Against Naturalized Epistemology



My aim in this essay is to examine and criticize the idea of naturalized epistemology, understood as embodying the idea that traditional episte­mology is fatally flawed and hence should be replaced by a properly naturalistic successor discipline.1 My thesis is that while the idea of naturalized epistemol­ogy reflects some genuine, albeit modest insights that need to be recognized and preserved, there is no interesting sense in which epistemology either can or needs to be naturalized. A thoroughgoing naturalization of epistemology strikes me as wrongheaded and ultimately self-destructive, and I will try to show here why this is so.


Unfortunately, however, like many fashionable positions, naturalized epis­temology is a rather diffuse and uncertain target, and considering all of the things that have been understood by this phrase would obviously go well beyond the allowable scope of the present essay. Thus I will focus here primarily on the accounts of naturalized epistemology offered by its original proponent, namely W. V. O. Quine, and by the philosopher who has offered the most complete and systematic recent account, namely Philip Kitcher.2 After an initial examination, in section I, of Quine's original introduction of and rationale for the idea that epistemology should be naturalized, I will turn, in sections II and III, to an examination of the two main ingredients of the idea of naturalized epistemology, as identified by Kitcher, together with the arguments in support of them that he discusses: (i) the rejection of "apsychologistic" epistemology; and (ii) the rejection of the idea of a priori justification, especially but not only as it applies to the theses of epistemology itself (58). I will argue that while there are insights of value in the area of (i), their significance is much more limited than Kitcher and others have taken it to be. Ingredient (ii), on the other hand, seems to me to be almost entirely a mistake. I will then conclude the essay by trying to show, in section IV, why the idea of naturalized epistemology, especially as it involves the rejection of a priori justification, is not only inadequately supported but inherently self-destructive.


Two other preliminary issues require brief consideration. One of these is the nature of the epistemological project itself. Here I am forced to be at least mildly dogmatic. I will assume here that at least one central aim of traditional epistemology is: (a) to decide whether or not we have good reasons for thinking that our various beliefs about the world are true; and (b), if the answer to this first question is affirmative, to say what those reasons are and to explain why they are good ones. I take this to be the central issue, the common thread that connects the concerns of Descartes, Locke, Hume, Reid, and others with twentieth-century epistemology. But it is enough for my purposes if it is one central issue, and it is on this issue, the issue of critical epistemology, that my discussion will focus.3


Second, while a good part of my discussion will take Kitcher's account of naturalized epistemology as its main target, it is important to acknowledge at the outset that my conception of the main dialectical alternatives is importantly different from his, in a way that means that I am approaching his arguments somewhat obliquely, rather than head on. For Kitcher, the main alternative to naturalized epistemology on the more conservative side4 is the analytic approach to epistemology, reflecting the so-called "linguistic turn," that is characteristic of the logical positivists and their contemporary progeny. His thesis, very roughly, is that this view is an ill-grounded aberration ("an odd blip in the history of philosophy" [56]) and that naturalized epistemology represents a return to the "restrained" version of naturalism exemplified by "Descartes, Locke, Leibniz, Hume, Kant, and Mill" (54). My own view, in contrast, is: first, that while the great early moderns certainly did not feel any need to fastidiously eschew the use of psychological concepts, there is no very interesting sense in which their philosophical claims depended on anything like empirical results from psychology and the other sciences, and thus no very interesting sense in which they were naturalists;5 and, second, that while the explicit methodological theses of positivism and analytic epistemology were indeed an ill-grounded aberration (one for which I suspect that I have substantially less regard than does Kitcher), these were largely superficial window dressing, beneath which there was very substantial continuity with a rationalistic approach to epistemology in particular and philosophy in general that extends back through the early moderns to much earlier times. (The defining thesis of rationalism, as I intend it here, is that a priori justification and knowledge genuinely exist and are not confined to claims that are in any useful sense merely conceptual or linguistic or "analytic" in character.6) An adequate consideration of these historical issues would far exceed the allowable bounds of the present essay. But it is important to be clear that when I oppose naturalized epistemology in favor of traditional epistemology, it is the older and deeper tradition of rationalistic epistemology that I mean to be defending. (As this suggests, it is the issue pertaining to a priori justification and knowledge, rather than that pertaining to psychologism, that is in my judgment by far the more fundamental.)




It was Quine who first introduced the idea that epistemology should be nat­uralized, and it will be useful to begin by examining his case for this claim, as presented in his essay "Epistemology Naturalized."7 Quine's introduction of this idea can, I believe, be plausibly viewed as an attempt to defend and make sense of his own epistemological position. Though elaborated somewhat in later works, Quine's main epistemological view seems to be essentially that suggested by the final section of "Two Dogmas of Empiricism": what we believe is a huge, interconnected "web" or "fabric" of sentences, a web that "impinges on experience only along the edges."8 We revise this web or fabric more or less continuously in an effort to keep the edge "squared with experience," and the results of these revisions represent our best epistemic efforts at any particular moment.


Many questions can be raised about this familiar but still undeniably fuzzy picture. But the central issue is what bearing it has or is supposed to have on the epistemic justification of our beliefs, i.e., with whether we have any reason to think that they are true. Quine's predominant view seems to be that the web picture is merely a psychological description: we simply do treat some sentences as more relevant or "germane" to a particular experience than others and some as generally less open to revision than others, and we do revise or modify our system of sentences accordingly. But the problem at this point is obvious: merely from the psychological fact that we do operate in this way, it does not follow in any obvious way that the beliefs that result are epistemically justified or rational, ones that we have any good reason for thinking to be true, so that, if true, they would constitute knowledge. Why, it may well be asked, should Quine be taken to have even offered an epistemology?


