Epistemology Naturalized and \"Epistemology Naturalized

MIDWEST STUDIES IN PHILOSOPHY, XIX (1994)

Epistemology Naturalized and “Epistemology Naturalized” RICHARD E. GRANDY

T

he phrase “epistemology naturalized” is habitually associated with Quine’s (1969) essay of that title, though the concept of naturalism is very

explicit and thorough in Ernest Nagel, Moms R. Cohen, John Dewey, and other earlier philosophers. In fact, if I were a historian of an earlier period, rather than the 1960s. I would perhaps argue that Locke was attempting to naturalize epistemology as he understood it, or, more accurately as he would have understood it if he had the concept of epistemology. The most famous passage of Quine’s essay deserves to be quoted in full for my main topic is the relation between epistemology, psychology, and the other cognitive sciences. I shall argue for a much broader construal of epistemology naturalized than that in “Epistemology Naturalized.” Carnap and the other logical positivists of the Vienna Circle had already pressed the term “metaphysics” into pejorative use, as connoting meaninglessness; and the term “epistemology” was next. Wittgenstein and his followers, mainly at Oxford, found a residual philosophical vocation in therapy: in curing philosophers of the delusion that there were epistemological problems. But I think at this point it may be more useful to say rather that epistemology still goes on, though in a new setting and a clarified status. Epistemology, or something like it, simply falls into place as a chapter of psychology and hence of natural science. (1969, p. 82) Which chapter of what psychology does epistemology become? Kornblith (1985, p. 1) structures the discussion around three questions: 1. How ought we to arrive at our beliefs? 2. How do we arrive at our beliefs? 3. Are the processes by which we arrive at our beliefs the ones by which we ought to arrive at our beliefs? 341

342

RICHARD E. GRANDY

Kornblith then distinguishes two theses about the relation between epistemology and psychology, which he calls strong replacement and weak replacement. On the strong replacement thesis, unemployment is inevitable for epistemologists since “psychological questions hold all the content there is in epistemological questions. On this view psychology replaces epistemology in much the same way that chemistry has replaced alchemy” ( 6 ) . One question I want to press is which chapter of what psychology epistemology is to become. Quine’s answer was Skinnerian behaviorism. “This human subject is accorded a certain experimentally controlled input-certain patterns of irradiation in assorted frequencies, for instance-and in the fullness of time the subjected delivers as output a description of the three-dimensional external world and its history” (82-83). Personally, I doubt that anyone, except Carnap in Der Logische Aufbau der Welt (1928), has eyer delivered as- output a description of the external world and its history. To put things into clearer perspective, let us consider some of the questions epistemology currently asks: For example, if we consider the Gettier/Goldman cases about knowledge in a case where George sees a red barn, but unknown to him there are many fake bams in the neighborhood, and his traveling companion was about to tell him about the fake bams. So instead of consulting our intuitions imagine that we send over to the psychology department the question whether George knows there is a red barn. I do not know of any department that is currently researching this kind of question. Or imagine asking the psychology department whether I am a brain in a vat-in this case the answer would be a plain, but unhelpful, “No.” And if we ask instead how I know I am not a brain in a vat, the answer is likely to be a referral to psychiatry. Perhaps this parody is unfair. But this strongly suggests that on whatever version of convergence or elimination one adopts, not all of epistemology is naturalizable, a point that was made by Stroud (1981). If epistemology is really at root all about skepticism, then none of the core will be naturalizable. I do not want to argue over the history or the core of epistemology, but hope to show that there are enough philosophically interesting problems in naturalized epistemology that it is worth pursuing regardless of how one ultimately delineates the historical connections. Let us return to the questions Kornblith posed, with my editorial emphasis: How ought we to arrive at our beliefs? How do we arrive at our beliefs? How well do the processes by which we do arrive at our beliefs match the ones by which we ought to arrive at our beliefs?

