JAMES
TRADITIONAL NATURALISTIC
BOGEN
EPISTEMOLOGY REPLIES CRITICS
AND
TO ITS SKEPTICAL 1
. Why is it m o r e reasonable to act on the basis of beliefs expressed by sentences for which there is good perceptual evidence than beliefs expressed by sentences which are empirically disconfirmed or unconfirmed? Why, for example, does the overwhelming perceptual evidence for the truth of the sentence ' m y H o n d a needs a tuneup' m a k e it reasonable for me to head for a garage as soon as I can afford it? Why does the perceptual evidence which supports sentences about the involvement of the substantia nigra in the release of dopamine to the corpus striatum, and the degeneration of the substantia nigra in Parkinson's disease m a k e it reasonable to try the administration of L - D o p a for Parkinson's disease? A n d why did what I see when I c a m e into my office this morning m a k e it reasonable to reach o v e r here and to begin typing? Traditional epistemology's answer involves two ideas. T h e first is that it is reasonable to think that sentences which are subject to empirical testing and are well confirmed by perceptual evidence are true, that those which are disconfirmed are false, and, all things being equal, that we have no good epistemic reason to form any expectation of the truth values of sentences which a r e neither confirmed nor disconfirmed. T h e second idea is that it is reasonable to do what has the best chance of furthering our aims, and that whether what we do does further them depends, in large part, upon how things stand in the world. It is the state of my motor, and what goes on at the garage which determines whether my taking my car there will solve my problems with it. And it is the goings on in the brains of Parkinson's disease patients which determines what effect L-dopa therapy will have on their neurological problems. According to traditional epistemology, the world in which we act and upon which the success of our actions depends is as it is said to be by true sentences and is not as it is said to be by false ones. T h a t is why beliefs expressed by true sentences are better guides to action than those expressed by false sentences. Thus, if well confirmed sentences tend to be true, and their confirmation is a good
Synthese 64 (1985) 195-224. 0039-7857/85.10 © 1985 by D. Reidel Publishing Company
196
JAMES BOGEN
reason to think they are true, that explains why it is reasonable to act on beliefs expressed by sentences which are well confirmed by perceptual evidence. In telling this story, the traditional epistemologist commits himself to: (TE 1) A realist conception of truth: What determines the truth value of a sentence subject to empirical testing is the state of a world of extralinguistic things whose behavior is determined by natural laws which are independent of the rules and practices which determine how we speak and how we evaluate evidence. Because the traditional epistemologist believes the success of an action depends upon the way the world is, he must reject coherence theories, and warranted assertability accounts of truth which analyze truth value solely in terms of evidence and standards for well warranting evidence. T h e rules and practices which determine which assertions are well-warranted by what sorts of evidence do not determine what things are in the world and how they stand - the extralinguistic circumstances upon which the success or failure of our conduct ultimately depends. Thus if the truth of a well-warranted sentence were nothing more than its being warranted, the traditional epistemologist could not explain why conduct based on it should further the agent's aims. Thus the thesis that it is at least logically possible for an ideally well confirmed sentence or theory to be false (the thesis which Putnam (1978, p. 125) used to characterize realism) provides a necessary condition for the correctness of the traditional epistemologist's view. Similar considerations apply to coherence theories of truth. T h e traditional position also assumes: (TE 2)
The no accident thesis: T h e tendency of well confirmed sentences to be true is due to and explainable in terms of connections between canons of evidence, testing procedures, the perceptual evidence used in confirmation, and the goings on in the world which determine the truth value of sentences. Thus it is no accident that well confirmed sentences tend to be true.