In response to this pretty obvious question, Quine offers a radical reinterpretation of what epistemology itself is all about. In "Epistemology Naturalized," he argues that epistemology, "or something like it," should be reconstructed as "a chapter of psychology," an empirical study of the rela­tion between "a certain experimentally controlled input--certain patterns of irradiation in assorted frequencies, for instance" and an output consisting of "a description of the three-dimensional external world and its history" (EN 83-84). His claim, in first approximation, is that while such a naturalized epistemology admittedly falls short of achieving the goals of traditional epistemology, it goes as far in that direction as turns out to be possible, and far enough to constitute a reasonable albeit less ambitious substitute. The rationale offered by Quine for such a reconstrual is basically that the epistemological project, as traditionally conceived, has failed more or less irredeemably and hence must be replaced by a more viable substitute. This view depends in part on a fairly narrow conception of traditional epistemology, roughly that put forward by positivistic empiricism, according to which epistemology (or at least the part of epistemology that is concerned with "natural knowledge," i.e., knowledge of the physical world) involves two correlative goals: (i) to explain the relevant concepts, e.g., the concept of a physical body, in sensory terms ("the conceptual side of epistemology"); and (ii) on the basis of this explanation, to justify claims about the physical world on the basis of sense experience ("the doctrinal side of epistemology") (EN 71). It is obviously these goals that motivate phenomenalism, as advocated by Hume, Ayer, Lewis, and many others. But, argues Quine, it is clear by now that neither of these goals, at least as traditionally conceived, can be achieved. The attempt to reduce physicalistic concepts to phenomenal ones fails to yield genuine translations; and, since sensory generalizations at least would be required, the attempt to prove physical statements on the basis of sensory evidence is defeated in any case by the problem of induction. What is left, once these goals are abandoned as hopeless, is the attempt "simply to understand the link between observation and science," and there is no reason not to appeal to psychology in achieving this end. In particular, the worry that an epistemological appeal to the results of natural science would be circular no longer applies once we abandon the goal of justification (EN 75-76).


There are many problems with this line of argument. A relatively minor one is that Quine's picture of "the conceptual side" and "the doctrinal side" of traditional epistemology as more or less equally important vastly exaggerates the importance of the former. Construed in the reductive way in which Quine construes it, "the conceptual side" of epistemology is a feature only of the narrowest and most implausible versions of empiricism, and even there is motivated primarily by the attempt to satisfy "the doctrinal side." Thus the failure to achieve the aim of "the conceptual side," to which Quine devotes most of his attention in "Epistemology Naturalized," does very little to show that traditional epistemology has failed and hence needs to be replaced by the suggested Quinean surrogate.


More importantly, Quine's discussion seriously muddies the waters by failing to distinguish a stronger and a weaker conception of "the doctrinal side" of traditional epistemology. According to the stronger conception, deriving from Descartes, the goal is to achieve certainty in our beliefs about the world, to establish that they are infallibly and indubitably true. For the weaker conception, on the other hand, the goal is the more modest one of showing that there are good reasons for thinking that our beliefs are at least likely to be true; complete certainty, while of course still desirable, is not essential. Though his discussion of "the doctrinal side" is too sketchy to allow much confidence on this point, Quine seems to slide illegitimately from the relatively uncontroversial claim that the stronger, Cartesian goal cannot be attained for "natural knowledge" to the much less obvious claim that the more modest goal is not achievable either. Thus we are told that statements about bodies cannot be "proved" from observation sentences, that "the Cartesian quest for certainty" is a "lost cause," that claims about the external world cannot be "strictly derived" "from sensory evidence" (EN 74-75); and on this basis it is apparently concluded that the entire "doctrinal side" of traditional epistemology, which Quine characterizes in one place as concerned with "the justification of our knowledge of truths about nature" (EN 71), must be abandoned. But this, of course, simply does not follow.9


What might cast doubt on this reading of Quine's argument is his employment of the term 'evidence' to characterize even the project of naturalistic epistemology. Thus he claims that despite the failure of traditional episte­mology, it remains unassailable "that whatever evidence there is for science is sensory evidence" (EN 75). And further on we are told that the goal of naturalistic epistemology is "to see how evidence relates to theory, and in what ways one's theory of nature transcends any available evidence" (EN 83); and also that "observation sentences are the repository of evidence of scientific hypotheses" (EN 88). I do not see any way, however, to take these remarks at face value, for surely the standard normative concept of evidence, i.e., the concept of a reason, perhaps of a certain restricted sort, for thinking that some claim is true, is not a concept of empirical psychology. Psychology can describe the causal relations between sensory stimulations and beliefs of various sorts, but it cannot offer any assessment of the rational acceptability of any such transition. Perhaps there is some other, naturalistically acceptable conception of evidence that Quine has in mind, but if so this would not count against the conclusion that Quine has entirely abandoned "the doctrinal side" of traditional epistemology.


It thus seems clear that at least Quine's version of naturalized episte­mology has nothing whatsoever to say about whether we have any reason to think that our beliefs about the world are true. And hence, if Quine is right that this sort of naturalized epistemology is the best we can do, the result is a thoroughgoing version of skepticism: we have a set of beliefs, i.e., we accept a set of sentences, that describe the external world;10 part of that set of beliefs describes how the beliefs are caused by observation, i.e., by sensory stimulation; but we have no cogent reason of any sort for thinking that any of these beliefs are true. And if knowledge necessarily involves the possession of such reasons, as most philosophers would still insist, then we also have no knowledge.11 This may indeed, as Quine suggests at one place be "the human predicament" (EN 72). But it is surely extremely unsatisfactory and also intuitively implausible from both a theoretical and a practical standpoint.12


To see how Quine would respond to this sort of objection, we need to look at his conception of skepticism:


