., LXIV, 21 (Nov. 9, 1967): 691-703; "The Individuation of Events," in Nicholas Rescher, ed., Essays i n Honor of Carl G. Hempel (Dordrecht: Reidel, 1969): 216-234; "Mental Events," in Foster and Swanson, op. cit. 7 Hilary Putnam, "Minds and Machines," in Sidney Hook, ed.. Dimensions of Mind (New York: ~w Press, 1960). Jerry A. Fodor, "Explanations in Psychology," in Max Black, ed.,Philosophy i n America (Ithaca, N.Y.: Cornell, 1965).
850
THE JOURNAL OF PHILOSOPHY
as in responses to observation; and to output, as in intentional action. A simple psychological model of a person would be an automaton with two basic sorts of internal states, corresponding to beliefs and desires. Perhaps there is a system of representation for things believed and desired. Representations of things believed are stored in one "place," and representations of things desired are stored in another. Observation can lead to the formation of new representations to be stored as beliefs. Things desired with things believed can lead to action output. And so on. More sophisticated models are possible. Indeed, one might even take a person himself to be the instantiation of a sufficiently complex automaton. Whatever one thinks of talk of psychological models, one must admit that there is a parallel between the explanation why a person believes as he does as the result of reasoning and the explanation why a certain automaton is in a particular state. Furthermore, as I shall now argue, the relevant sort of automaton will be nondeterministic. T o specify a nondeterministic automaton is to specify the possible states of the automaton, possible input, and possible output, and also to specify which states and what output can follow any given state and input. If one also specifies the likelihood that any state and input will lead to another state and output, one has specified a probabilistic automaton. If at most one state and output can follow any given state and input, one has specified a deterministic automaton. T h a t is, a probabilistic automaton is a special case of a nondeterministic automaton; and a deterministic automaton is a special case of a probabilistic automaton. If for purposes of psychological explanation we conceive a person as an automaton, it is important that we do not conceive him as a deterministic or even probabilistic automaton. We do not believe that certain psychological states must be followed by others. Nor do we even have views about the relative likelihood that certain psychological states may be follo~vedby others. Ordinary psychological explanation is neither deterministic nor probabilistic. There may or may not be an underlying neurophysiological determinism. I do not know what to make of quantum physics; and the experts do not seem to agree. I am talking about psychological determinism. A physically determined device can instantiate a nondeterministic automaton, where that automaton is not also a deterministic automaton. My point is that explanation that refers to reasoning does not presuppose either an underlying determinism or
KNOWLEDGE, REASONS, AND CAUSES
851
an underlying indeterminism. I t makes no presupposition either way. I t is helpful to think here of a simple probabilistic automaton, such as a roulette wheel. We assume that the wheel can take any of a number of states, all equally probable. T h e wheel instantiates a purely probabilistic automaton. Perhaps, under some complex description, it also instantiates a deterministic automaton. Our explanations of various outcomes presuppose the probabilistic way of looking at the wheel without committing us either way with respect to the deterministic viewpoint. Similarly, explanations of belief as the result of reasoning are nondeterministic without committing us either way with respect to an underlying determinism. Reasons may or may not be causes; but explanation by reasons is not causal or deterministic explanation. It describes the sequence of considerations that led to belief in a conclusion without supposing the sequence was determined. This is uncongenial to the covering-law theory of explanation. But that theory always seemed implausible when applied to ordinary psychological explanation. If we continued to accept it, that was because we did not like the apparent alternatives. Seeing a person as a nondeterministic automaton helps us to see why the covering-law theory of explanation breaks down in this case. It breaks down, not because psych~logicalexplanation appeals to a mysterious kind of Verstehen, but because it is nondeterministic. When one considers that there can be a kind of explanation of why a purely nondeterministic automaton is in a particular state, one can see that psychological explanation can be a perfectly intelligible kind of explanation without being deterministic. Reasoning is a mental or psychological process. Since psychological states and processes are functionally defined, to say what reasoning is to say how it functions psychologically. T o do that is to say how ascriptions of reasoning function in our accounts of people. I want to argue that the function of reasoning in giving a person knowledge is particularly relevant. That is why I think that the ordinary concept of knowledge plays an important role in common-sense psychological views and that it ought to be taken more seriously in theoretical psychology. Consider the difficulties facing a philosopher who wishes to discover principles of warranted inductive reasoning. He soon discovers that it is extremely difficult to reach any sort of agreement on cases if one attempts to reach such agreement directly. Some philosophers even deny that there are any cases in which inductive