If there were no explanation for the fact that well confirmed sentences tend to be true, we should have to consider their truth a matter of luck or happy accident. Presumably, it would be unreasonable
TRADITIONAL
EPISTEMOLOGY
197
to depend heavily upon luck in deciding what to do, and so the traditional epistemologist must be able to show that there is a good reason why well confirmed sentences, though sometimes false, tend in general to be true. Otherwise, he would be unable to account for the rationality of acting upon well confirmed beliefs. The traditional view has been attacked, at least since the time of Pyrrho, by skeptical arguments which remain unrebutted. The reason traditional epistemology has been unable to refute one class of skeptical arguments (classified in the next section of this paper as 'non-Humean') lies, I think, in a further assumption. (TE 3) Because skeptics question the legitimacy of procedure by which empirically testable theories must be confirmed, no empirically testable parts of any scientific theory may be appealed to in answer to the skeptic. My own view is that while (TE 1) and (TE 2), along with the traditional strategy for explaining the rationality of acting upon well confirmed beliefs, are highly plausible, and fundamentally correct, (TE 3) is both mistaken, and tangential to the traditional view. To show this I will examine tWO representative non-Humean skeptical positions, and argue that naturalism - by which I mean the rejection of (TE 3) and an appeal to empirically testable theories of perception - provides an adequate reply to them which preserves (TE 1) and (TE 2). Contrary to the fears which probably motivated (TE 3), I will try to show (in section 7) that naturalism begs none of the questions raised by non-Humean arguments. Furthermore (section 9), contrary to a suggestion derived from Carnap, naturalistic replies to skepticism are consistent with the realism which is essential to the traditional view. The concluding section of this paper sketches a Humean argument, and explains why naturalism seems inapplicable to it. . For our purpose, skeptical arguments should be classified according to the stance they take with regard to what I shall call 'theories of evidence'. Skeptics I class as non-Humean grant their traditional opponent's theory of evidence, while Humeans argue against it. A theory of evidence is required for the confirmation or disconfirmation of any belief expressed by an empirically testable sen-
198
JAMES
BOGEN
tence, s, such that both s and its negation are logically consistent with the observation sentences constituting the strongest evidence that can be obtained for or against its truth. Beliefs about the existence, the behavior, and the features of physical objects (including living plants and animals) their macroscopic and microscopic parts, and such highly theoretical items as the atoms, particles, etc., thought by science to explain their b e h a v i o r are of this kind; it is with such beliefs that traditional epistemology is chiefly concerned. What I mean by theories of evidence are bodies of sentences, rules, practices, etc., by appeal to which observation sentences are brought to bear upon the confirmation or disconfirmation of sentences which they neither entail nor contradict. A m o n g m u c h else which I shall not consider, an adequate theory of evidence must provide for at least two kinds of distinctions a m o n g observation sentences, z First, it must tell us how to determine whether an observation sentence, or batch of observation sentences is "reliable" evidence for the sentence to be tested. Roughly, an observation sentence, o, is reliable only relative to a sentence, s, to be tested, and is reliable only if its truth or falsity is indicative of the likelihood that s is true or false under the theory of evidence according to which s is tested. H e r e are some examples. W h e r e s is a sentence used to claim that a certain thing has a certain t e m p e r a t u r e and o reports a t h e r m o m e t e r reading, o is reliable evidence for or against s only if the t h e r m o m e t e r is in good working order, is properly applied, etc. If these conditions are not met, the truth of o (e.g., the truth of 'the t h e r m o m e t e r read 90 ° at time t') should be considered irrelevant to the confirmation or disconfirmation of s. Or, suppose that o reports what was seen by s o m e o n e looking through a microscope at a piece of stained tissue, and s makes a claim about a neuroanatomical structure, o is reliable evidence for s only if what was seen was not the result of an artifact of the staining or magnification procedure. A n d despite what my best friends tell me, I take it that accurate reports of the positions of heavenly bodies at the precise m o m e n t of my birth are unreliable with regard to claims about my psychology and personality because these have nothing to do with one another. In the sort of case often considered by traditional epistemologists, o is usually unreliable evidence for s if s is used to claim that a physical object of a certain kind was present in the p e r c e i v e r ' s immediate environment, and o reports what the perceiver merely d r e a m e d or hallucinated) Since unreliable observation sen-
