A L V I N I. GO L D M A N ' S E P I S T E M O L O G Y A N D AN INTRODUCTION
COGNITION:
MICHAEL P. LEVINE 'Epistemics: an enterprise linkingtraditional epistemology,first with cognitive science and, second, with social scientific and humanistic disciplines that explore the interpersonal and cultural processes impinging on knowledge and belief' (Epistemology and Cognition, p. vii) I
In this book Goldman takes up the first part of'epistemics', a term he coined, discussing the links between traditional epistemology and cognitive science. (A sequal on social epistemics is underway.) He focuses on reliabilism, particularly the reliabilist account of epistemic justification, of which he is the primary architect and proponent. He sees reliabilism as well within the framework and concerns of traditional epistemology, and argues for their continuity in the account of his conception of epistemology. He discusses reliabilism's relation to significantly related metaphysical views as well, e.g. truth and realism - - defending it against certain metaphysically based objections (chapter 7). For example, he defends truth-linked epistemic evaluative standards against charges of circularity and vacuity. He argues for a nonepistemic account of truth and a realist metaphysics showing their link with reliabilism and his approach to epistemology generally. His discussion of realism and anti-realism, truth and correspondence (sections 7.2 and 7.3) is not just a survey or explanation of how these issues fit with his epistemology, but a substantive contribution to these metaphysical questions. It is very useful to have such a complete account of reliabilism readily available, and if this was all the book was about it would more than suffice to make it valuable. His comparison and critique of various 'alternative reliable-process theories' of knowledge (section 3.2) is brief but excellent (more than a survey), and his discussion of epistemic 'Objectivism, Relativism and Related Issues' (section 4.3) is significant in its own right. In this latter section comparisons with the issues' counterparts in ethics are a helpful propadeutic. However, the main 209
MICHAEL P. LEVINE purpose of the book is not to reiterate and defend reliabilism. Although reliabilism is a major theme and Goldman's version is defended, the adequacy of some type of reliabilism is more or less presupposed. Goldman's project is far more ambitious. 'The aim of this book is to redirect and restructure the field of epistemology.., epistemology should be a multidisciplinary affair, not the province of pure, a priori philosophy'(l) I shall have something to say about whether Goldman's extraordinarily bold aim has been - - or should be - - met, after an account of only a few of his major themes, and raising some questions about aspects of just a few of his many arguments. (Part 1 is packed with arguments, and my questions are not always about particularly important aspects of his important arguments.) Whether or not his stated aim succeeds, Goldman has produced a formidable and important work, and one that will be influential. The book has enormous breadth and depth. Goldman writes with an'impi'~ssive degree of clarity and'readability' for such a closely argued and difficult work. (The only exception to this might be in his exposition of some of the examples he uses from cognitive science for illustrative purposes which, on occasion, I found either overly detailed and unecessarily technical, or not detailed enough.) On Goldman's account I am justified in believing that his book will be influential if the cognitive processes causally operative in producing that belief are sufficiently reliable. The processes relevant to reliabilism are 'purely internal processes.' 'Factors external to the cognizer's psychology' should not be invoked in determining in each case which process type is to be evaluated in terms of reliability (50). Goldman conjectures that in each case 'the critical type is the narrowest type that is causally operative in producing the belief token in question' (50). His discussion of the relationship between cognitive science and epistemology, whether or not one finally agrees with it, will have many rethinking their views. It will give others a framework with which to think about epistemology as multidisciplinary, and its interface with cognitive science. It raises questions about the very nature of epistemology - - its scope and limits - - and some may wonder, if Goldman's views are right, whether traditional epistemology survives the interface rather than giving way to cognitive science altogether. Goldman would deny that it does. His epistemology differs significantly from others which also favor interdisciplinary approaches, such as Quine's naturalistic and Donald Campbell's evolutionary approaches, in maintaining the normative and 'evaluative mission of epistemology' rather than a purely descriptive one. All of this is important since 210