Scepticism is an offshoot of science. The basis for scepticism is the awareness of illusion, the discovery that we must not always believe our eyes. Scepticism battens on mirages, on seemingly bent sticks in water, on rainbows, after-images, double images, dreams. But in what sense are these illusions? In the sense that they seem to be material objects which they in fact are not. Illusions are illusions only relative to a prior acceptance of genuine bodies with which to contrast them . . . The positing of bodies is already rudimentary physical science; and it is only after that stage that the sceptic's invidious distinctions can make sense,13


Thus skepticism. in Quine's view, arises only from within science; "sceptical doubts are scientific doubts," and hence can best be answered by science itself:


Retaining our present beliefs about nature, we can still ask how we can have arrived at them. Science tells us that our only source of information about the external world is through the impact of light rays and molecules upon our sensory surfaces. Stimulated in these ways, we somehow evolve an elaborate and useful science. How do we do this, and why does the resulting science work so well? These are . . .  scientific questions about a species of primates, and they are open to investigation in natural science, the very science whose acquisition is being investigated.14


Thus, Quine claims, naturalized epistemology is in principle quite adequate to deal with skepticism.


But this view of the skeptical challenge is seriously inadequate in two distinct ways. In the first place, while it is of course true that skeptics have often appealed to various sorts of illusions to motivate their doubts, such an appeal is in no way essential to the basic thrust of skepticism. The fundamental skeptical move is to challenge the adequacy of our reasons for accepting our beliefs, and such a challenge can be mounted without any appeal to illusion. A prominent example of such a challenge is Hume's skepticism about induction, mentioned in passing by Quine himself (EN 71-72), but there are many, many others. Such a challenge can in principle be raised against any alleged piece of knowledge: is the reason or justification that is available for the belief in question adequate to show that it is (at least) likely to be true? To the difficult issues raised by these other versions of skepticism, Quine's version of naturalized epistemology has nothing at all to say. This is a very serious deficiency if one takes the project of traditional epistemology at all seriously, and the point is that Quine has offered no reason for not taking it seriously.


Moreover, even if we restrict our attention to the more limited versions of skepticism that essentially involve an appeal to illusions, the sort of response that is offered by naturalized epistemology totally misses the main issue which is, of course, reasons or justification. What the skeptic questions is whether, once the possibility of illusion is appreciated, our sensory experience any longer constitutes a good reason for accepting our various beliefs about the world. Such a skeptic need not doubt that our beliefs are caused in some way, nor still less that an account of how they are caused can be given from within our body of beliefs about how the world operates. What he doubts is whether we have any reason for thinking that any of our beliefs about the world, including those that are involved in such an account, are true, and to this issue of justification, the Quinean version of naturalized epistemology once again has nothing at all to say.


Another, quite different way to appreciate the irrelevance of this con­ception of naturalized epistemology to traditional epistemological issues is to consider its application to bodies of belief where a substantial degree of skepticism seems warranted, e.g., to religious beliefs and beliefs in occult phenomena of various sorts. For just as naturalized epistemology can say nothing positive about the justification of science and common sense, and is thus impotent in the face of skepticism, so also it can say nothing distinctively negative about the justification of these less reputable sorts of belief. There is. after all, no reason to doubt that occult beliefs are caused in some way by the total sensory experience of the individual, and thus no reason to doubt that psychology can offer an empirical account of how they are produced.15 Such an account would no doubt differ in major ways from that which would be given for more properly scientific beliefs, but the differences would not, within psychology, have any justificatory significance. Thus the only epistemology that is possible on Quine's view apparently cannot distinguish between science and occult belief in any way that would constitute a reason for preferring the former to the latter.


I conclude that at least Quine's original version of naturalized episte­mology is both inadequately defended and extremely unsatisfactory in light of its radically skeptical implications. It should not be assumed, however, that all those who have followed Quine's lead (or at least adopted his slogan) are guilty of the same mistakes. (Kitcher, in particular, makes abundantly clear that he wants to preserve the normative dimension of epistemology.) Thus I want to examine, in the next two sections, the two main ingredients of naturalized epistemology. as identified by Kitcher, together with the arguments that allegedly support them.16 Following that, I will return in the final section to the question of whether any version of naturalized epistemology involving these components can avoid the skeptical consequences that we have seen to follow from Quine's version.




I turn then to the first of Kitcher's two main ingredients, the rejection of "apsychologistic" epistemology. Here there are a number of issues that need to be carefully sorted out. My main claim will be that while there are a number of ways, all of them pretty obvious, in which psychological concepts and psychological theorizing are relevant to epistemological issues, none of them are even approximately central enough for their recognition to constitute an interesting psychologizing or naturalizing of epistemology.


We may begin with what is probably the most widely discussed recent argument for some degree of psychologization. In Kitcher's presentation. it runs as follows:17


By the mid 1970s a powerful argument for psychologistic episte­mology had emerged. Take any set of favored logical relations among propositions that a subject believes. It is nonetheless possible that the subject lacks knowledge and lacks justification because the psychological connections among her states of belief have nothing to do with the logical relations. Thus, to take an extreme example, assume that a subject justifiably believes that p, justifiably believes that p-->q, and believes that q. It might seem that the belief that q must be justified because there is an elementary logical inference to q from propositions that are believed. Nonetheless, it is easy to understand that the causes of the subject's belief may have nothing to do with this elementary inference, that she fails to make the inference and believes that q because of some thoroughly disreputable generative process. ...(60)


If a powerful argument is one whose conclusion is hard to resist, then al­most anyone will agree that this argument is indeed powerful. Unfortunately, however, as is indeed the case with many powerful arguments, its importance might well be questioned on the ground that its conclusion is something that it is hard to believe anyone has ever disputed. Has there ever been a significant epistemologist, or indeed philosopher of any kind, who seriously (as opposed to inadvertently18) advocated the view that the mere presence of a logical relation among a person's beliefs might yield justification, even if it went entirely unnoticed by the individual in question? I at least can think of no remotely plausible candidate for such a dubious honor. Thus, I suggest, the degree of psychologism that follows from this argument, which we might for obvious reasons call minimal psychologism, is entirely uncontroversial. But also, as Kitcher himself seems to acknowledge (62), minimal psychologism involves at most a quite minor departure from traditional epistemology, if indeed it involves any departure at all.