TRADITIONAL
EPISTEMOLOGY
199
tences assign no degree of confirmation or disconfirmation to the sentence to be tested, an adequate theory of evidence must tell us what sorts of observation sentences are reliable (relative of course to the sentence to be tested). A second thing a theory of evidence must tell us is what degrees of confirmation or disconfirmation are provided by reliable observation sentences. For example, both a report of a barometer reading and a reading of a satellite photograph may count as reliable evidence for or against a claim about the weather. A theory of evidence should tell us which is to be weighted more heavily in case one tends to confirm and the other to disconfirm the weather claim. In cases where batches of evidence confirm (or disconfirm) a sentence, the theory should tell us which provides a stronger degree of confirmation. It may be (and probably is) the case that no precise metric applies to the confirmation of a great many empirically testable sentences. Thus, while the visual and tactile evidence I have right now for the presence of my electric typewriter provides much stronger confirmation for the claim that it is here before me than do observation sentences reporting the hum of its motor, there is no way and no need to specify just how many degrees stronger it is. But where no precise metric applies, an adequate theory of evidence should rank the claim as better or much better confirmed by the visual and tactile evidence, even though it can't say how many degrees better it is. As noted earlier, the skeptical arguments I have called "nonH u m e a n " do not question the correctness of any theory of evidence. Instead, they deny the applicability of theories of evidence to the assessment or explanation of the rationality of holding empirical beliefs. One of them, the "Abnormality A r g u m e n t " , whose best known representative is the dream argument of Descartes' Meditations (I and VI) argues that even if observation sentences, o, would provide high confirmation for a sentence s, if s were reliable, we can never tell whether the evidence or the theory of evidence under which it is thought to be reliable has been correctly applied in any case of empirical testing. Arguments of this sort are discussed in section 5. A second n o n - H u m e a n argument derives from Gettier's counterexamples against the "justified-true-belief" account of knowledge. What I shall call the "Gettier argument" grants the correctness of a theory of evidence, and unlike the Abnormality argument, it also grants its correct application to particular cases of confirmation. But, says the
200
JAMES BOGEN
Gettier skeptic, o need have nothing to do with what goes on in the world to secure the satisfaction of s's truth conditions, even if s is both true and well confirmed by o. If this is so, then contrary t o . t h e no accident thesis (TE 2) the truth of s is inexplicable and accidental as far as o and the theory of evidence go. This argument will be discussed in section 6. It is arguments of this sort which I think can be answered by a naturalistic appeal to empirically testable claims about perception and what we perceive, and historically, a good many skeptical arguments have been of this n o n - H u m e a n kind. 4 In contrast, H u m e a n arguments question, and demand justification for elements of a theory of evidence required to determine degree of confirmation. It is these (see section 10) for which I see no clear application for the naturalist strategy. This is regrettable; to show that naturalism is a panacea for all skeptical ills would be much nicer than to admit that H u m e a n arguments seem immune to it. But it is no reason to reject naturalism. T h e refutation of n o n - H u m e a n arguments is greatly to be desired, and if naturalism provides it, it should not be thrown out because there are other sorts of arguments it will not refute.
.
Here are the n o n - H u m e a n arguments I want to consider. Abnormality arguments against the possibility of distinguishing reliable from unreliable observation sentences have premisses of the forms: (la) (lb)
Observation sentences of kind K are unreliable according to any adequate theory of evidence, and T h e r e are no tests (admissible under adequate theories of evidence) by which observation sentences of kind K can be distinguished from observation sentences which are reliable.
Its conclusion is that: (lc)
T h e r e is (or can be) no acceptable theory of evidence under which we can have good reason to think any given observation sentence is reliable.