EPISTEMOLOGY AND COGNITION: INTRODUCTION questions concerning the relationship between epistemology and empirical sciences are not only not going to disappear, but already predominate. The book is divided into two parts; 'Theoretical Foundations' and 'Assessing Our Cognitive Resources.' He begins with an account of 'the elements of epistemology, 'skepticism and knowledge. He then gives an account of a rule framework for justification of belief and defends reliabilism in the context of such a framework. He describes and defends specific criteria, truth-linked evaluative standards, that have to be taken into account in formulating the rules of a right justification - rule (J-rule) system. These include the capacity for problem solving, power and speed. Part I is not only a defense of reliabilism, but a general introduction to epistemology as well in which various positions, problems and proposed solutions are categorized and discussed. This establishes the theoretical foundations for part II in which he gives an account, by way of a variety of examples, of how reliabilism is necessarily reliant upon cognitive science in delineating rules for justification. Formulating right rules of justification apart from cognitive science simply is not possible. He argues that various findings in cognitive science about, for example, perception, memory, deductive reasoning, probability judgments, belief acceptance and uncertainty, etc. can be used for assessing the strengths and weaknesses of our cognitive processes in terms of the reliabilist criterion of justification. He undertakes such an assessment in an admittedly partial and preliminary way, noting that such assessment will naturally undergo expansion, revision and refinement in view of further developments within cognitive psychology. His assessment of human cognitive resources is summarized in the following: '...[M]y own tentative assessment will be neither warmly glowing nor severely disparaging. It will be more qualified and moderate in tone. The mind uses all sorts of cognitive processes and process variants. In terms of, say, reliability, they are a mixed bag: some good, some not so good. Moreover, most of them have significant trade-offs. The less reliable processes are often quicker...' (183). Trade-offs and mutual supports between the evaluative dimensions of reliability, speed and power, are discussed - - e.g. in 'Memory' (Chapter 10). II 'The notion of justification is directed, principally, at beliefs. But evaluations of beliefs.., derive from evaluations of belief-forming processes. Which processes are suitable cannot be certified by logic alone. Ultimately, justificational status depends (at least in part) on 211
MICHAEL P. LEVINE properties of our basic cognitive equipment. Hence, epistemology needs to examine this equipment, to see whether it satisfies standards of justifiedness' (4). This is part of the task of what Goldman terms 'primary (individual) epistemology.' 'It is fully appropriate, then, for [primary] epistemology to inquire into cognitive architecture, to assess its strengths and weaknesses.., within primary epistemology.., the objects of epistemic evaluation are cognitive processes, structures and mechanism.' (5) In discussing 'terms of epistemic appraisal' he argues for a widening of the scope of the terms that 'academic epistemology' has traditionally considered. 'Most evaluations.., have been in the category of evidential evaluations.., many terms of widespread intellectual evaluation have not figured prominently in analytic epistemology, although there is no good reason why they should not' (21-2). For example the term 'intelligent is especially worth mentioning here because it does not fall (wholly) in the evidential category' (22). 'Originality' is another significant term of epistemic appraisal that has been neglected. Goldman draws a parallel between the notion of justifiedness in ethics and in epistemology. Epistemic evaluation is treated by Goldman both in the context of rules (i.e. deontically), and also in terms that '...apply no deontic category...[but] merely identify certain goodmaking characteristics.., evaluations.., in the virtue-vice style' (25). He maintains that there is a factual basis that underlies both evaluative and epistemic evaluation - - and upon which justifiedness supervenes (23). 'Consider justifiedness. What has to be true of a belief in order for it to qualify as justified? What factual standard determines justifiedness?' (23). One prominent constraint on any answer to this question is that '[i]t is not admissible to answer by using other terms of epistemic appraisal.., we need nonevaluative terms or conditions' (23). A specification of conditions under which a belief is justified must use only non-epistemic terms. Three standards of evaluation Goldman discusses are reliability, power and speed. He applies these to psychological processes. '...[R]eliability is one component in a complex standard appropriate to justification' (26). Reliability helps one to avoid error. 'Power is the capacity of a process.., to produce a large number of true beliefs...'(27). '...[T]he power standard, especially the problem-solving variant, is associated with the evaluational term 'intelligent' (27). Chapter 2 is a discussion of reliabilism's impact on the problem of skepticism as traditionally conceived. 'Where skepticism is predicated on the frailties or inadequacies of human cognitive powers, the psychology of cognition may address this claim' (30). 212
EPISTEMOLOGY AND COGNITION: INTRODUCTION In Chapter 3 Goldman presents his causal reliability approach to knowledge. 'Whether a true belief is knowledge depends on why the belief is held, on the psychological process that causes the belief... '(43). Goldman argues that a 'no-relevant-alternatives conditions,' the 'local reliability requirement' in his account of knowledge can handle Gettier and post-Gettier examples, though admittedly he does not have a 'detailed theory of relevance' (55). (Such a theory would presumably reflect a degree of inexactness in the concept of knowledge itself that Goldman calls attention to.) S's true belief may not be knowledge if there are relevant alternative counterfactual situations that S cannot discriminate from the actual state of affairs (54-55). In chapters 4 and 5 Goldman explains his rule framework account of justification (J-rules). 