There is a second kind or level of psychologism that is equally undeniable, but also, I believe, equally innocuous from the standpoint of traditional epis­temology. As Kitcher points out, various philosophers have made the logical or conceptual point that perception depends on causal relations, e.g., "that one can see that p only if there is some lawful dependency between one's belief that p and p" (61).19 Analogous points could obviously be made about introspection and memory and, in a somewhat different way, about logical inference itself. It follows that the philosophical consideration of such concepts and, especially, the application of the philosophical results to actual cases will have to make reference to psychological facts about, e.g., the causation of belief. But this does nothing to show that the distinctively philosophical content of such discussion depends on empirical psychological results in any interesting way. Thus, as Kitcher seems again to acknowledge (62), this second kind of psychologism, which we may term conceptual psychologism, again represents no significant advance toward the naturalist's main conclusion in this area, viz. that epistemology must appeal in important and thoroughgoing ways to empirical psychological results.


There is yet a third kind of psychologism that must be acknowledged, one which, while in a way more substantive than those discussed so far, still poses no real threat to traditional epistemology, as understood here. Consider again Kiteher's idea of the meliorative epistemological project, the project of improv­ing the reliability and success of human cognitive functioning. It is once more obvious and something that it is hard to imagine anyone denying that serious attempts in this direction must at some point take notice of the human cognitive behavior in question and especially of human cognitive limitations. Thus, for example, it does no good for this purpose to describe a complicated schema for, e.g., inductive or explanatory inference, however logically impeccable it may be in itself, if it is one that human beings are for some psychological reason incapable of conforming to or a least reasonably approximating. And, to take the other side of the coin, it is presumably an important part of this general meliorative effort to provide critical assessments of inferential patterns and other modes of cognitive behavior that are actually exemplified in practice, for which purpose some knowledge of the relevant facts about such practice, psychological and otherwise, is clearly needed. All of these points, however, have to do again with applying epistemological assessments to actual practice, not with how those assessments are themselves arrived at and justified. And thus there seems to be nothing about what we may call meliorative psychologism, understood in the way just indicated, that has any serious bearing on the nature of epistemological criticism and argument when considered in itself--and thus once again, nothing that supports any significant kind of naturalization.


The question is whether, once minimal, conceptual, and meliorative psy­chologism are set aside as essentially irrelevant to the main issue between tra­ditional epistemology and the proposed naturalistic successor discipline, there is any argument left in this area in favor of the latter, any reason for thinking that epistemology needs to be psychologized in some deeper, more fundamental way. Kitcher's response to this question consists mainly of a critique, primarily within the context of the meliorative project, of an a priori epistemology that construes its claims as "analytic" or "conceptual" truths pertaining to our concepts of justification or rationality. To this sort of view, exemplified perhaps most clearly in the ordinary language solution to the problem of induction advanced by Strawson and others,20 "an appropriate challenge is always, 'But why should we care about these concepts of justification and rationality?'" (63) Or, more specifically, why should we think that seeking beliefs that are justified or rational according to these concepts is likely to lead us to the truth?


But while it is easy to agree that a priori claims, when construed in this moderate way, appear to have little significance either for the meliorative project or for the central epistemological project as explained above, Kitcher seems to me to conclude much too quickly that the only alternative to an appeal to merely analytic or conceptual truths is the sort of appeal to empirical scien­tific and especially psychological results that is characteristic of naturalism. In addition, he seems to me not to worry enough about whether and how such an empirical epistemology could stand on its own, without being grounded on at least some a priori principles. Both of these issues can be restated in terms of the traditional rationalist conception of a priori justification (and knowledge) alluded to above: Is there any argument that the naturalist can give for excluding the very possibility of such justification? And is there any way that either the empirical results that the naturalist depends on or the naturalist epistemological theses themselves can be justified without such an a priori appeal? These questions will be the concern of the final two sections of this essay.




The foregoing discussion suggests one clear way (though not, I believe, the only way) in which the issue of a priori justification is apparently prior to and more fundamental than that of psychologism: when relatively innocuous versions of psychologism have been set aside, the argument for anything stronger in this di­rection depends on rejecting the idea that significant epistemological principles or premises can be established on an a priori basis, since this would represent an obvious alternative to naturalism. It is time to ask whether the proponents of naturalism can offer any compelling arguments for such a rejection.


Unfortunately, however, as already briefly suggested above, the arguments of the naturalists are mainly aimed, not at the central conception of a priori justification itself, but instead at the rather idiosyncratic form that the idea of a priori justification took within the context of analytic epistemology. The specific target is the view that I have referred to elsewhere as moderate empiricism: the thesis that the only claims or propositions that can be justified a priori are those which are analytic (in some sense of this exceedingly flexible term).21 One consequence of this focus is that the resulting case for naturalized epistemology is seriously incomplete, in that a significant alternative is not adequately considered. But a more subtle problem, as I will explain, is that the kinds of arguments that Quine and others have mounted against moderate empiricism, if successful, have the consequence that there was never any good ease for a moderate empiricist (as opposed to a rationalist) conception of a priori justification in the first place. The naturalists are thus in the dialectically embarrassing position of concentrating almost entirely on a view that, if their own arguments are any good, is not the main alternative.