852
THE JOURNAL OF PHILOSOPHY
reasoning warrants belief in a conclusion. On the other hand, philosophers have much less difficulty in reaching agreement about whether various examples are cases of knowledge. Nota bene: I am not saying that philosophers always agree; I am only saying that there is less disagreement about possible examples of knowledge than there is about possible examples of warranted inductive inference, when such examples are described for our direct intuitive judgment. Therefore, we can use our intuitive judgments about knowledge in order to get clearer about reasoning. Reasoning warrants belief in its conclusion if that reasoning could give a person knowledge of its conclusion. Since we can agree that inductive reasoning sometimes does give a person knowledge, we can agree that inductive reasoning does sometimes warrant belief in its conclusion. Through consideration of various examples we learn that such reasoning tends to infer the most coherent relevant total explanatory a c c o ~ n t s .A~ conclusion is warranted if it is included in a warranted total account. Here is one way in which a theory of reasoning makes use of our ordinary concept of knowledge. Edmund Gettier and others have described a number of examples which establish that knowledge is not simply justified true be1ief.O For example, from premises one knows to be true one might inductively infer something false and be justified in doing so. One might go on to infer something true from that first conclusion and still be justified. One's second conclusion is true and one is justified in believing it; but one does not know that it is true. Given the theory of reasoning mentioned in the previous paragraph, such examples lead us to accept roughly the following principle: (P) Reasoning that involves any false conclusions, intermediate or final, cannot give one knowledge. Many Gettier examples are not obviously accounted for by PI since it is not always evident that there has been any relevant reasoning. We can account for such cases by means of principle P if we assume that the relevant reasoning has occurred. Notice that it is not easy to decide directly whether reasoning has occurred in various examples. It is much easier to decide whether the examples are examples of knowledge and also whether the Gettier effect is operating. This suggests that we use intuitions about knowledge and See my "Induction," in Swain, op. cit. I n addition to articles mentioned in footnotes 1 and 4, see Brian Skyrms, "The Explication of 'X Knows that p'," this JOURNAL, LxIv, 12 Uune 22, 1967): 373-389. 8
KNOWLEDGE, REASONS, AND CAUSES
853
the Gettier effect to decide when reasoning has occurred and what reasoning there has been. T h a t would be to treat reasoning as a functionally defined process that is partly specified in terms of its role in giving a person knowledge. For example, a man can come to know something as the result of being told it by someone else. He might be equally justified in believing what he is told whether or not the man telling him believes what he says. But even if what is said is true, the first man does not come to know it if the second man is not saying what he believes. Here is a Gettier example where there is no obvious reasoning on the first man's part. However, if we ascribe reasoning to him we can account for our intuitions about knowledge. Roughly speaking, such reasoning would infer that the speaker says what he does because he believes it, he believes as he does because he has reason to believe it, he has such reason because of the truth of what he believes. If the speaker does not believe what he says, the first man's reasoning infers something false; so, by principle (P), his reasoning cannot give him knowledge. Ascribing this reasoning to the first man enables us to account for our intuitions about knowledge by means of a simple principle. A functional account of reasoning enables us to make sense of such ascription. Words like 'reasoning', 'argument', and 'inference' are ambiguous. They may refer to a process of reasoning, argument, or inference; or they may refer to an abstract structure consisting of certain propositions as premises, others as conclusions, perhaps others as intermediate steps. A functional account of reasoning says how a mental or neurophysiological process can be a process of reasoning by virtue of the way it functions. That is, a functional account says how the functioning of such a process allows it to be correlated with the reasoning, taken to be an abstract inference, which the process instantiates. T h e relevant correlation is a mapping F from mental or neurophysiological processes to abstract structures of inference. If x is a process in the domain of F, then F ( x ) is the (abstract) reasoning that x instantiates. Such a mapping F is a reasoning instantiator. T o give a functional account of reasoning is to say which of the infinitely many possible mappings from processes to abstract inferences can serve as reasoning instantiators. For example, the conclusions arrived at by reasoning must be beliefs reached by the process of reasoning. In other words, we are interested only in mappings F such that for any x in the domain of F there is a proposition p that is contained in the con-