I do not think that contemporary cognitive psychology is yet prepared to provide much information beyond fairly elementary domains about how we do arrive at our beliefs. Consider my colleague’s belief that the sun will rise on July 4, 1994. He has lived long enough that a straight induction would not be unreasonable, and he is a philosopher of science so he knows what the

EPISTEMOLOGY NATURALIZED

343

straight rule of induction is. He also has sufficient knowledge of astronomy that he could give a theoretical backing to the claim, as well as an unpacking of the mistaken, or misleading, semantics of “sunrise.” But he almost certainly formed the belief long before any of these was the case. And, of course, we are abstracting from the fact that although he would undoubtedly assent to this statement, it has almost certainly never occupied his conscious attention yet. The fact is that psychologists, having practical goals such as publishing, have a strong preference for answerable questions-whereas philosophers generally, and epistemologists in particular, are prone to be concerned with questions which do not have direct experimental answers. They tend very much, therefore to pay attention to beliefs that have been formed in the recent past in situations that they have (attempted to) experimentally control and about which the subject had no prior beliefs. As part of the design, typical experiments eliminate any doubt in the experimenters’ mind about what is true ab initio. I agree with Quine that there are no natural boundaries between philosophy and psychology, but at present there is a largely uninhabited and unexplored area where the two interact only minimally, though there is potential for considerably more cooperation. Some philosophical pioneers in this regard are Goldman (1986) and Bach (1984).The main reason that I believe that epistemology would have much to learn from psychology if psychologists knew more about belief formation is that I believe that in epistemology as in ethics ought implies can. Epistemic agents cannot and ought not be faulted on the grounds that they did not follow epistemic strategies which are not cognitively possible for them. Among the older injunctions which fall into this category is “Make sure your beliefs are consistent.” for there is no general decision procedure for consistency. Thus the simple natural division of labor in Kornblith’s scheme which assigns problem (1) to psychology and (2) to philosophy does not work because of the empirical elements of (2). We are natural creatures, and the process of belief formation, knowledge acquisition, etcetera, are natural processes and can be studied by scientific methods. Note that I intend “scientific methods” to be taken broadly and I believe that one of the major tools of science is mathematics. Many naturalists are physicalists, but I regard it as an open question how naturalism will ultimately deal with questions in the philosophy of mathematics. Certainly Quine’s view that (some proper subset of) extant mathematics is to be accepted because it is necessary for physics is a matter of rational construction of our overall view, and does not bear, nor is it intended to bear, any connection to the way we form our mathematical beliefs. As natural creatures, we have some continuities with other species, though there are evidently very significant discontinuities too. This is yet one more angle from which to argue for the importance of images and other non-linguistic processing, as well as the centrality of non-conscious processing. Many of our intuitive natural language judgments about epistemic matters tend to be too coarse grained. For example, we tend to dichotomize remembering and not remembering. But there is an extensive literature on ‘implicit memory’,

344

RICHARD E. GRANDY

the characteristic of which is that subjects show evidence in various ways of previous exposure to material even though they cannot recall it on request. For example, subjects may be shown a list of words to remember. At a suitably later time they are asked to recall as many as they can. The important task though consists of subsequently completing word fragments. Subjects show a consistently higher ability to complete word fragments with words which they have seen on the list but could not recall in comparison to fragments which can only be completed with words not on the list. For example, if “donkey” was on the list (but not remembered) while “rabbit” was not on the list, then more subjects successfully complete the word fragment “d,,k,y” than complete ‘k-b-t” (Kolers and Roediger, 1984). The conclusion seems to be that memory is not a unitary phenomenon, but is dependent on the kind of task required as well as the kind of original exposure. The cognitive perspective tends to be that we are information processors, which tends to overlook, or at least to deemphasize, the fact that we are ambulatory and goal directed. I regard it as preferable instead to think of us as information hunters, gatherers, and processors. Since computers have been the dominant model in most of cognitive science, it is not surprising that there is a very strong emphasis on the processing of information, to the neglect, I argue, of the importance of how we acquire the information and what we do with it. This point may suggest a strong affinity with evolutionary epistemology. While something of importance may yet emerge from this general line of thought, almost all of the arguments which have been produced thus far under that rubric make little or no real use of the fact of evolution. At a conference on the topic at the University of Pittsburgh in 1989, an analysis of the papers seemed to show that the arguments about the appropriateness of our beliefs or our belief-forming mechanisms were only slightly diminished if one substituted the supposition that our species had survived intact in an indifferent environment for six millennia rather than several hundred thousand years. Returning closer to our opening line of discussion, Quine’s own formulation of the argument for naturalizing epistemology is flawed, in my opinion, by his insistence on identifying evidence with a class of sentences. This is related to an objection by Jonathan Dancy (1985) specifically to Quine’s claim that there is a contrast between “the meager input and torrential output” (83) of the subject. there is no contrast between meagre and torrential. The input is (together with other things) sufficient to cause the brain states which are its effects, and in following this causal story we are not any more studying a gap between input and output, if that gap is thought of as analogous to the gap between evidence and theory. (238) This is an important issue for understanding Quinian epistemology, for his full remark (“The relation between the meager input and the torrential output is a relation that we are prompted to study for somewhat the same reasons that always prompted epistemology; namely in order to see how evidence relates