'S's believing p at t is justified only if (a) S's believing p at t is permitted by a right system of J-rules, and (b) this permission is not undermined by S's cognitive state at t' (63). A J-rule 'permits transitions from prior cognitive states to beliefs... [or] to any doxastic attitudes'(77). In order to identify the right system of J-rules a criterion of rightness must first be determined. The system(s) of rules which satisfies the criterion may then be investigated. Goldman's efforts focus on the criterion of rightness. We want only nonepistemic notions to appear in a criterion of rightness' (63) ...[N]o J-rules may be derived from logic alone... logic cannot serve as even a partial criterion of J-rule rightness (81) ...Logic is completely silent about states of belief, or other cognitive subjective matter. Yet that is precisely the concern of J-rules, namely, transitions to states of belief (or other doxastic attitudes) (82) ...[W]e must conceive of... [J-rules] as specifying cognitive processes [not cognitive-state transitions], where by 'process' we mean a determinate kind of causal chain. This is the sort of thing that logic, by its nature, cannot provide, but precisely the sort of thing naturally sought from psychology.., no criterion will be plausible unless the rules it authorizes are permission rules for specific (types of) cognitive processes (85). Goldman notes that'the process approach goes against a substantial tradition concerning the nature of justification (or rationality)'(89). He criticizes evidence proportionalism.., the idea that justifiedness consists in proportioning your degree of credence in a hypothesis to the weight of your evidence.., whatever the degree of confirmation of evidence on hypothesis.., these relationships cannot fix the 213
MICHAEL P. LEVINE justifiedness of alloting a selected degree of credence vis-a-vis the hypothesis. The question will always arise: What understanding does the cognizer have of these relationships? If he lacks proper understanding, then he is not generally justified in assigning the indicated degree of credence. But 'understanding' can only arise from suitable cognitive processes. So a specification of justifiedness must make reference to such processes (89-90). But on certain interpretations of evidence proportionalism the cognizers correct understanding of the evidence is presupposed. With this supposition in mind the evidential proportionalist would insist that, normatively, J-rules require proportioning belief to evidence. If one ignores evidence one understands, or perhaps even should understand, then one will not be justified in one's belief. Of course justification will, in part, be a function of what the cognizer understands of the relation of evidence to hypothesis. Still, Goldman misrepresents evidence proportionalism in suggesting that it would hold that '...all talk of 'processes', especially psychological processes, seems to be irrelevant' (90). This may be true of some versions, but it is not true of all - - or the most plausible. I do not see how a criterion of rightness of J-rules which does not take into account evidence proportionalism (its main idea) can meet the constraint that Goldman puts on any adequate theory of justification - - which is that it must conform to our ordinary standards of justification (more or less). Granted that 'a specification ofjustifiedness must make reference to such [cognitive] processes' does this undermine the evidential proportionalist claim that justification requires proportioning belief to evidence where understanding is presupposed, or alternatively, it is supposed that one should understand? I do not see that this is wrong or where it fits into Goldman's theory of justification. It may fit in his account of secondary justifiedness. Whereas primary justifiedness 'results from the use of approved processes... Secondary justifiedness results from the use of approved methods' (93). (Counterexamples to reliabilism being sufficient or necessary for justifiedness are discussed in section 5.6.) Perhaps Goldman would say the same thing about evidence relations he says about coherence. 'Any theory of justification that neglects coherence relations entirely is liable to be defective. But it does not follow that the coherence requirements should be built into the criterion of rightness. My contention is that coherence is not the highest standard ofjustificational rightness but a derivative, subsidiary, standard. The fundamental standard concerns the formation of true belief. Coherence 214
EPISTEMOLOGY AND COGNITION: INTRODUCTION enters the picture only because coherence considerations are generally helpful in promoting true belief' (100). However, whatever the highest standard ofjustificational rightness is (and how is this to be determined?) an adequate theory of justification must not violate Goldman's constraint that it must conform to our ordinary standards of justification (more or less). But can a theory do this if its criterion ofjustificational rightness does not include coherence and evidence relations as part of it, and not merely as derivative or subsidiary standards? I am not clear that it can, or more importantly, I am not clear why Goldman thinks it can. What is a subsidary standard and what isn't? Why needn't a criterion of justificational rightness include its subsidary standards as part of it? Aren't they internal and not external to it? Is it enough for J-rules to promote true belief in order to qualify as right? As a theory of justification does this conform to our ordinary standards of justification? Apart from considerations or derivative standards like coherence, explanation, evidence etc., it seems that the criterion is too broad to base an adequate theory of justification on - - one which meets Goldman's constraint. Why is truth-conduciveness, by itself, apart from coherence conduciveness etc. relevant to the criterion? Why are these other considerations secondary and relevant only at the level of J-rules rather than that of the criterion ofjustificational rightness itself?. 