SKEPTICISM AND NATURALIZED EPISTEMOLOGY DOUGLAS G. WINBLAD

Naturalized epistemology as Quine conceives it is a branch of natural science. Explanation, not justification, is its raison d~tre. Its explanandum is the acquisition of human knowledge, which for Quine is epitomized even exhausted by, natural science - - physics, broadly construed, the science for which Descartes tried to provide a philosophical foundation. Quine does not think science needs a foundation, nor, like Hume, does he think one can be devised. His approach to epistemology is of a piece with his rejection of analyticity. If epistemology is thought of as the analysis of the concepts of science, it appears as an autonomous discipline. In giving up the notion of analyticity, Quine erases the boundary between science and philosophy, and abandons, at least in theory, the methods of conceptual analysis. Nowhere are the results of this approach more striking than in his treatment of Cartesian skepticism. For Quine, "sceptical doubts are scientific doubts,"~ and can therefore be quieted by scientific means. Consider, for example, the skeptical possibility that one is now dreaming. Quine responds to dream skepticism by saying we need not take it seriously as long as predictions extracted from current physical theory are more successful than ones based on dreams. In effect, Quine treats skeptical counterpossibilities as rival empirical hypotheses, which, because they now possess less predictive power than contemporary science, need not command our attention. In Quine's eyes, skeptics are simply "overreacting. ''2 But this rejoinder will hardly convince the skeptic who is not ready to accept the standards by which Quinejudges him. From the skeptics point of view, the best scientific theory may simply not be good enough. The question arises, then, whether some form of naturalized epistemology other than Quine's fares better with skepticism. In this connection, we shall examine Goldman's epistemics. I. GOLDMAN~S EPISTEMICS: SOME P R E L I M I N A R I E S While allying himself with naturalized epistemology, Goldman acknowledges the inadequacy of Quine's reply to the skeptic. Since, in Goldman's view, "Some, but not all, skeptical doubts are scientific

99

DOUGLAS G. WINBLAD doubts, ''3 any adequate response to skepticism must involve some conceptual analysis. In particular, for Goldman, doubts based on our apparent inability to exclude certain logical possibilities cannot be dealt with properly except by means of a satisfactory analysis of knowledge. Goldman's attitude toward conceptual analysis is distinguished by two important features. Firstly, despite the fact that he finds it doubtful whether necessary and sufficient conditions for many terms are to be had, he proceeds by trying to give them. Secondly, Goldman endorses what he calls the "Goodman-Rawls conception of 'considered judgments in reflective equilibrium'," according to which "initial intuitions are not final, TM but can be discarded in the interests of theoretical economy in something like the way recalcitrant scientific data sometimes areP Goldman embraces the standard necessary conditions for S's knowing that p: S believes that p; S's believing p is justified; and p is true. His analysis also requires, however, that S knows that p only if S's belief that p is acquired by means of reliable cognitive processes. There are two kinds of reliability at issue here, which he terms global and local Local reliability concerns a process's reliability, not with respect to bodies of beliefs, but with respect to particular beliefs. Goldman's local reliability requirement has it that S knows that p only if there are no relevant alternative situations in which p would be false but the process by means of which S actually acquired the belief would "cause" S to believe p anyway. Only if there are no such relevant alternatives is a process locally reliable for S with respect to p. It is important to note that the relevant alternative situations are counterfactually specified, but do not include all possible situations in which p is false. They only include alternatives to the truth of p that are relevant to the case under consideration. Since for Goldman the local reliability condition is a necessary condition for knowledge, but not for justifiedness, in his view S can be justified in believing p even if the process that caused S to believe p is not locally reliable. Global reliability, however, is in his view a necessary condition for justifiedness. Roughly, a process is globally reliable just in case it has a propensity to generate a greater number of true beliefs than false ones in what Goldman calls normal worlds. Normal worlds are those that conform to our general beliefs about the world, regardless of what the world is really like. Thus, the actual world may not be a normal world. Goldman's analysis of justification is complex. He begins with the "framework principle" below:

100

SKEPTICISM AND NATURALIZED EPISTEMOLOGY S's believing p at t is justified if and only if (a) S's believing p at t is permitted by a right system of J-rules [justificatory rules], and (b) this permission is not undermined by S's cognitive state a t t. 6

He then introduces a non-verificationist "criterion of rightness" schema for systems of J-rules: A J-rule system R is right if and only if R permits certain (basic) psychological processes, and the instantiation of these processes would result in a truth ratio of beliefs that meets some specific high threshold (greater than .50). 7 In conjunction with the framework principle, this criterion schema captures the global reliability requirement in Goldman's theory. It also enables him to bridge the gap between traditional epistemology and cognitive psychology. If one construes epistemology as being normative, in the sense that, like moral theory, its task is at least in part that of specifying principles for appraising the performances or traits of agents, 8 and one accepts Goldman's strategy of identifying epistemic principles by means of a reliability criterion, one will likely conclude that epistemology must bind itself to cognitive psychology. For it seems plausible to think there is no other way the epistemologist can determine whether the cognitive processes to which the rules in a candidate J-rule system refer actually are globally reliable. We began this sketch of Goldman's theory of knowledge by distinguishing his approach from Quine's. There are many respects, of course, in which they are similar. Both require attention to the results of empirical psychology - - behavioral psychology in Quine's case, cognitive psychology in Goldman's. And both exhibit a certain sort of procedural circularity. Quine's inquiry is an instance of its object: scientific theorizing. Goldman's epistemologist employs the same cognitive processes he is studying in order to determine whether they are reliable. But whereas a commitment to naturalism makes it difficult for the Quinean to develop a normative epistemology, Goldman's commitment to normativity is at least in part what leads him to take a naturalistic turn. Quine is constrained by the Humean structure against deriving 'ought' from 'is'. The Kantian dictum that 'ought' implies 'can', however, forces Goldman to try to determine whether we possess the ability to follow certain epistemic rules before deciding whether they apply to us. 9 101

DOUGLAS G. WINBLAD The issue of normativity here is most pronounced with regard to skepticism. As we have seen, Goldman thinks certain paths to skepticism can be closed only by an adequate analysis of knowledge. If, given his analysis, it is possible for us to know certain things, and if his analysis is correct, then varieties of skepticism according to which we cannot know anything should be rejected. There are other forms of skepticism, however, with respect to which Goldman claims the conceptual analysis component of his theory is neutral. As he sees it, his analysis allows both for the logical possibility that we have first and second order knowledge, and for the logical possibility that we do not. For Goldman, whether we possess any knowledge at all depends on the nature of the processes by means of which we acquire and "maintain" our beliefs. In his view, the only way we can learn enough about these processes to tell is to engage in empirical investigation. To this extent, Goldman, like Quine, tries to turn the skeptical problematic into an empirical matter. II. THE PROBLEM OF INDUCTION REVISITED In Quine's words, "If the epistemologist's goal is validation of the grounds of empirical science, he defeats his purpose by using psychology or other empirical science in the validation. ''~~ Convinced that "The Humean predicament is the human predicament, '''~ he responds by simply "granting the efficacy of induction" and then employing it in order to "explain why induction is as efficacious as it is."12 To proceed this way is not to solve the problem of induction, but to bypass it altogether. Goldman, however, claims his own approach to justification provides a solution to the problem. 13He divides the traditional riddle of induction into two problems: that of whether we can have justified beliefs based on inductive inference, and that of whether we can have a justified belief that we can have justified beliefs based on inductive inference. He suggests Hume confused these problems with one another, or at least held that a positive answer to the first depends on a positive answer to the second. Goldman thinks there is no such dependence, and tries to give positive solutions to both problems. According to his reliability theory ofjustifiedness, what ultimately endows beliefs produced by a process with justification is the global reliability of the process. Thus, in his view, it is logically possible for us to have justified beliefs obtained by means of an inductive process, since it is logically possible for us to have a sufficiently reliable inductive process at our disposal. In the event that we do have such a process in our cognitive repertoire, Goldman's reliabilist rightness criterion will pick out a right system of J-rules in which there are rules that permit one to possess beliefs produced by that process. And the framework