EPISTEMOLOGY NATURALIZED

345

to theory”) certainIy suggests a conflation of causal and evidential relations. A neo-Quinian might respond that the output depends on brainstate, or on the previous history of the subject as well, and one can discuss Duhemianism in the brain. Rather than explore Quinian options here, I want instead to contrast the cognitivist point of view which construes information more broadly. From this perspective the input is vast, and the perceptual systems discard most of the information before it moves very far toward the central processing system. From a vast amount of information about what is currently happening at, e.g., the retinal surface, we get (sometimes) a modest amount of information about what is currently happening at a distance, or even about what has or will happen. The function of the perceptual systems is to process the information-since a great deal is redundant or irrelevant, the first level. of processing is primarily a matter of finding changes, i.e., edges, motion or color change. Looked at this way the process is a causal one, but it is the causal embodiment of an information-processing algorithm. There are assumptions about the world, e.g., of continuity and some underlying invariances, a few of which are hardwired and some of which are acquired through early experience. Having interposed this level of explanation in terms of the physical embodiment of algorithms which process the physical representations of information, one can then ask questions about optimality of the algorithms. The most forceful development of this approach is Marr’s Vision, and although many details of his specific conclusions may be faulted, the overall approach seems very promising for naturalizing epistemology beyond the perceptual. Traditional epistemologists may well still be unimpressed, for the conclusions one reaches tend to be hypothetical or conditional-if one has the following goals and these cognitive capacities, then the best you can do is.. . . But I am not certain that we have ever had anything better than such conclusions in traditional epistemology. Notice here that there is a shift in emphasis from questions about formation of individual beliefs to emphasis on the methods by which beliefs are formed in general. Thus the assessment seems to shift somewhat from the product (the belief) to the process (the process by which beliefs of this kind are formed). Although few psychologists have addressed the questions posed by Kornblith, some, most notably or notoriously Tversky and associates, have addressed the issues and concluded that we often do not form our beliefs as we ought. Having emphasized that epistemology could learn from psychology, I want to raise the, probably unnecessary, caution that we ought not to believe everything our colleagues tell us. In one experiment, subjects were confronted with two lights, and in each trial one of the two would light. The mechanism was set to randomize whether the left or right light went on, but with a bias so that (depending randomly on the subject) one light went on 70 percent of the time. On average subjects tended in the long run to guess that the more frequent light would be the next to light about 70 percent of the time. This means that they will correctly