'As in the case of coherence, it may turn out that explanatory factors have an appropriate place at the level of J-rules, should explanatory processes indeed be truth-conducive. But explanatoriness should not replace verific values at the level of the criterion' (101). But even if a 'replacement' is not in order might not a (serious) complication be? Goldman says that '...the verific approach to justificational value in the objective consequentialist style - - is the most promising. True belief is the value that J-rules should promote - - really promote - - if they are to qualify as right...'(103). If his criterion is correct (complete) then this J-rule requirement would follow. But this requirement is not useful as a support for the correctness of the criterion. I have mentioned the constraint Goldman places on an adequate theory of justification. It must conform to our ordinary standards of justification (more or less). How does this fit with his opting for 'an absolute, resource-independent criterion ofjustifiedness' (105)?'A resource-independent criterion fixes an acceptable truth ratio without any regard to the resources of the (type of) cognitive system in question. A resource-relative criterion fixes an acceptable truth ratio as a function of the target cognitive system's resources, in the present case, human cognitive resources... A resource-independent criterion is liable to be absolute rather than 215
MICHAEL P. LEVINE comparative. It would set an absolute truth-ratio that must be met, whether or not human resources are capable of meeting it' (104). Why does Goldman opt for the resource-independent criterion? He says: 'One virtue of this approach is that it makes the challenge of skepticism both serious and credible. If some fixed level of truth ratio is necessary for justifiedness, it is an open question whether human processes ever yield justified belief, whether they are even capable of yielding justified beliefs' (105). But I do not see how in opting for 'an absolute, resource-independent criterion of justifiedness" he can both meet his constraint and take skepticism seriously. For in any conflict between such a criterion and the constraint, one or the other would have to be abandoned. If the criterion told us that none, or nearly none, of our beliefs were justified, the constraint would be violated. However, this constraint in turn, puts constraints on the absolute, resourceindependent criterion of justifiedness. Indeed it appears that the criterion must be resource dependent and comparative, rather than independent and absolute, given Goldman's constraint on an adequate theory of justification. Goldman's rightness criteria are 'verific consequentialist criteria, and these are a species of truth-linked criteria' (116). He discusses some possible difficulties concerning their application, and the charge of circularity in their determination (section 5.8). He successfully refutes Putnam's charge of definitional circularity given that truth is not defined as 'rational acceptability.' However, I do not think he successfully deals with one of Firth's objections. 'Firth appears to be charging that in order to substantiate that criterion [a reliabilist criterion ofjustifiedness] we must presuppose it. This charge rests on a misplaced assumption. It is apparently assumed that we can only substantiate the criterion by generalization from observed 'data'. We observe warranted beliefs, we observe their truth, and we infer a correlation between the two... [But] We do not have to observe warranted beliefs and determine their actual truth-values to decide whether the criterion is correct. As in other cases of analysis, one can proceed entirely by hypothetical cases...'(117). This response is not adequate since Firth could easily agree that we can proceed by hypothetical cases without undermining the objection. Firth's point is simply that in supposing certain beliefs to be justified, and so suitable for substantiating the criterion, the criterion must be presupposed in accepting certain beliefs, hypothetical or not, as justified in the first place. Whether or not beliefs are in fact justified given that the criterion is in fact the correct one, has nothing to do with the fact that the criterion has to be presupposed as Firth says. Goldman counters Firth's charge of circularity in other ways as well, but I think 216
EPISTEMOLOGY AND COGNITION: INTRODUCTION unsuccessfully. Also, he says '...truth-linked criteria are no worse off than any other criteria in the matter of application. To apply any sort of criterion, some reasoning and other cognitive processes are required. How else could one decide whether a particular rule-system satisfies a given criterion...' (118). Yes - - but what has this fact to do with the charge of circularity? The charge of circularity does not arise in the application of the criterion but in its substantiation or validation. If there is a'problem of the criterion'concerning justification in which one must either (a) presuppose some of the things that satisfy the criterion in order to arrive at the criterion or (b) presuppose a criterion in order to determine what satisfies it, then it is a problem for Goldman's criterion, like any other. Goldman criticizes Putnam's epistemic account of truth. Putnam says, 'Truth is an idealization of rational acceptability. We speak as if there were such things as epistemically ideal conditions, and we call a statement 'true'if it would be justified under such condition.' Goldman says: 'In order to clarify this proposal, let me reformulate it as follows .... If someone were in epistemically ideal conditions vis-a-vis the question of p versus not-p, then he would be justified in believing p' (146). But