102

SKEPTICISM AND NATURALIZED EPISTEMOLOGY principle will then yield the result that such beliefs are justified. Of course, whether we have access to such a process is for G o l d m a n an empirical matter. How, then, can we determine whether one is actually available to us? By using the very inductive process the reliability of which we are attempting to assess? Goldman's answer is this: "If we stick to the f r a m e w o r k principle, there is no reason why a belief in the justification-conferring power of induction could not be justified as a result of the self-same inductive process! ''14 G o l d m a n acknowledges that Humeans will object that there is such a reason, namely, that justifying the second order belief in the justification conferring power of induction in this manner is viciously circular. Relative to his system, G o l d m a n maintains, giving in to this objection would require "a further restriction on justifiedness," such as "No belief about the permissibility of a process is justified if the belief results from the self-same process." G o l d m a n resists such a restriction for two reasons. He insists the restriction in question "seems quite arbitrary", and alleges that accepting it "leads to the dilemma posed by Lewis Carroll in " W h a t the Tortoise Said to Achilles.'"~5 Let us examine these claims in turn. Earlier we noted that G o l d m a n ' s approach to conceptual analysis allows him to discard certain intuitions in the interests of theoretical economy. Presumably some intuitions are so central, however, that virtually no proper reflective equilibrium can be reached that does not preserve them. Developing criteria for determining which intuitions must be, and which need not be saved is no simple task. In any case, what of the intuition fairly c o m m o n among philosophers - - that a circular justification is no justification at all? Suppose we can find no reason why our concept of justification should not apply to circular reasoning independent of our intuition that it does not. Would it then follow that an adequate analysis of justification need not take account of this intuition? Suppose it is in fact a necessary condition for the correct application of our concept of justification to certain inferences that they not be circular. T h a t the concept possesses this feature may in some sense be arbitrary, but why should this matter? F o r the realistic conceptual analyst, the concept is what it is and not another thing. Of course, like measurements in a laboratory,, intuitions may err. This possibility undergirds the realist rationale for jettisoning intuitions and d a t a that conflict with otherwise successful theories. Nevertheless, whatever its shortcomings, there is reason to think the restriction on justifiedness we are considering is not wholly unmotivated. This may be seen by reflecting on the role of justification. In at least a wide range of cases, justifications serve to support claims or views that 103

DOUGLAS G. WINBLAD are open to question before they have been justified. Proofs in mathematics, for example, c o m m o n l y function to establish the theoremhood of statements regarding the truth of which mathematicians initially suspend judgment. Of course, new proofs are also often devised for old theorems widely acknowledged as such. But this does not tell against the importance of cases in which a conclusion is at issue prior to its demonstration. Now if we question whether induction is reliable, it is difficult to see how we can satisfy ourselves that it is solely by means of an inductive argument to that effect. As long as we are unwilling to recognize any argument of a certain type as legitimate, consistency dictates that we refuse to accept an argument for the legitimacy of such arguments that belongs to that very type. This is the predicament someone is in who has skeptical doubts about whether induction is acceptable. 16 The question of how one could respond to such a person leads us to Goldman's second point. Recall that G o l d m a n claims that if one does not countenance justifications of the sort in question, one ends up with the sort of d i l e m m a Carroll describes in his tale of Achilles and the Tortoise. 17 In effect, the tortoise refuses to draw the conclusion 'q' from the permises 'p -- q' and 'p', even though he accepts these premises, demanding that a further premise, 'p - - ((p -- q) -- q)', be introduced first. When it is, he affirms it along with the earlier premises, but remains unwilling to conclude that 'q' is true, insisting on the introduction of still another premise, and so on. One can read Carroll's tale as a meditation on the difference between premises and rules of inference. Goldman, however, seems to believe it expresses the worry that one cannot give a justification of modus ponens that is not circular. ~8 Relative to this interpretation, one might see the tortoise's insistence on further premises as an impossible demand that one step outside the circle. If this situation worries Carroll, presumably it does so because he finds circular justifications suspect. But in Goldman's view there is no cause for alarm here. " W o u l d one say that a person could not be justified in believing in the validity of modus ponens if he used modus ponens to arrive at this belief?." G o l d m a n asks.19 Let us try to put a finer point on the matter by supposing we are not talking about a situation in which someone examining a formal system in which there is an inference rule corresponding to modus ponens gives a metalogical argument that has the form 'p', 'If p then q', 'Therefore, q', to the effect that the rule is truth-preserving. Let us suppose instead that we are talking about someone who questions whether any arguments that have the form of the aforementioned metalogical argument are valid. And let us suppose 104