As I have suggested elsewhere,22 and as Kitcher's account of Quine also amply corroborates, Quine's arguments against the idea of a priori justification tend to assume what amounts to a hypothetical version of moderate empiricism: the view that if there were any a priori justified claims, they would have to be analytic. It is this implicit premise that makes his arguments against analyticity into arguments against a priori justification. But why should such a premise be accepted? What Quine does not seem to realize is that the case for moderate empiricism, especially in the face of the numerous counterexamples offered by recalcitrant rationalists, rests almost entirely on the claim that the account of a priori justification that can be given using the concept of analyticity is clear and unmysterious in a way that the general concept of a priori justification is not, so that only the moderate empiricist reduction of apriority to analyticity can make clear sense of how a priori justification is possible. It is, I suggest, this moderate empiricist argument, and not the idea of a priori justification itself, that is defeated by Quine's arguments (assuming that they are otherwise successful). This is most conspicuously true for the central argument of "Two Dogmas of Empiricism,"23 the argument that the very idea of analyticity cannot be defined or explicated in an intelligible way. For obviously, if the very idea of analyticity is unintelligible, there can be no warrant for the claim that the appeal to analyticity provides a clearer and less problematic account of the possibility of a priori justification than that offered by the rationalist.


The upshot is that the naturalist cannot justify the rejection of the idea of a priori justification by arguing against moderate empiricism, for the claim that moderate empiricism is the preferred account of such justification, superior to that of the rationalist, will not survive such arguments, if they otherwise have any force. What naturalism needs, then, is a direct objection to rationalism -- or, more of less equivalently, an argument that the idea of a priori justification is untenable even when not construed in a moderate empiricist way. Once the issue is posed in this way, arguments of the sort required are anything but thick on the ground. In the present essay, I will limit myself to the two possibilities for such an argument mentioned by Kitcher, one from Quine that Kitcher alludes to but does not really discuss in any detail and one that he rather tentatively attributes to Kuhn.24


First. If Quine's arguments against analyticity are set aside as essentially irrelevant to the main issue, only one very clear possibility remains within Quine's own writings for an argument against a priori justification: the argu­ment, at the end of "Two Dogmas of Empiricism," that depends on the thesis that any claim can be (rationally) given up.25 The idea, apparently, is that an a priori justified claim would have to be one which could never be (rationally) given up, so that if no claim has this status, then nothing is justified a priori.


There are many problems with this strangely influential line of argument.26 But the central one pertains to the way in which Quine defends the thesis that any claim can be (rationally) given up. He appeals to an extended (or perhaps exaggerated) version of the celebrated thesis of Duhem concerning the way in which background assumptions are relevant to the assessment of an observation or experiment that conflicts with and apparently refutes some particular claim or thesis. Duhem's argument is that the claim or thesis in question can always be retained by abandoning instead one of these background assumptions. Quine generalizes this by arguing that in principle any element of the system or "web" of belief could be such that modifying or abandoning it would help to resolve the conflict with experience, so that such modification or abandonment might well be the most rational course. But whether or not Quine's hyperbolically holistic view of our accepted body of beliefs is correct, the present line of argument is in fact clearly and utterly question begging. For the most that Duhem's thesis, in however extreme a form, could show is that it might be rational to give up any claim in the web -- if the only consideration relevant to rationality were how best to accommodate such conflicting experience. But to assume that this is so is obviously tantamount to assuming, rather than showing, that a priori considerations have no independent rational force.


Second. The Kuhnian or a least Kuhn-inspired argument is formulated by Kitcher as follows:


Kuhn's passing remarks about the details of earlier accounts of scientific methodology stress the mismatch between the deliverances of methodology and the reasoning that scientists actually employ. Unless one can show that attention to the historical record will close the gap between philosopher's methodologies and scientific practice ... methodologists are confronted with a dilemma. Either they can continue to insist that philosophers know a priori the principles of confirmation and evidence, concluding that the actual reasoning of scientists is cognitively deficient, or they can abandon the a priori status of methodological claims and use the performances of past and present scientists as a guide to formulating a fallible theory of confirmation and evidence. Since the first option has an uncomfortable air of arrogance, it is hardly surprising that most responses to Kuhn have followed the latter course. (73)


How exactly the latter, preferred alternative is supposed to go is less than clear, at least to me, though some of the discussion of the final section will be at least relevant. But if it is interpreted in the seemingly obvious way as involving a total rejection of any a priori appeal, then there is the more immediate problem that the alternatives posed in the alleged dilemma fail to be exhaustive. For in addition to the alternatives of (i) sticking stubbornly and perhaps dogmatically to one's initial, allegedly a priori precepts or (ii) adopting what may seem to amount to an abandonment of philosophy altogether in favor of a kind of psychology or sociology of scientific practice, there is the seemingly obvious further alternative of (iii) reconsidering one's initial a priori assessment in light of what scientists seem to be doing, while still insisting that any acceptable mode of scientific reasoning must ultimately be able to be seen or shown to be rationally cogent in an a priori way that transcends the mere fact that it is employed in practice.