how is this a'reformulation' of Putnam's proposal at all? Surely Putnam means that /ff one is justified in believing p in such ideal epistemic circumstances then p is true, and tffone is justified in believing not-p in such circumstances is not-p true. Instead Goldman says'The correctness of this proposal [i.e. Goldman's reformulation of Putnam's proposal] depends on how 'epistemically ideal conditions' is defined.., the equivalence could succeed... [if] we define 'S is in epistemically ideal conditions vis-a-vis p as: S is so situated and so constructed that he would believe p if and only if p is true...' (146-7). But this does not, I think, capture Putnam's meaning. In rejecting Putnam's formulation Goldman says: 'The problem.., is that the proposed definition of 'epitemically ideal conditions' reintroduces the notion of truth.., there is no reason to think that Putnam's equivalence can succeed unless its crucial epistemic notion is cashed out circularly, in terms of the (realist) notion of truth'(147). But Putnam would deny Goldman's reformulation, in terms of an equivalence that hinges upon the notion of 'epistemically ideal conditions,' is an accurate reformulation at all. For one thing he is not defining 'truth.' He is giving a non-realist account of truth that makes Goldman's reformulation beside the point. Putnam would reject Goldman's definition o f ' S is in epistemically ideal conditions vis-a-vis p...' as presupposing the very realist notion of truth he wants to eschew. Putnam's argument for his account of truth does not rest on the 217
MICHAEL P. LEVINE equivalence G o l d m a n presents as necessary to P u t n a m ' s account at all. That equivalence is not necessary to Putnam's account. Note, I am not trying to dispute the correctness of a truth-linked theory of justification or a realist notion of truth, but only seek to show that Goldman's criticism of P u t n a m (and Dummett), no matter how mistaken that account is, seems to have missed the mark. In 7.4 G o l d m a n defends a version of scientific realism and notes that given his account of justification this presents some prima facie difficulties. ...[H]ow can it be epistemically legitimate for scientists to aim at getting theoretical truths? How is it possible for them to be justified in believing scientific theories so construed? This seems to be a particular problem given my theory ofjustifiedness. After all, if beliefs are justified only when caused by reliable processes, and if scientific methodology has a poor historical track record, the processes in question must not be reliable. But then belief in scientific theories, at least construed realistically, is not and cannot be justified... The p r o b l e m is crystallized in what Putnam calls the "disastrous meta-induction" of scientific theorizing. Suppose scientists decide that no term in the science of more than fifty years ago referred, and no theory of that vintage was true... shouldn't we be led to the meta-induction that no theoretical term now or in the future will refer, and no present or future theory will be true? (158) G o l d m a n successfully argues (contra Putnam) that 'within my full theory of justification, a poor track record of scientific beliefs.., need not preclude the possibility of justified beliefs in scientific theories' (160). However, he also claims that Putnam's 'disastrous recta-induction is really self-undermining' (160) and I don't think this is right. He says: 'We judge past theories to be false only by the lights of our present theory. If we a b a n d o n out present theory, we are no longer in a position to judge past theories false. So if we use the meta-induction to conclude that no present scientific theory is true, we thereby eliminate all grounds for believing the past theories were false. But then we are no longer entitled to believe the premises of the meta-induction' (158). G o l d m a n has neglected the ' m e t a ' character of the induction altogether. He assumes that the premises of the meta-induction must presuppose our current scientific theory to be more or less correct, but this is no part of the premise of the induction. The premise does not suppose the truth of current theory in supposing past theories to be false. That is what makes the induction ' m e t a ' rather than induction per se. Meta-theories are 218
EPISTEMOLOGY AND COGNITION: INTRODUCTION often mistakenly taken to be self-undermining in the way Goldman thinks this meta-induction is. For example, relativism of various types is often mistakenly taken to be self-defeating and the argument is similar to that of Goldman's. In chapter 8 'The Problem of Content' Goldman defends his view of epistemology and its need for cognitive psychology against a number of views in philosophy of language, mind and cognitive psychology itself concerning - - especially but not exclusively - - problems about the alleged content of mental states that would undermine his project. An example of a cognitive process (e.g. perception) that Goldman examines in part II (chapter 9) will illustrate the kind of task he undertakes with respect to a range of such processes. Within perception there is ' b o t t o m - u p ' and ' t o p - d o w n ' processing. In b o t t o m - u p processing 'information flows from little perceptual pieces to large units built from them.., in top-down processing.., higher-level beliefs, or background beliefs, influence the interpretation of low-level perceptual units. Psychologist's have particularly shown that (knowledge of) a pattern's context influences how one perceives that pattern' (186). Goldman's conclusion is that although 'reliance on context can breed error. It does not follow.., that top-down processing generally promotes error. On the contrary, the contexts.., may