SKEPTICISM AND NATURALIZED EPISTEMOLOGY further that this person goes on to produce an argument to the effect that they are that has this very form. W o u l d we want to say in this case that the person is justified in believing the truth of his conclusion? It may be that one can justify neither deduction nor induction along these lines. Cases in which one gives a justification in the metalanguage for an object language rule seem less problematic, for in such cases circularity does not rear its head in any straightforward way. But once one really questions the validity of an argument form as such - - instances, of which may occur in either object or metalanguages - - a circular justification remains unsatisfactory. Of course, most of us are able to and do recognize valid deductive and strong inductive arguments when we see them. The real worry behind Carroll's tale m a y be that if we encounter someone who refuses or is unable to do the same, nothing we say may be able to bring them to respond the way we do. There may be minds in which even the light of logical necessity cannot shine. To say that one can justify neither deduction nor induction in the sense in which we are speaking of doing so is not to say there are no valid deductive or strong inductive arguments. It does not conflict with the view that some conclusions are justified and others are not. Nor does it require one to have beliefs about the nature of one's inferential processes in order to use them in acquiring justified beliefs. While, as G o l d m a n suggests, a skeptic would favor the view that deductive reasoning has no justification over the view that it has a deductive or nondeductive one, 2~ one need not be a skeptic to favor it. While it may not satisfy the skeptic, our inability to justify fully the forms of inference that confer justification on our beliefs may be in other respects unproblematic. One can take something like the Quinean approach: give up the project of justification and take the inferential procedures for granted - - simply use them, as most of us do. Let us set these considerations aside, however, and imagine we have chosen instead to use an inductive process to determine whether it is reliable. If it is reliable, chances are turning it on itself will issue in a positive verdict. But if it is not reliable, what then? Given G o l d m a n ' s analysis of justification, if by means of our inductive process we were to acquire the true belief that the process has a propensity to produce no more than fifty percent true beliefs, the belief would not be justified. Because it would be the product of an unreliable process, no right system of J-rules would permit it. And the framework principle would then imply that it would not be justified. Suppose, however, that our inductive process is reliable. Relative to G o l d m a n ' s account, could it produce in us a justified belief to the effect that it is not reliable? Clearly it could only do this if it is not completely reliable. 105