346

RICHARD E. GRANDY

guess that light .7x.7, or 49 percent of the time; guessing the other light 30 percent of the time will produce a correct response 9 percent (.3x.3) giving a total of 58 percent correct responses. The simpler strategy of always guessing the light which had been lit most frequently up to that point would give 70 percent correct and the authors conclude the subjects perform suboptimally (Kahnemann and Tversky 1972). While this is true, it only shows irrationality or suboptimality if one assumes that the subjects knew, which they did not, that the sequence was random. If one is simply confronted with a sequence without knowing it is random then it is far from clear what is the optimal strategy. Before we can make our judgment as to what strategy for belief formation ought to be used we need to know not only which strategies we can use, but what the set of problems is, what the time constraints for solutions are, what the acceptable error rate is, and the relative value of different answers. I suspect, and have argued elsewhere (Grandy, 1987), that the many observations that subjects significantly misestimate with very low probabilities may have to do with survival value of not underestimating some risks. On the other hand, I do intend to argue that we are always optimal in our belief formation. The strongest example I know of to the contrary is the conjunction fallacy. In the most familiar example subjects make inconsistent probability judgments about a feminist lawyer. This example, however, allows some explaining away in terms of conversational implicatures or contextual features. The example I prefer is one in which subjects were given the following task (in early 1981). Suppose Bjorn Borg reaches the Wimbledon finals in 198 1. Please rank the following outcomes from most likely to least likely. Borg will win the match. Borg will lose the first set. Borg will win the first set but lose the match. Borg will lose the first set but win the match. In this study (Kahnemann and Tversky [1982], 96) 72 percent of the subjects ranked the conjunction “Borg will lose the first set but win the match” as more likely than its first conjunct. It could be argued, of course, that “but” is not interpreted as a simple conjunction in this case, but I find that line of argument much less persuasive in this case than in some others. Returning to my theme of several pages ago that we are information hunters and gatherers, as well as processors, I would also argue that we are also social animals. Thus the goals of the epistemic process can be dependent on social questions as well as individual ones. It has probably not been sufficiently appreciated that even Quine’s behavioristic characterization of observation sentences, the repository of all evidence, has a social dimension. This is explicit in the formulation in Quine (1969). “an observation sentence is one on which all speakers of the language give the same verdict when given the same concurrent stimulation. To put the point negatively,

EPISTEMOLOGY NATURALIZED

347

an observation sentence is one that is not sensitive to differences of past experience within the speech community” (8687). While it is less obvious that the equivalent definition in Word and Object incorporates reference to a community, Quine thinks it worth remarking on the contrast between the concepts of stimulus meaning and of observation sentence. ‘‘[Tlhe notion of stimulus meaning itself, as defined, depends on no multiplicity of speakers. Now the notion of observationality, in contrast, is social” (1962, 44-45). An alternative way of phrasing what Quine’s concept of an observation sentence captures is that those are the sentences where disagreement must come to an end within the linguistic community. The fact that a sentence is in that set is, however, something that is perhaps only known to an outside observer. A different, more usefully internal conception of an observation sentence can be given if we eschew the allegedly austere vocabulary.of behaviorism-and give a griceful twist to the idea. We can define an observation sentence instead as one about which everyone in the linguistic community believes that everyone in the linguistic community will give the same verdict given the same concurrent stimulation. These sentences and the relevant stimulations could be sought by members of the community as paths to resolving disputes. It is important to note from a naturalistic perspective, and in this instance perhaps the evolutionary one is helpful, that goals of groups and goals of individuals can be at odds with one another. Survival, or thriving, of a group may be best served by individual behavior which is not optimal for the individual. Here epistemology naturalized may have much to learn from the literature on such topics as prisoner’s dilemma. A natural metaphysical question that arises at this point is whether one sees the group or the individual as the basic unit of analysis and explanation. Two fairly representative but extreme positions are the following: What I propose. . . is a much more thoroughgoing contextualism than the one which urges us to remember that scientific inquiry occurs in a social context, or even that scientists are social actors whose interests drive their scientific work. What I urge is a contextualism which understands the cognitive processes of scientific inquiry not as opposed to the social, but as themselves social. This means that normativity, if it is possible at all, must be imposed on social processes and interactions, that is, that the rules or norms of justification that distinguish knowledge (or justified hypothesis-acceptance) from opinion must operate at the level of social as opposed to individual cognitive processes. (Longino, 1992) In contrast The conclusion is simple. The most promising approach to a general theory of science is one that takes individual scientists as the basic units of analysis. It follows that we must look to the cognitive sciences for our most basic models, for it is these sciences that currently produce the