generally foster reliability... At present, then, we have no reason to think that top-down processing per se is a negative characteristic of perception, at least from the vantage point of reliability and j ustifiedness' (190) and it is psychology that shows this to be the case. Goldman says: 'The reliabilist criterion of rightness is neutral about the precise character, or content, of right J-rule systems. As far as reliabilism goes, a system - - or its rules - - could be either foundationalist or coherentist in character... Within my approach, F-ism and C-ism are best viewed as these about the contents of right J-rule systems' (195). However, foundationalism and coherentism as theories of the structure of epistemic justification, are also neutral about content and one might as easily say that the J-rules, or principles, of epistemic justification in foundationalism are reliabilist in nature - as many surely are. Foundationalism and coherentism are normative accounts metaaccounts as well) of the structure of epistemicjustification. They offer a c r i t e r i o n of justification (that Goldman rejects) and as such are n o t compatible with reliabilism. The manner in which he viewed foundationalism and coherentism is an equivocation on the nature of those theories as accounts of justification. On a supreme normative (or meta-epistemological) level the accounts are incompatible. To describe 219
MICHAEL P. LEVINE foundationalism and coherentism as theses about the contents of right J-rule systems is at best misleading since in so doing one is no longer talking about foundationalism and coherentism as normative accounts of the structure of justification at all. Goldman proposes that'the foundationalist-coherentist controversy depends for its resolution on psychological facts.., but.., not... exclusively'(195). But this cannot be true, any more than reliabilism as a criterion ofjustifiedness can depend upon psychological facts for a determination of its adequacy and correctness. 'My reliable-process criterion of rightness implies that whether a set of J-rules is right depends on two things: (1) the cognitive processes available to human beings, and (2) their reliability. Whether there are any right F-ist J-rule systems, or any right C-ist J-rule systems, also depends on these facts. Hence, whether right J-rule systems are F-ist or C-ist in character depends on psychological facts' (197). The character may well depend upon psychological facts for the reasons given by Goldman - - but the characterization of these systems as C-ist of F-ist is unwarranted. In chapter 13 Goldman discusses the 'epistemological significance of psychological research on deductive reasoning... The psychology of reasoning is relevant to justifiedness.., because it inquires into the basic cognitive processes that can occur in reasoning.., lit] is also linked to the evaluative dimension of rationality' (278). In this chapter he also discusses the alleged distinction between a priori and aposteriori knowledge, with an eye towards undermining it. The distinction is discussed primarily in terms of the role of perception in knowledge acquisition. The prospect of distinguishing the two realms in terms of perceptual knowledge seems dim. There is no need to deny, though, that simple, logical truths may be humanly knowable without any perception, and this trait may indeed, contrast with contingent truths about external matters. Even granting this point, though, can it grant the a priorist's hopes? No. The traditional hope.., is to draw an epistemological dichotomy between the entire realm of logic and mathematics, on the one hand, and the realm of contingent fact, on the other... Notice also that perception is only one of many cognitive faculties.., there is extensive sharing of other cognitive operations... Why harp so persistently on differences concerning the role of perception while downplaying the shared utilization of other faculties? (302) I don't see that Goldman has undermined the distinction. He thinks that because of constraints on human cognitive capacities there is at 220
EPISTEMOLOGY AND COGNITION: INTRODUCTION best 'nonperceptual knowability of a small set of logical truths.' For example, to know any complicated logical truths calculation is often required and perception is necessary to perform such calculations. But I don't see that this is pertinent to the alleged distinction since perceptual knowledge or other data acquired, in a sense, experientially is obviously necessary for a priori knowledge. A priori knowledge acquisition does not presuppose knowledge apart from any experience whatsoever. Concepts are acquired in part by means of perception etc. He says: 'Could one distinguish the realms.., by saying that propositions known a posteriori are, or can be, known by perceiving the known facts themselves.., by contrast, apriori knowledge only utilizes perception as an 'aid'?.... This obviously will not work. Some facts known aposteriori - - for example, facts involving electrons and nomological facts generally - - cannot themselves be perceived...'(302). But there is a sense in which facts about electrons etc. are perceived in a way that apriori knowledge is not and experience can figure in our knowledge of them in ways different from a priori knowledge. Electrons are not directly perceived but they are perceived indirectly by means of the senses. Goldman says: 'The standard explication of apriori knowledge is knowledge that does not involve, or depend upon, perception, or 'experience' (301). There may well be difficulty in determining just how a priori knowledge allegedly does not require or depend upon 'experience' as a posteriori does, indeed they may not even be a satisfactory way to do it. However, Goldman's effort in this regard merely seems to miss the point of the distinction since it is safe to assume, for instance, that no one who thinks there is such a thing as a priori knowledge is committed to the view that experience plays no role whatsoever in that kind of knowledge acquisition. Why Goldman thinks that an undermining of that distinction is important for his theory of justification is, I think, clear. Whereas foundationalism and coherentism are usually (not always) meant to give accounts of the structure of justification for empirical knowledge, Goldman intends his reliabilism to be considerably more far reaching claiming that other (alP) types of knowledge are within its scope. 