DOUGLAS G. WINBLAD If a reliable inductive process led us to believe the process is unreliable, however, the second clause of Goldman's framework principle would come into play to undermine the permittedness, and hencejustifiedness, of the belief. F o r Goldman, a belief's permittedness is undermined if the cognizer is permitted to believe that it is not permitted, believes that it is not permitted, or believes that certain conditions are not satisfied, where those conditions are in fact necessary for belief permissibility3~ Were we to endorse Goldman's reliability theory and apply it to the inductively produced belief that induction is unreliable, presumably we would be permitted to believe this belief is not permitted, thus undermining its justifiedness. A n d his account, regardless of whether we embrace it, implies that if we are led by an inductive process to believe the process is unreliable, the justifiedness of the belief is undermined because the belief conflicts with the reliability requirement for belief permissibility. It is a belief that a certain condition is not satisfied, but the condition is the global reliability condition,which is necessary for belief permissibility. F o r Goldman, then, it is impossible for a reliable inductive process to produce in us a justified belief in its unreliability. But, as we have seen, relative to his account it is also impossible for an unreliable inductive process to produce in us a justified belief in its unreliability. According to his theory, an inductive process simply cannot provide us with a justified belief that it is unreliable, nor, consequently, can it yield knowledge to this effect. But then if, as G o l d m a n believes, the p r o b l e m of induction is at least in part an empirical problem, 22 it is unclear to what e x t e n t it can be solved by induction. In G o l d m a n ' s analysis, one can use induction to discover that induction is reliable, but one cannot use induction to discover that it is not. The reliability of induction is not disconfirmable by inductive means. But if this is the case, then even if we set aside the concerns about circular justification discussed earlier, it is unclear whether it is confirmable by such means either.

III. CARTESIAN SKEPTICISM REVISITED The considerations adduced at the end of the preceding section can be generalized in such a way as to call into question whether we could discover that all our cognitive processes are unreliable. Relative to Goldman's account, we can be justified in believing, and can even know, that a process is unreliable, but only if our belief is produced by a reliable process. If we know a process is unreliable, our belief that it is must be produced by some other process. But now consider the belief that all our cognitive processes are unreliable. Given G o l d m a n ' s analysis, whether or not this belief is true, it can never be justified for us. 106

SKEPTICISM AND NATURALIZED EPISTEMOLOGY If it is true, the processes that produced it are unreliable, and if it is false, whatever permittedness it might seem to have at first is undermined. In either case, Goldman's theory implies that the belief cannot be justified. If we are so constructed that our cognitive processes all tend to fall short of the truth, according to his view, we can never be justified in believing we are. However we are constructed, if we believe we are justified in believing this, in Goldman's eyes we are mistaken. 23 There seem, then, to be limits to the extent to which his account enables us to answer the question of whether we are justified in what we believe empirically. In his view, it is possible that all our cognitive processes are unreliable. If they are, we are incapable of knowing anything, including this. But for Goldman, whether we are capable of knowing anything is a contingent matter. He is committed, then, to the view that there are some empirical hypotheses about us, some contingent propositions about us, that we cannot discover to be false. The skeptic may insist that just because we can never know it to be false that we are capable of knowledge it does not follow that it must be true. And presumably Goldman would concur. But, given his reliableprocess analysis of knowledge, at least as he interprets it, we c a n know that we can know things. That is, for Goldman it is at least logically possible for us to know that we are capable of knowledge. His account denies the skeptic at two junctures. The global reliability analysis of justifiedness allows cognizers to be justified even when Cartesian skeptical counterpossibilities are realized. And, at least as Goldman interprets it, his local reliability requirement rules out these possibilities as irrelevant to the issue of whether we know anything. As we have seen, for Goldman a process is globally reliable just in case it has a tendency to produce a greater number of true beliefs than false ones in normal worlds. In conjunction with the rest of his analysis ofjustifiedness, this conception enables Goldman to maintain that even if one lived in a world in which one were deceived by an evil demon, one's beliefs would still be justified if they were produced by processes that were globally reliable in normal worlds. If we are being deceived by an evil demon, the actual world is not a normal world. But, in Goldman's view, the cognitive processes that produce our beliefs in this world may nevertheless be reliable in worlds that are. So much for the global reliability requirement. Goldman's local reliability requirement embodies a diagnosis of the Gettier and postGettier counterexamples to the traditional analysis of knowledge in terms of justified true belief. Consider, for instance, Harman's example of someone who is looking into a mirror that stands between him and a candle in front of him and sees the reflection of another candle placed to 107