Here it may be useful to briefly consider an example, and fortunately there is a relatively perspicuous and pivotal one available, though not perhaps one that Kuhn himself would be entirely happy with. Consider what is often referred to as theoretical reasoning: reasoning that moves from observational premises to a conclusion about objects or processes that are not directly observable, the rationale being that such objects or processes provide the best explanation of the observational evidence. Notoriously there were generations of philosophers who denied, allegedly on a priori grounds, that such reasoning could be rationally cogent, or perhaps even intelligible, unless the conclusion were so reinterpreted as to no longer pertain to genuinely observation-transcendent entities, but instead to amount simply to a redescription of the observations themselves. This was of course the rationale for instrumentalism or fictionalism in the philosophy of science, behaviorism in the philosophy of mind, phenomenalism in the philosophy of perception, and other less important but basically similar views. But in spite of the philosophers' admonitions, both scientists and ordinary people continued to make the inferences in question and to construe the conclusions in the supposedly unacceptable way.27 What eventually happened, of course, was not an abandonment of a priori epistemology in favor of psychology or sociology, but rather the realization of a gradually increasing group of philosophers beginning with Peirce, that the initial philosophical view was seriously oversimplified, and that an a priori case (or at least apparent case) could be made for the kind of reasoning in question (what has come to be called "inference to the best explanation"). I do not mean to suggest that such reasoning is now entirely unproblematic or that the issues in question have been fully resolved. The point for now is just that this case illustrates a third alternative to the two mentioned in Kitcher's alleged dilemma.


If this third alternative is indeed available, then Kuhn's historical argu­ment fails to establish the strong thesis that the ideas of a priori justification and a priori epistemology should be abandoned, as opposed to the much weaker thesis that an apparent conflict with scientific practice can and should provoke a reconsideration of the specific a priori results in question, possibly but certainly not inevitably leading to their correction or abandonment. All this assumes, of course, that it makes sense to speak of correcting or abandoning a priori claims, i.e., that a priori justification need not be construed as incorrigible or infallible. My view is that a conception of a priori justification as both corrigible and fallible is quite tenable (and is indeed easily seen to be unavoidable in the face of various kinds of apparent mistakes). But a full discussion of this issue is beyond the scope of the present essay.28


My conclusion is that at least the main naturalist arguments against the possibility of a priori justification are rather spectacularly unsuccessful, so that the case for the second main element of the idea of naturalized epistemology has not been made out. It is possible, of course, that there are other, better arguments that the naturalist might invoke instead, but a consideration of whether this is so will have to await another occasion. For now, I want to turn in the final section to an argument that naturalized epistemology in general and the rejection of a priori justification in particular leads directly to epistemological disaster.




The argument in question depends on a distinction between two classes of beliefs: those which report the results of direct observation or experience; and those whose content transcends the results of direct observation or experience. It might be thought that such a distinction has been shown to be highly problem­atic or even completely untenable by the extended controversy in the philosophy of science concerning the observational/theoretical distinction, but this would, I believe, be a mistake. That discussion indeed shows that it is hard to be sure precisely where the observational/non-observational line is to be drawn, and in particular that many claims that it is initially plausible to count as direct observations may be infected with background theory in a way that renders their status as observations at least uncertain. But none of these considerations provides the slightest reason for doubting that there are many, many beliefs that we confidently hold which cannot be construed as strictly observational or experiential in any sense that has the slightest plausibility: beliefs about the remote past, beliefs about the future, belief about present situations where no observer is present, beliefs about general laws, the vast majority of the beliefs that make up theoretical science, and perhaps others. (Any belief whose status in this respect is seriously uncertain may for present purposes simply be consigned to the observational or experiential side of the ledger.)


I will assume here, without worrying about the details, that the fact that a belief is a report of direct observation or experience constitutes an adequate reason for thinking it to be true. But what about the non-observational or non-experiential beliefs? If we are to have any reason for thinking these latter beliefs to be true, such a reason must apparently either (i) depend on an inference of some sort from some of the directly observational beliefs or (ii) be entirely independent of direct observation. A reason of sort (ii) is plainly a priori. And a reason of sort (i) can only be cogent if its corresponding conditional, a conditional statement having the conjunction of the directly observational premises as antecedent and the proposition that is the content of the non-observational belief as consequent, is something that we in turn have a reason to think to be true. But the reason for thinking that this latter, conditional statement is true can again only be a priori: if, as we may assume, all relevant observations are already included in the antecedent, they can offer no support to the claim that if that antecedent is true, then something further is true. Thus if, as the naturalist claims, there are no a priori reasons for thinking anything to be true (or, as Kitcher sometimes seems to suggest, none of any epistemological importance), the inevitable result is that we have no reason for thinking that any of our beliefs whose content transcends direct observation are true.29


This is epistemological disaster in itself, but a further consequence is that the vast majority of the claims about the nature of the world, the nature and reliability of human psychological processes, etc., upon which naturalized epistemology so lovingly focuses, are things that we have no reason at all for thinking to be true-as, indeed, are the very theses that epistemology must be naturalized or that traditional epistemology is untenable (together with all normative claims of any sort). In this way, naturalized epistemology is self referentially inconsistent: its own epistemological claims exclude the possibility of there being any cogent reason for thinking that those claims are true.30


The foregoing argument seems to me as obvious and compelling as any in the whole of philosophy, so much so that it would be embarrassing to advance it in print were it not so often ignored. The question is whether there is anything that the proponent of naturalized epistemology can offer in reply. Here I will focus on some remarks of Kitcher's, in the essay under discussion here, which while not aimed at precisely the argument formulated above, are at least enough in the same general vicinity to be worth considering. While I do not believe that the response that I will extrapolate from Kitcher is adequate, I know of no one who has done any better.


Kitcher's immediate target is a skeptic who challenges a naturalistic defense of principles of "cognitive optimality" on grounds of circularity, but his response has wider application:

The demand is for synchronic reconstruction of beliefs: take the totality of things you believe, subtract this claim and everything that you cannot defend without assuming it, and now show that the claim is correct. With respect to some claims, synchronic reconstruction is possible.


Traditional naturalists, however, cognizant of the history of mathematics, science, and methodology, should know in advance of skeptical embar­rassments that some forms of the problem of synchronic reconstruction are solvable and others are not. On their account there is no substantial body of a priori knowledge, so that successful synchronic reconstructions must always appeal to empirical findings. . . .On naturalism's own grounds, there are bound to be unanswerable forms of skepticism.