854
THE JOURNAL OF PHILOSOPHY
clusion of F ( x ) and x leads to belief in p (or would so lead if p were not already believed). Furthermore, I have been arguing that reasoning must be partially specified in terms of its role in giving a person knowledge. F must be such that whether process x can give knowledge depends among other things on the abstract reasoning F(x). For example, x can lead one to be justified in believing p only if the reasons specifying F ( x ) would warrant one in believing p. The extent to which possible knowledge of p depends on whether one knows other things that one believes must be reflected in the fact that these beliefs appear as premises, intermediate steps, or conclusions in F(x). Finally, I suggest that it must be possible to explicate Gettier's examples by means of the relevant F ( x ) and principle P . I claim that ordinary talk about reasons and reasoning is to be explicated by way of the notion of a reasoning instantiator. T o ascribe reasoning r to someone is to presuppose the existence of a reasoning instantiator F and to claim that his belief resulted from a process x such that F ( x ) = r. I have already argued that explanation by reasons involves such ascription of reasoning. If I am right about both these p i n t s , it follows that the ordinary concept of knowledge plays an important role in common-sense explanation by reasons. Nothing in this account keeps it from being extended to cover cases in which one "directly" perceives that something is the case. Here too there are Gettier examples that can be accounted for if reasoning of the appropriate sort is ascribed to the perceiver. In a familiar example, a man looks and acquires the belief that there is a candle about twelve feet in front of him. Indeed there is; but a mirror intervenes between that candle and the man in question. What he sees is the reflection of a similar candle off to one side. He is justified in believing that there is a candle about twelve feet in front of him, and that belief is true. But he does not come to know the candle is there. This Gettier-type case is accounted for if the reasoning instantiator ascribes to the man in question reasoning that infers that things look the way they do (or that the stimulation on his eye is as it is) because there is a candle twelve feet in front of him, etc. There is no requirement that the man be aware that things are looking a certain way to him, or that the stimulation on his eye has the relevant character. A good reasoning instantiator will ascribe such reasoning to him in this and similar cases. This simplifies the account of observational knowledge and unifies it with the rest of the theory of knowledge.
KNOWLEDGE, REASONS, AND CAUSES
as
Here one might object that this is to extend the notion of reasoning far beyond anything permitted in ordinary usage. But ordinary usage even permits us to ascribe calculation and reasoning to inanimate machines (computers). And ascription of inference and reasoning to the perceiver will come as no surprise to anyone familiar with the psychology of perception. The psychologist's talk of cues, etc. is to be justified by way of a functional account of the reasoning process extended as I have just suggested. It is important that this functional account defines reasoning in terms of its function in yielding knowledge in the ordinary sense. T o make sense of visual cues and related psychological phenomena, one must take the ordinary concept of knowledge seriously. At the beginning of this paper I distinguished three questions which I now repeat: (1) What is it for belief to be based on certain reasons? (2) When do reasons justify belief? and (3) What conditions must be added to the analysis of knowledge as justified true belief if the Gettier counterexamples are to be avoided? I have argued that these questions must be answered together. That is, I have argued that they are to be answered through the specification (in more or less detail) of a reasoning instantiator F in such a way that all three questions are answered with reference to F and principle P. This leads to a theory of knowledge not subject to any simple kind of counterexample. One cannot show that the resulting answers to the three questions conflict with ordinary judgments about when people know things, since the theory makes use of these ordinary judgments in order to decide when inference occurs and what the criteria of good inference are. This cannot seem to be a defect unless one wrongly supposes that there is some way to answer one of these three questions independently of how one answers the others. T o sum up: I have been interested in reasons for which a person believes something and in reasons that give a person knowledge. I have argued that both sorts of reasons are constituted by reasoning that can be ascribed to the person in question. Explanation by reasons may or may not be causal; but it is not deterministic explanation. Such explanation requires a functional account of reasoning, according to which an important function of reasoning is to give a person knowledge. GILBERT H. HARMAN
Princeton University