DOUGLAS G. WINBLAD one side. 24 Because he seems to see a candle in front of him, and there really is one there, his belief that there is one there is not only justified but true; nevertheless, it does not constitute knowledge. By adding the local reliability requirement to the traditional three conditions for knowledge, Goldman believes he can disarm such counterexamples. Recall that S's belief that p satisfies the local reliability condition just in case there are no counterfactually specifiable relevant alternative situations in which p would be false but the process or processes by means of which S acquired the belief that p in the actual situation would cause S to believe p anyway. Harman's case can be treated quite easily by means of this analysis, for one can think of a relevant alternative situation of the sort in question by simply imagining the situation as described above without the candle in front of the subject. The diagnosis of such counterexamples implicit in Goldman's analysis is that in each of them there are relevant alternatives of the sort we are considering. The core intuition behind the local reliability condition is that one does not know that p if one cannot discriminate or distinguish between the truth of p and alternative situations in which p is false. However, as we have seen, Goldman does not require that the cognizer be able to distinguish between the truth of p and all alternative situations in which p is false in order to know that p. Such a requirement would be very difficult, if not impossible to satisfy. It is not easy to imagine cognitive processes that could not be "fooled" by any means whatsoever. As a result, Goldman requires only that the processes that lead one to believe something not lead one to believe it in the relevant alternative situations. The question then arises as to how, in particular cases, we are to decide which alternatives are relevant. What about the cases in which I am dreaming, or being deceived by an evil demon? In Epistemology and Cognition, Goldman writes: In particular, I assume that I can know there is a computer keyboard in front of me because I do not have to discriminate that state of affairs from the possibility that I am merely a brain in a vat being artificially stimulated to make it appear as if there is a keyboard in front of me. The brain-in-a-vat alternative just is not a relevant alternative. (I do not, however, have a detailed theory of relevance.) 25 Why is this alternative not relevant? What can we tell the skeptic to dissuade him from believing that it is relevant? We have seen that the Quinean response to this skeptical worry - - roughly, that in the absence of a brain-in-a-vat theory with greater predictive power than current physical theory, we need not take the"envatment" hypothesis seriously 108

SKEPTICISM AND NATURALIZED EPISTEMOLOGY

- - is unlikely to move the skeptic. And it is precisely this feature of Quine's position that lends support to Goldman's claim that an adequate treatment of skepticism requires the resources of conceptual analysis. In the absence of some rationale that justifies ruling out the skeptical counterpossibilities as irrelevant, however, the skeptic can have his way. The mere claim that they are irrelevant will seem as arbitrary to him as the restriction on circular justifications discussed in the previous section seems to Goldman. Even Quine's response will appear to be more strongly motivated, rooted as it is in his commitment to scientific method. None of these observations are meant to suggest that the skeptical counterpossibilities are relevant, or to deny that any account of relevance is to be hadfl 6 In this connection, it is worth noting that we do not ordinarily take such possibilities seriously. Were a defense attorney to cross-examine a witness by asking him how he knows he was not dreaming the night of the crime instead of witnessing a murder, the prosecuting attorney would probably be so surprised he would barely be able to bring himself to object that the question is irrelevant. Our ordinary practice of assessing claims leads us to consider unusual counterpossibilities only when there are, in Austin's phrase, special reasons to think the possibilities are actualities. 27 Interestingly enough, this requirement is structurally similar to Quine's insistence that there be evidence for skeptical hypotheses before we consider them seriously. But the question for the skeptic is whether either ordinary or scientific practice can provide us with a final court of appeal in these cases. In "Discrimination and Perceptual Knowledge," Goldman discusses two sorts of accounts of what constitutes a relevant alternative. According to the first, "the semantic content of 'know' contains (implicit) rules that map any putative knower's circumstances into a set of relevant alternatives." The second has it that "the verb 'know' is not so semantically determinate. ''28 According to the latter sort of view, while there may regularities governing the alternative hypotheses an attributer or denier of knowledge deems relevant, they are not part of the semantic content of'know'. Goldman is attracted to a variant of this kind of position, but says the following: I leave open the question of whether there is a "correct" set of relevant alternatives, and if so, what it is. To this extent, I also leave open the question of whether skeptics or their opponents are "right." ...I shall assume some (psychological) regularities concerning the selection of alternatives. Among these regularities is the fact that speakers do not ordinarily think of "radical" 109