Traditional naturalists should therefore decline blanket invitations to play the game of synchronic reconstruction. . . . (90)


And in his recent book on the philosophy of science, he remarks that "Skeptics who insist that we begin from no assumptions are inviting us to play a mug's game,"31 the suggestion being that of course this sort of skeptical problem cannot be solved because it is unreasonably posed in the first place.


But it does not seem to me that this response will do. While this sort of answer may be appropriate for some skeptical problems, the issue of whether and why we ever have any reason to think that a conclusion that goes beyond observation is true is far too fundamental and inescapable to be dismissed as some clever dialectical trick. Nor does the appeal to the synchronic/diachronic distinction seem to help, since the same issue can of course be raised for our ancestors, however remote, and it is hardly reasonable to say that the fact that something was accepted at some earlier point despite there being no reason to think that it was true somehow gives a reason now to think that it is true. It is quite true, of course, that it is part of the naturalist's own position, or so immediate a consequence as to make no difference, that the skeptical problem posed above cannot be solved, but their explicit adoption of this consequence does nothing to make it less catastrophic or less self-defeating. Thus it is the naturalists who are asking us to play a "mug's game" -- an invitation that, I suggest, should be firmly declined.


Summarizing, I have argued, first, that Quine's original argument for naturalizing epistemology fails either to show that this is necessary or to establish a viable alternative; second, that the various reasons offered by Kitcher and others fail to show the need for psychologizing, and so naturalizing, epistemology in any important sense; third, that the main arguments of the naturalists fail to show that a traditional, rationalist conception of a priori justification is untenable; and, fourth, that the abandonment of any sort of a priori justification leads directly to epistemological disaster and also undercuts the very premises used to argue for it.


I will conclude with two further remarks. First, one thing that it is important to bear in mind about the issue of a priori justification is how easy it is to rely on a priori insights without explicitly acknowledging them, even to oneself. This is particularly easy where such insights pertain to fundamental patterns of reasoning and argument. Thus it becomes fatally easy for a pro­ponent of naturalized epistemology to rely on the intuitively obvious rational credentials of logic, induction, and explanatory reasoning, while at the very same time denying the very possibility of the only sort of nonquestion-begging justification which such reasoning could have. The argument offered at the beginning of this section can serve as a useful antidote to this kind of mistake.


Second, it is important to emphasize that I have not attempted to say anything here about the scope of a priori justification, in particular about how specific and numerous a priori epistemological principles may be. Nothing about the argument advanced here excludes the possibility that such principles might be few in number and very general in character, perhaps even limited to logic (including probability theory) and general principles of inductive and explanatory reasoning.32 Thus it might be that such principles do not take us very far -- and in particular that much of the job of "meliorative epistemology" must be done empirically. This would mean that while the general thesis of naturalism is false, many specific naturalistic results may be correct and valuable. But it would still remain the case that the a priori underpinning is indispensable to the rational acceptability of these more specific results, so that naturalized epistemology, as Kitcher understands it, would be unacceptable.


At one point in his essay, Kitcher speaks of a "broader vision" that underlies naturalism, one part of which is the following:


How could our scientific understanding of ourselves -- or our reflec­tions on the history of the sciences -- support the notion that answers to skepticism and organons of methodology (or, indeed, anything very much) could be generated a priori? (58)


To this, my response is that unless some general answers to skepticism can be justified a priori (even if these may not add up to anything that deserves to be called a methodology), we will have no reason for thinking that "our scientific understanding of ourselves" (or, indeed, anything else beyond the results of direct observation) is true.33




1. I have no wish to deny that some of the things that are or might be done under the rubric of naturalized epistemology might have independent value of their own, so long as they are not intended as a replacement for traditional epistemology. But an exploration and assessment of this possibility is beyond the scope of the present essay.


2. Philip Kitcher, "The Naturalists Return," Philosophical Review 101 (1992): 53-114. Otherwise unspecified references in the text are to the pages of this article.


3. This is not to deny that epistemologists have had other aims as well, in particular that many of them have also sought to advance what Kitcher calls the "meliorative epistemolog­ical project" (64f.), i.e., the project of improving the reliability of human cognitive efforts. Indeed, as we will see further below, the relevance of psychology to this second project is a good deal clearer than it is for critical epistemology. But it seems obvious that no appeal to a further project of this sort provides any reason for abandoning critical epistemology, which is where I shall argue that naturalization would in effect lead.


4. There are also more radical alternatives that would take naturalization farther than Kitcher wants to go, e.g., by abandoning any idea that epistemology has a normative dimension. But apart from Quine's own view, which seems to have such a consequence, I will not be concerned here with these other views, which seem to me to be even further from the truth.


5. A partial exception is the explicitly psychologistic accounts that Hume offers of inductive reasoning, the apparent inference to the external world, etc. But these are precisely the pans of Hume's philosophy, I submit, that have the least value in relation to the philosophical issues in question and indeed which are the hardest even to recognize as even constituting philosophy.


6. For a fuller characterization. see my paper "A Rationalist Manifesto," Canadian Journal of Philosophy, Supplementary Volume 18 (1993): 53-88.


7. Reprinted in Quine. Ontological Relativity and Other Essays (New York, 1969), 69-90. Further references in the present Section to toe pages of this article will use the abbreviation "EN," and will be placed in the text.


8. Quine, "Two Dogmas of Empiricism." reprinted in Quine, From a Logical Point of View, 2nd ed. (Cambridge, Mass., 1961), 42.