At first sight, it would appear that premisses of the form (lb) would be clearly false. In the examples mentioned earlier, for instance, an observation sentence reporting a thermometer reading is unreliable if it
TRADITIONAL
EPISTEMOLOGY
201
is of kind K where Ks are sentences obtained from reading defective thermometers. In another case, observation sentences were considered unreliable because ~hey reported what was seen because of staining or magnification artifacts. Presumably, there should be no great difficulty in devising a test for defective thermometers or misleading microscopic techniques, and if successful, such tests provide counterexamples to (lb). But (lb) is notoriously recalcitrant in connection with abnormal observation sentences, namely sentences reporting what is dreamed, hallucinated, etc. Where instances of K are abnormal observation sentences, we have what I shall call the argument from abnormality, for which Descartes' dream argument is an instance: Under any adequate theory of evidence: (Ala) (Alb)
(Alc)
Abnormal observation sentences (e.g., sentences reporting what was in fact dreamed or hallucinated) are unreliable. For~any feature, F, discernible to a perceiver, and thought to distinguish an abnormal perception (e.g., dream or hallucination experiences) from a perception whose report would constitute normal evidence, F itself could occur as part of a dream or a hallucination. Therefore, it is impossible to tell whether any observation sentence is abnormal or normal, and therefore, it is impossible to tell whether any observation sentence is reliable or unreliable.
If this argument is sound, traditional epistemology is not just mistaken; it is misguided and irrelevant. A theory which purports to explain why sentences confirmed by reliable perceptual evidence are likely to be true would be of academic interest at best if the Abnormality argument is sound and there is no way to tell the difference between unreliable and reliable evidence. Knowing why beliefs expressed by well confirmed sentences are good guides to conduct provides cold comfort to the traditional epistemologist if the Abnormality argument is correct, and we cannot tell which sentences are well confirmed. Gettier (1963) devised cases that were intended to be counterexamples to the thesis that justified true belief is necessary and sufficient for knowledge. The following is typical. If observation sentences, o, provide strong confirmation for a sentence, sl (e.g., 'Jones is in France'), then, together with the law that if a
202
JAMES BOGEN
sentence is true its disjunction with any other sentence is true, o provides excellent evidence for the sentence ( s l v s2), where s2 is a sentence (e.g., 'Thelonius is in Mexico') for which we have no evidence whatsoever. T h e disjunction, sl v s2 turns out to be true, but only because s2 is true; sl, the sentence o confirmed, turns out to be false. This, and the intricate and ingenious examples in the post-Gettier literature were thought to show that what epistemology needs is a new condition which, together with justified true belief, would be sufficient and necessary to distinguish knowledge from justified true belief. For reasons articulated by Austin (1946) I think this is the wrong moral to draw from the Gettier examples. Austin suggests that a person knows that s just in case he can m a k e a knowledge claim which should be accepted. W h e t h e r s o m e o n e ' s claim to know that s should be accepted depends upon whether or not he is in a position to rebut challenges to his claim to know. Just which challenges must be rebutted depend upon the sort of thing he claims to know. Thus the reasons for challenging and rejecting s o m e o n e ' s claim to know that the bird on yonder tree is a goldfinch differ greatly f r o m the reasons for challenging and rejecting the claim to know that there is one proton in the hydrogen atom, that a single man wrote the works of H o m e r , that a line in Sappho was miscopied by an early scribe, that cigarette smoking causes cancer, that the ' H o t Pecans' play Will Bradley tunes, etc., even though all of these claims are empirically testable. Furthermore, just which challenges may arise in connection with a particular claim to know depends upon the situation in which the claim is made. Suppose s o m e o n e claimed to know there is a diamond in front of him. All things being equal, the challenge 'how do you know that's not a laser h o l o g r a p h ? ' needs rebuttal if the knowledge claim is made in a m u s e u m which uses holograph equipment, but not on the site of a mineral deposit in the wilds of Africa. In view of this, I see no reason to think any single set of conditions which are at all informative can define knowledge, or constitute necessary and sufficient conditions for all empirical knowledge claims. But the question 'what is knowledge?' is plausibly construed as requiring an account of successful knowledge claims. Accordingly, I am inclined to think of knowledge as a context-relative notion of far less interest to the problems of traditional epistemology than recent writers have thought it to be. If this is so, the search for necessary and sufficient conditions for knowledge deserves abandonment. W h a t the Gettier examples do show, however, is that in at least some