DOUGLAS G. WINBLAD alternatives, but are caused to think of such alternatives, and take them seriously, if the putative knower's circumstances call attention to them. Since I assume that radical or unusual alternatives are not ordinarily entertained or taken seriously, I may appear to side with the opponents of skepticism. M y official analysis, however, is neutral on the issue o f skepticism. 29 If ' k n o w ' is not semantically determinate in the way described above, G o l d m a n m a y neither need nor be able to include an account of relevance in a conceptual analysis of knowledge. We have seen that he doubts whether necessary and sufficient conditions are to be had for many terms. ' K n o w ' m a y well be a case in point. W h a t of Goldman's assumption that speakers are caused to think of skeptical counterpossibilities and take them seriously if the cognizer's circumstances call attention to them? The case we examined earlier of the person who believes there is a candle in front him involves a situation that may in some sense "call attention" to certain alternatives he cannot discriminate from the truth of what he believes, alternatives that seem relevant to whether he knows what he believes to be true. And these alternatives are surely more relevant to the epistemic status of his belief in this situation than they are to that of any number of other beliefs he might have in this or other circumstances. Moreover, it is important to note that most of these other beliefs and circumstances do not call attention to the same alternatives as those to which the case we are considering does. Whether there is a candle in front of someone is hardly relevant to whether he knows that 2 + 2 = 4 because his teacher said so, for example; nor does the case call attention to it. But what situations can be said to call attention to the skeptical alternatives? There is nothing a b o u t someone's sitting by the fire who believes he is doing so that calls attention to the possibility that he is dreaming any more than do countless other situations in which he might find himself. In fact, it is difficult to see what in such a situation would call attention to this possibility, unless we count someone's cognitive states as part of his situation, and, like Descartes, our cognizer is considering it himself. These reflections hardly amount to a theory of relevance. Nor do they presume that Goldman's assumption about when speakers are led to consider skeptical counterpossibilities is meant to count as one itself. G o l d m a n has insisted repeatedly that he has no such theory to offer. Nevertheless, he has changed his position somewhat since writing "Discrimination and Perceptual Knowledge." His analysis of knowledge is still to some extent neutral with respect to skepticism, since he holds 110

SKEPTICISM AND NATURALIZED EPISTEMOLOGY that it is an empirical issue whether we c o m m a n d cognitive processes that can give us knowledge. But he no longer allows the skeptic to question whether we do so on grounds such as that we cannot discriminate between reading an essay and being deceived by an evil demon into believing that we reading it. Because the skeptic cannot see why such grounds are inadmissible, he and G o l d m a n have reached an impasse. Let us suppose, however, that there is a "fact of the matter" about what alternatives are relevant. What becomes of the issue of skepticism then? Relevant alternatives are relative to a cognizer and an epistemic situation. Let us suppose, then, that there is a correct set R of relevant alternatives in the case of S's belief that p. According to Goldman's account, S knows that p only if S can discriminate the truth of p from every member of R. And we possess second order knowledge of S's knowing that p only if we can distinguish S's being able to make such discriminations from relevant alternatives in which S cannot make them. Presumably we can distinguish between these situations only if we ourselves can discriminate the truth of p from the members of R. Next let us suppose that S knows that p. Now consider the following objection. Suppose there is a counterfactual situation in which there is a relevant alternative regarding S's belief that p, not contained in R, that n e i t h e r w e n o r S can distinguish from p's truth. Adopting the second account of relevant alternatives discussed earlier for the moment, presumably such a situation could arise if the empirical regularities that at least in part determine the membership of R did not obtain. Such a counterfactual situation would surely constitute a relevant alternative f o r us to S's knowing that p. But if such a relevant alternative regarding our belief that S knows that p exists, given Goldman's analysis we will be unable to know that S knows that p. F o r we will be unable to discriminate between the truth of the proposition that S knows that p and a relevant alternative in which this proposition is false. The putative relevant alternative might be a skeptical one - - one in which, say, we and S are dreaming that p is true, but nevertheless p is false - - but it need not be. Of course, if the members of R are fixed by necessity, as they would be relative to the account according to which ' k n o w ' is semantically determinate - - in which case perhaps the skeptical counterpossibilities simply c a n n o t be relevant alternatives the objection fails because there could be no such counterfactually relevant alternative given our initial assumptions. In any case, in the absence of a more detailed theory of relevance, it is unclear how to deal with this difficulty from the standpoint of Goldman's theory of knowledge, the 111