9. Though this is by no means apparent from the actual texts, it is possible that Quine would want to argue that the goat of even the more modest construal of "the doctrinal side" is rendered unachievable by the complete intractability of the problem of induction. I do not think that such a pessimistic view of the problem of induction is warranted, but a discussion of that issue is impossible here, See my "A Reconsideration of the Problem of Induction," Philosophical Topics 14 (1986): 93-124.


10. Or rather we believe that (accept a sentence saying that?) we have such beliefs or accept such sentences.


11. It is important, however, to see that the main issue here does not turn on the term 'knowledge'. Even if, as some believe, the ordinary meaning of 'knowledge' does not require epistemic justification in the sense advocated here, but only something like reliable or truth-conducive causation of belief, it would remain true even for beliefs that constitute knowledge in this sense that we have no reason at all for thinking them to be true, and that result is enough in itself to constitute a very deep and intuitively paradoxical version of skepticism.


12. Notice also in passing that the belief that this is the best that we can do, that naturalized epistemology is all that is possible, is obviously not itself a psychological claim and thus cannot be part of the content of such an epistemology.


13. Quine, "The Nature of Natural Knowledge," in Mind and Language, edited by Samuel Guttenplan (Oxford, 1975), 67-81; the quoted passage is from p. 67.


14. Ibid., p 68.


15. Of course, some sorts of occult beliefs may stand in conflict with the sort of psychology that Quine has in mind. It is, however, not clear why such a conflict poses any problems once issues of justification are set aside; and in any case, there will be or could be other, occult versions of psychology that Quine can offer no reason for not taking just as seriously as the scientific brand.


16. The second component and some of its supporting arguments also originate with Quine and were almost certainly influential in leading him to advocate the naturalization of epistemology, even though they are not mentioned in any very explicit way in "Epistemology Naturalized" itself.


17. For further discussion of this same argument, see Hilary Kornblith, "Beyond Foundationalism and the Coherence Theory," Journal of Philosophy 72 (1980): 597-6l2; reprinted in Hilary Kornblith (ed.), Naturalizing Epistemology (Cambridge, Mass., 1985), 115-28, esp. pp. 118-19.


18. I am not denying that people may have sometimes carelessly said things that might suggest the view in question, partly due to the grip of the analytic methodological preconceptions briefly alluded to above and partly because the precise sort of case that Kitcher describes had not occurred to them (though it no doubt should have). What I am denying is that there has ever been a significant philosopher who, once clearly presented with the argument, would have rejected the conclusion or who would have felt that it was a significant emendation of his views to accept it. Ironically enough, a good part of the motivation for the formulations that might suggest the contrary was a reluctance to formulate matters in terms of intentional mental states, states which were thought to be philosophically suspect for reasons that are quite close to some of those that often motivate naturalists.


19. Kitcher, of course, does not accept the category of logical or conceptual truth, and so places descriptions of the quoted point as logical in scarequotes. But this issue, to be discussed later, is irrelevant here.


20. See P. F. Strawson, Introduclion to Logical Theory (London, 1952), chap. 9, Part 11.     For a critique of Strawson's account from the standpoint of traditional epistemology as understood here, see "A Reconsideration of the Problem of Induction," pp. 104-06.*


21. For a detailed discussion of the vicissitudes of the concept of analyticity, see "A Rationalist Manifesto," 62-69; and also Appendix A of my book, The Structure of Empirical Knowledge (Cambridge, Mass., 1985).


22. See "A Rationalist Manifesto," 71-72.


23. Quine, "Two Dogmas of Empiricism," 2~37. Oddly enough, Kitcher does not mention this argument.


24. For a discussion of other arguments against rationalism, see my book In Defense of Pure Reason, forthcoming (eventually).


25. "Two Dogmas of Empiricism," 37-46, especially pp.42-46. A second possibility worth noting is the argument for the indeterminacy of radical translation, which some have taken to be an argument against the a priori. Both this argument and its relevance to the a priori seem to me too uncertain and problematic to be worth discussing here. But see "A Rationalist Manifesto," 76-79.


26. See "A Rationalist Manifesto," 72-76, for a fuller discussion.


27. At least this is true for the cases of instrumentalism or fictionalism and behaviorism. Whether the same thing is true in the ease of phenomenalism (or instead whether something like "direct realism" is true in a way that avoids the need for such an inference) is an issue that I cannot go into here.


28. Kitcher has argued elsewhere in effect that the only way to make sense of the concept of a priori justification is to construe it as guaranteeing truth and so ruling out fallibility. See his "A Priori Knowledge," Philosophical Review 86 (1980): 3-23; and The Nature of Mathematical Knowledge (New York, 1983), chaps. 1-4. I believe that this claim is mistaken, but an adequate consideration of it is impossible here.


29. I have formulated the argument in terms of a reason for thinking that the belief is true, rather than in terms of the belief's being (epistemically) justified, because I do not want to enter here into the somewhat vexed controversy between externalist and internalist conceptions of justification.  My view is that the result arrived at in the text is enough to constitute epistemological disaster whether or not the beliefs in question may be said to be justified in some other sense of justification that does not involve our having a reason to think that they are true.


30. I owe my appreciation of the fundamental importance of the issue of self-referential consistency or inconsistency to the teaching, long ago and far away, of Richard Rorty. Unfortunately, however, Rorty's own philosophical views seem to me to have long since ceased to reflect any concern for this kind of issue.


31. Philip Kitcher, The Advancement of Science (New York, 1993), 35.


32. Many philosophers have suggested that enumerative or instantial induction may be justified by appeal to explanatory reasoning. For my own version of this idea, within a general rationalist context, see "A Reconsideration of the Problem of Induction," 11-2l.


33. I am grateful to Ann Baker and Larry Colter for helpful comments on earlier drafts of this essay.