348

RICHARD E. GRANDY

best causal models of the cognitive activities of individual human agents. (Giere, 1989) The view I am advocating accepts neither model, but sees the continuing dynamic interaction between group and individuals as critical. It is essential to see that although a group is in a sense constituted at a given time by a set of individuals, as a group changes over time members are attracted to the group or become part of it because of the properties of the group as a whole. The group, and the perception of the group, shape the cognitive behavior of those who join it. Moreover, epistemic evaluation seems appropriate for both individual and group processes, although the units and the measure of evaluation differ. Let me emphasize again that from a naturalistic perspective true beliefs are not necessarily an end in themselves but a means to the ends of the agent. For some agents true beliefs about a subject, e.g., prime numbers, may be an end in itself. But for many agents the beliefs are instrumental to some other end. Thus the goal is not necessarily maximizing the number of m e beliefs, that could be done by correctly identifying blades of grass or calculating whether numbers are prime, but to maximize the usefulness of beliefs. True beliefs, or perhaps even probable beliefs, about important issues may be more important than more probable beliefs about less important issues. This point leads to one criticism I would make of the fine survey "The Naturalists Return" by Kitcher (1992). While I agree with most of his delineation of recent naturalists, I have reservations about his list of basic theses of naturalized epistemology. His third principle is: (3) The central epistemological project is to be carried out by describing processes that are reliable in the sense that they would have a high frequency of generating epistemically virtuous states in human beings in our world. (75-76)

I do not see why a naturalist needs to be committed to this narow sense of virtue-less reliable beliefs about necessities may be more important than highly reliable beliefs about other matters. The central epistemological project would rather, on my account, be one of finding the most reliable methods of providing the important information at a reasonable cost in time and effort. In some circumstances, e.g., at the frontiers of new subjects, or in the difficult regions of old ones, there may be no strategies that are very reliable, but some may still be judged epistemically preferable! 1 NOTE 1. Earlier versions of this essay were read to the Philosophy Department Colloquium

at Rice University and as the 1983 George Myro Memorial Lecture at UC-Berkeley. I am indebted to members of both audiences for helpful comments.

REFERENCES Bach. K. 1984. "Default Reasoning: Jumping to Conclusions and Knowing When to Think Wice." Pacific Philosophical Quarterly 65:37-58.

EPISTEMOLOGY NATURALIZED

349

Carnap. R. 1928. Der Logische Aufbau der Welt. Berlin. Dancy, J . 1985. Introduction to Contemporary Epistemology. Oxford. Giere, R. 1989. “The Units of Analysis in Science Studies.” In The Cognitive Turn, edited by S. Fuller et al. Dordrecht. Grandy, R. 1987. ‘‘Information-based Epistemology, Ecological Epistemology and Epistemology Naturalized.” Syntheses 7 0 19 1-204. Goldman, A. 1986. Epistemology and Cognition. Cambridge, Mass. Kahnemann, D., and A. Tversky. 1972. On Prediction and Judgement. OR1 Research Monograph. Kahnemann, D., P. Slovic, and A. Tversky. 1982. Eds. Judgment under Uncertainty: Heuristics and Biases. Cambridge. Kitcher. P. 1992. “The Naturalists Return” Philosophical Review 10153-1 14. Kolers, P., and H. Roediger, 111. 1984. “Procedures of Mind.” Journal of Verbal Learning and Verbal Behavior 23~425-49. Komblith, H. 1985. Ed. Naturalizing Epistemology. Cambridge, Mass. Longino, H. 1992. “Essential Tensions-Phase Two: Feminist, Philosophical and Social Studies of Science.” In The Social Dimensions ofScience. edited by E. McMullin. Notre Dame, Ind. Quine, W . V. 1962. Word and Object. Cambridge, Mass. . 1969. “Epistemology Naturalized.” In Ontological Relativity and Orher Essays. New York. Stroud, B. 1981. “The Significance of Epistemology Naturalized.” Midwest Studies in Philosophy 6455-7 1. Tversky, A., and D. Kahnemann. 1982. “Judgements of and by Representativeness.” In Judgement under Uncertainty, edited by D. Kahnemann et al. Cambridge.