DOUGLAS G. WINBLAD m o s t s o p h i s t i c a t e d s o far. N o r is it c l e a r t h a t n a t u r a l s c i e n c e c a n r e s o l v e it, if it c a n b e r e s o l v e d at all. VASSAR COLLEGE POUGHKEEPSIE, NEW YORK 12601 USA

NOTES W.V. Quine, "The Nature of Natural Knowledge," in S. Guttenplan, ed., Mind and Language (Oxford: Clarendon Press, 1975), p. 68. 2 W.V. Quine, "Reply to Stroud," Midwest Studies in Philosophy, vol. VI, 1981, p. 475. 3 Alvin I. Goldman, Epistemology and Cognition (Cambridge, Mass.: Harvard University Press, 1986), p. 57.

a lbid.,p. 66. 5 Goldman may see more than a mere similarity here, insofar as he seems inclined to think of linguistic analysis as applied psycholinguistics. See Epistemology and Cognition, p. 57. 6 Epistemology and Cognition, p. 63. 7 Ibid., p. 106. Very roughly, for Goldman basic processes are ones that are unacquired. In his view, the criterion schema is a schema because it does not specify a particular truth ratio threshold. 8 One might also try to specify rules that serve as instructions for agents, but for Goldman this is a task for regulative theory of knowledge, which is not his concern in Epistemology and Cognition. See Alvin I. Goldman, "Epistemics: The Regulative Theory of Cognition," reprinted in Hilary Kornblith, ed., Naturalizing Epistemology (Cambridge, Mass.: MIT Press, 1985), p. 218. l0 W.V. Quine, "Epistemology Naturalized," in Ontological Relativity and Other Essays (New York: Columbia University Press, 1969), pp. 75-76. ~ "Epistemology Naturalized", p. 72. ~2 "The Nature of Natural Knowledge," p. 70. 13 Epistemology and Cognition, pp. 393-394. 14 Ibid., p. 393. 15 Ibid., p. 394. ~6 Although he might have to recognize new ways belief permissibility can be undermined, invoking the second clause of his framework principle may enable Goldman to honor the intuition that a skeptic with regard to induction is not permitted to acquire a belief in the legitimacy of induction by means of an inductive process. If so, a restriction on circular justification might be unnecessary. But, as long as it is construed as a skeptical problem, the problem of induction would remain unsolved. ~7 L. Carroll, "What the Tortoise Said to Achilles," Mind, 3 (1895): 278-280. 112

SKEPTICISM AND NATURALIZED EPISTEMOLOGY 18 See also Epistemology and Cognition, p. 285. 2o Ibid., pp. 285-286. 21 See ibid., pp. 62-63 and p. 111. 22 See ibid., p. 108: "...it is still a contingent and empirical question whether the processes we take to be justification-conferring really are so." 23 Cf. Epistemology and Cognition, p. 120. 24 Gilbert H a r m a n , Thought (Princeton: Princeton University Press, 1973), p. 22. 25 Epistemology and Cognition, p. 55. 26 Cf. ibid., p. 56. 27 J.L. Austin,"Other Minds,"in Philosophical Papers (Oxford: Oxford University Press, 1961). 28 Alvin I. Goldman, "Discrimination and Perceptual Knowledge," The Journalof Philosophy, 73 (1976), p. 776. 29 Ibid., pp. 775-776.

I13