'The exact extent of shared cognitive operations in logic tasks and empirical tasks requires more experimental and theoretical spadework... the cognitive ingredients of problem solving, at a deep level of analysis, are fundamentally similar across different domains. Hence, the classical attempt to dichotomize epistemology is fundamentally misguided, another example of the potential impact of psychological work on epistemological theory' (304). But one need not deny that the cognitive operations, or many of them are shared - - and still maintain that there -
-
221
MICHAEL P. LEVINE is an important distinction between apriori and aposteriori knowledge that has nothing to do with those aspects of the cognitive operations that are shared, or even some that are not shared. The distinction is not just one that allegedly rests on acquisition, but (possibly) on analyticity etc. as well. It is not clear whether Goldman thinks the distinction is fundamentally at odds with reliabilism, either formally or 'in spirit.' I do not think that there is any formal conflict, but the ways in which it might be seen to conflict with reliabilism in a less stringent manner is more complex and less clear. Goldman's point that 'perception is only one of many cognitive faculties' is more important for his reliabilist account, but it has nothing to do with the a priori/a posteriori distinction. III A reliabilist account of justification says, roughly, that a belief is justified if it results from reliable cognitive processes (37). '...[T]his account of justification would lead epistemology down a path that fosters collaboration with cognitive psychology'(37), since the question of what cognitive processes humans have is a question for psychology. But what is the nature of this collaboration? Granted that the question of what cognitive processes humans have is a question for psychology how does this indicate a collaboration? Perhaps it is merely the case that their various findings must not conflict. For example, in a specification of just which processes are reliable, the criterion of reliability used must not conflict with what epistemology has shown the criterion to be. But how exactly would cognitive science help to determine the criterion? Does it help in Goldman's own account of reliability as the criterion? Psychology does not play a role in determining what the criterion of rightness is. Only after such a determination has been made is it relevant in deciding which system(s) meet the criterion (66). But how then is this a collaborative effort? Does epistemology play any role in determining which system meets the criteria? No. Goldman says: 'Which rule system(s) satisfies the criterion is an empirical question. It is part of epistemology, of course to address this question' (80). But how can it b~? 'Suppose... the availability or nonavailability of such processes [sufficiently reliable cognitive processes] - - is settled, whether in the affirmative or in the negative. Would it follow that nothing is left for epistemology to do? Certainly not.., suppose we were guaranteed that there are sufficiently reliable processes to make knowledge and justified belief possible. Still, much work would be left. Epistemology should be 222
EPISTEMOLOGY AND COGNITION: INTRODUCTION concerned not only with whether there are reliable processes or methods, but which ones these are. Epistemology should be interested in specifying the good procedures, not simply ensuring that there are such' (39). 'A worthy epistemological task is to identify the comparatively reliable processes, to discriminate the better from the worse' (40). I do not see how or why this task of specification or discrimination could plausibly be seen as a task for epistemology rather than cognitive science. Once epistemology determines the correct standard of justification isn't it wholly a matter for cognitive science to delineate methods and processes which meet the criteria?, Goldman cites a fear of surrendering the 'autonomy' of epistemology as the worry of some who eschew a collaborative effort between epistemology and cognitive psychology. However, it seems that in this case more than autonomy is surrendered. Isn't the enterprise being handed over to cognitive science - - like an automobile on an assembly line? In a sense the project in its entirety is collaborative, but a more accurate description of the different domains of epistemology and cognitive science could indicate, as I think it does, that there is little in the way of collaboration. Instead what one has is one discipline leaving off where the other picks up. How would epistemology or cognitive science interact once a complete and accurate account of, for example, the reliabilist account of justification, were given? Goldman takes up the question o f ' W h y shouldn't epistemology stop at the selection of a rightness criterion? Why should epistemology include the job of identifying the J-rules... that actually satisfy the criterion?' (96) He notes that 'epistemology has traditionally been concerned not simply with abstract criteria.., but with the endorsement of particular methods or procedures.., it is just part of a full theory of justifiedness to say which J-rules... are right' (96). But how does this show that the task of specification is one for epistemology rather than cognitive psychology? '...