TRADITIONAL
EPISTEMOLOGY
203
cases, well confirmed sentences can be true for reasons their confirming evidence would never lead us to expect. This raises the horrible possibility that for all we can learn from the observation sentences which confirm a sentence and the theory of evidence under which it is confirmed, the truth of the confirmed sentence can be a matter of happy accident. T w o conclusions might be drawn from this. T h e first is that contrary to what the no accident thesis (TE 2) suggests, it is not rational to believe what is said by an empirically well confirmed sentence. After all, if the truth of such a sentence may, for all we know, be due to sheer luck, it should be no more reasonable to believe and act on what is empirically well confirmed than to rely on luck to go our way. This is analogous to the idea that because there is no earthly reason why astrological predictions should turn out to be true, it would be unreasonable to act on the predictions of an astrologer, even if the astrologer were usually right. Alternatively, we might say that because well confirmed sentences usually do turn out to be true, it would be folly not to believe them and act upon what they say. If so, the moral to be drawn f r o m the Gettier examples would be that although it is reasonable to believe what is said by empirically well confirmed sentences, it is a mystery why they should turn out to be true. (Compare: there is no explanation for the truth of the astrologer's predictions, but since she's usually been right, it would be foolish to ignore them.) T o accept this conclusion is to abandon the p r o g r a m of traditional epistemology whose aim requires an explanation of the tendency of well confirmed sentences to be true. . Considerations like the two n o n - H u m e a n arguments just presented suggest drastic measures, one of which is the a b a n d o n m e n t of traditional epistemology. T h o s e who accept the Abnormality a r g u m e n t may retain (TE 1), holding that the state of the world determines the truth values of our beliefs but that it is never epistemically reasonable to think we h a v e good evidence that any sentences (except for observation sentences and their logical consequences) are true or false. I shall call this view "Pyrrhonism". T h e Gettier argument suggests a position which I shall call " P r o t a g o r e a n i s m " . 5 Protagoreans reject ( T 1 ) and hold m o r e or less complicated versions of the view that a sentence is true only relative to a theory, and that relative to a theory, T, a testable
204
JAMES
BOGEN
sentence is true just in case its assertion is well warranted according to T's standards of confirmation. T h e Protagorean is able to say we can often tell whether a sentence is well confirmed, and he has no difficulty in explaining why well confirmed sentencess are apt to be true. But he rejects the traditional account of the rationality of accepting and acting upon well confirmed beliefs. If truth is well-warranted assertability, and is therefore theory relative and determined solely by observation sentences and a theory of evidence, there is no room for the traditional version of the thesis that conduct based upon true beliefs is apt to succeed because the world tends to be as well confirmed sentences say it is. T h a t is because the traditional account assumed not only that well confirmed sentences tend to be true, but also that truth is a function of the state of the world in which we rely on our beliefs to guide our conduct. If truth is a function of evidence and a theory of evidence, and nothing more, the fact that a sentence is well confirmed can lead us to expect certain things of the evidence and the theory of evidence, but it supports no conclusions about the state of the world in which the agent acts. Different as they obviously are, Protagoreanism, Pyrrhonism, and traditional epistemology are all nonnaturalist because they share the contention (TE 3) - the idea that because they require empirical testing, the sciences are rendered problematical by skeptical arguments, and therefore, that they are of no interest for epistemology except insofar as their confirmation procedures count as data for which the epistemologist should try to account. (See Price, 1961, p. 2.) T h e traditional epistemologist sees his task as vindicating the sciences by showing that their confirmation procedures are really guides to truth. But, he thinks, until they are vindicated it would beg the question to appeal to scientific theories in answering skeptical arguments. T h e Pyrrhonian believes skeptical arguments show we can never establish the soundness of any argument based on empirically testable premisses. And for the Protagorean, the realistic conception of truth upon which traditional epistemology depends is false, and cannot be established by any arguments, let alone scientific ones. Let us see what happens if we reject (TE 3) and treat n o n - H u m e a n skepticism naturalistically. . In an earlier paper. Morton Beckner and I (1979) argued that if there were a comprehensive and well confirmed account of perception which