[K]nowledge and justification, at least on my account, critically invoke one particular standard of appraisal: reliability. But that is not the only standard of epistemological interest. Even if no cognitive processes are sufficiently reliable to confer knowledge or justifiedness, indeed, even if all are equally unreliable.., distinctions could still be drawn according to other standards... [e.g.] power and speed'(40). Why is this an epistemic task? Again, power and speed can be empirically determined by cognitive science. The specification of the standards of appraisal may be an epistemological task, but not the specification of the actual processes that meet those standards. What has the epistemologist to do with this? Furthermore, is it clear that any normative judgment concerning the 223
MICHAEL P. LEVINE application of these standards of appraisal (e.g. which process should be used given its greater power, speed etc.) is an epistemic task rather than one for cognitive science? I can see where, in some situations, it would be an ethical task, but why epistemic? In his discussion of 'originality' Goldman says: 'One dimension of epistemological appraisal is the assessment of problem-solving procedures... When problems are difficult and previously unsolved by others, this may mean that unusual solution candidates need to be constructed. Hence, cognitive procedures involved in such construction fall under the scope of epistemology' (249). But I do not see how this is a problem for epistemology or how it falls under its scope. Goldman says: 'The principal way that cognitive science can contribute to epistemology, I claim, is to identify basic-belief forming, or problem solving, processes. Once identified, these processes would be examined by primary epistemology according to the evaluative dimensions and standards adduced... How do the processes fare in terms of justification, rationality, and intelligence? How well do they stand up, as judged by the three standards of reliability, power, and speed?' (181). I see this, as I think cognitive scientists do, as a wholly empirical task - - itself a part of cognitive psychology. Goldman thinks this aspect is epistemological because it is 'evaluative.' In what sense is this part of the task evaluative? Isn't it really just a measuring vis-a-vis certain standards rather than evaluation? The determination of whether a process meets a standard is empirical and not evaluative. This is another indication of how, on Goldman's conception, as I see it epistemology gives way entirely to cognitive science. However, there are some clear examples that Goldman gives where I can see the relevance of cognitive psychology - - 'epistemology needs contributions from cognitive science to select its descriptive apparatus for doxastic states' (324). More significantly; 'Our intuitive judgments ofjustifiedness and unjustifiedness, I believe, rest partly on a tacit grasp of the ease or difficulty of constructing relevant hypotheses. However, if we want to spell out the basis for these intuitive judgments, we need to identify the repertoire of operations and their standard preference ordering. That would give us a systematic ground for the indicated judgments... But this is a job for cognitive psychology' (250-I). Presumably, if one's belief were formed by a process such that one did not take suitable easily formulated alternative processes into account, then it wouldn't be justified on Goldman's account. Clearly cognitive psychology is relevant here, but is this the kind of fact that the claim that epistemology and cognitive science are, or should be interdisciplinary subjects is based on? Does it change the character of 224
EPISTEMOLOGY AND COGNITION: INTRODUCTION the way in which epistemology and cognitive science are to be conceived, or does it merely show that like in other fields (e.g. ethics and physiology) there is an interface? Goldman says, It is not the job of empirical science to make normative judgments... When it comes to epistemic normative judgments... this is the task of epistemology. However, if certain epistemic rules or principles are agreed upon, then the psychologist may be in a position to tell us whether or not human beings conform with these rules.., with what frequency or in what circumstances. But if there is any dispute about which normative principles are correct, this dispute goes outside the realm of psychology per se... principles of primary epistemology are partly determined by psychological facts. So on my view the psychologist can help epistemology ascertain correct normative principles... But this flows from the criteria of rightness (at least J-rulle rightness) that I have defended, and the defense of such criteria falls outside psychology altogether (306-7). Thus, Goldman himself appears to suggest that the task of epistemology is finished once the epistemic rules or principles are agreed upon. Furthermore, as I have said, I doubt that it is correct that empirical science qua empirical science cannot and does not make normative judgments. Indeed, once the criterion of rightness is stipulated, which is the task of epistemology, normative principles associated with cognitive processes will be stipulated by cognitive science. If cognitive science delineates J-rules (though not the criterion of J-rules rightness) why shouldn't it also delineate principles of sound processes etc. in accordance with those right J-rules?*
UNIVERSITY OF WESTERN AUSTRALIA NEDLENDS WESTERN AUSTRALIA 6009
AUSTRALIA
* My thanks to Tim Oakley for helpful comments. 225