TRADITIONAL
EPISTEMOLOGY
205
showed that the Abnormality skeptic's premise (Alb) that there is no discernible feature, F, by which normality can be determined is either false or of only negligible probability, we should be free to reject the Abnormality argument as unsound. The only reason the skeptic gives us to believe (Abl) is that any discernible mark of normal (nonhallucinatory, etc.), perceptions would itself have to be detected through perception. This is certainly true, but the skeptic supposes that any feature of a single perceptual experience could itself be dreamed or hallucinated, and that if additional perceptions were required for the detection of F, they too could be abnormal. If, contrary to this, F could not be abnormally experienced there would be no reason to accept the Abnormality argument. What the Abnormality skeptic says is logically possible, but all this means is that what he says can be expressed in meaningful sentences whose denials are not required by logical, syntactical, or semantical rules. There are many such sentences (e.g., "Bach is the composer of 'The House of Blue Lights'", 'H. H. Price is an existentialist theologian', 'Ultra violet light can be detected by the unaided human eye under normal conditions', 'there are espresso bars on the moons of Jupiter', etc.) which are clearly false or too improbable to worry about. Beckner and I contended that (Alb) and the premisses used by the skeptic to support it are among them. Suppose the millenium came and there were a well confirmed and highly comprehensive, detailed theory, T, about what sorts of physical objects and systems of objects in the perceptual environment cause normal perceptions. Suppose T told us how perceptual experiences are produced in the nervous system as a result of light, sound, and other stimuli emitted by objects in the perceptual environment, and set out physical, biological, and psychological laws which govern the behavior of the physical things which produce perceptual stimuli. Perhaps this is not too much to hope for; a good deal is known about such things now. Suppose T also included a theory of what must go on in the nervous system in order for dreams, hallucinations, and other abnormal experiences to occur, and the laws which govern their causes. Although there is no such theory now, what we do know suggests that if T were fully developed and well confirmed, it would establish that dreams, hallucinations, and other abnormal experiences are caused by internal happenings in the CNS whose occurrences are random with regard to crucial regularities which characterize the production of normal experiences. Consider, for
206
JAMES
BOGEN
example, the normal perception of a raccoon in bright, full moonlight sneaking through an oak grove in the San Gabriel mountain foothills toward the chicken coop, prying apart a seam in the chicken wire, and squeezing through it, killing a chicken, eating it, and slinking away. The reasons why these events are perceived in this order, and why the raccoon, the chickens, the oak trees, etc., appear as they do are to be found in the physical, psychological, and biological laws which govern the behavior of these things, their emission of stimuli to which the observer's perceptual systems are sensitive, and the capabilities of the CNS to respond and process the effects of the stimuli. According to the best theories we have now (and presumably, according to T) it would be nomologically impossible for this system of objects to cause perceptions under normal conditions, e,g., of the raccoon disappearing in a cloud of magenta smoke, levitating, turning into a cantaloupe, crowing like a rooster, smelling like a rose, suddenly occupying positions at great distances from the one in which it appeared to be at a given time, etc. It would be impossible for a raccoon to produce sights but no sounds, for the objects just mentioned to be visible or audible to one, but not to any other well located normal observer, etc. These are the sort of regularities relative to which hallucinations, dreams, and other abnormal experiences are produced at random. According to T, it would be impossible, or nearly so for the internal disturbances which cause abnormN experiences to cause a sequence of experiences which is as complicated as this and which also exhibits the regularities to which a normally caused experience of the relevant objects would be subject. It would also be impossible (or nearly so) for a sequence of internally caused perceptions to fit what the observer knows or has good reason to believe had been going on before the experience, and what he observes going on after it. Accordingly, the chance that sufficiently detailed sentences reporting detailed observations of the raccoon's progress to the coop and his attack on the chickens are unreliable because the perceptions were abnormal is negligible at best. To supply an F, the distinguishing feature the skeptic claims we can't find, we need only to characterize the regularities for various sorts of perceptual experiences which, according to the theory of perception, internally caused experiences should fail to exhibit. To rebut the skeptic's claim that such regularities could be features of abnormal experiences, all that would be needed is strong confirmation of T. If it turned out that no theory of perception could give us these
TRADITIONAL
EPISTEMOLOGY
207
results, we should have to concede that ( A l a ) could be true. But in view of what is already known of the relevant areas of physics, biology, and psychology, it would be astonishing if a well confirmed, comprehensive theory of perception did not lead to the anti-skeptical results Beckner and I hoped for. A n d in any case, what I want to argue is not that there is now a theory like T, but that an appeal to T would (if we had T and it were well confirmed) provide a legitimate refutation of the Abnormality argument. . If the Gettier problem is understood as I argued it should be (in section 3) - as an objection to (TE 2), the no accident thesis - the confirmation of T would solve it as well as answering the Abnormality argument. In fact, it would show that a solution is already available in the works of Goldman, Swain, and others who have suggested causal accounts of knowledge in response to Gettier. In an early paper, G o l d m a n (1967) suggested that good evidence for a true belief is often causally connected to what makes the belief true in such a way that relative to the evidence, the truth of the belief is clearly no accident. H e r e is a very simple example of what G o l d m a n noticed. T h e sentence which claims that the leaves on the Eucalyptus tree outside my office are green is well confirmed by reports of normal observers who see green leaves when they look at the tree. W h a t makes the sentence 'the leaves are green' true is the pigment of the leaves, the very thing which causes normal observers to have the perceptions which the confirming observation sentences report. H e r e the truth of a well confirmed sentence is no mere piece of luck; it is well explained by its truth and its confirming evidence having the same cause. Thus there is no reason to think that believing and acting upon the sentence would be unreasonable in the way the Gettier examples suggest it might be. Of course, not all cases are this simple. Nor do they all involve the same causal pattern. But G o l d m a n (1967) and Swain (1978) have shown that in a great m a n y cases, there are causal connections between confirming evidence and what makes the confirmed sentences true. Intricate and involved as some of these connections are, they provide strong support for the no-accident thesis. In addition, the recent literature suggests plausible arguments for the view that natural selection has attuned as well enough to the normal perceptual environment to m a k e it unlikely
208
JAMES BOGEN
that a great many perceptually well confirmed beliefs could be true only by accident. Goldman (1976) argues that for recurring features of the environment of the kinds we need to be able to discriminate, our perceptual systems are biologically such that under favorable conditions, different features of the environment are bound to produce different perceptions, which in turn are bound to produce " . . . suitably different beliefs about the environment". Sober (1980) makes an analogous claim about the reliability of certain of the cognitive faculties which figure crucially in perceptual confirmation and the theoretical use of confirmed sentences. 6 Both suggest, quite plausibly, that our species would not have survived the process of natural selection unless such mechanisms tended to produce true beliefs under favorable conditions. These may be considered further plausibility arguments both for the contention that things should turn out to be as I suppose T will claim them to be, and for the claim that the causes of perceptions are closely enough connected with the causes of truth of what they confirm to make the occurrence of well confirmed sentences which are only accidentally true the exception rather than the rule. I do not wish to claim that any of this provides a solution to the Gettier Problem conceived, as it originally was, as requiring a universal but informative enumeration of necessary and sufficient conditions for knowledge. As a problem for the traditional epistemologist who wants to explain the rationality of acting according to well confirmed beliefs, the Gettier argument requires instead a defense of the no-accident theses, and this does not require an enumeration of necessary and sufficient conditions for knowledge. What I want to claim is that Goldman, Sober, Swain, and other causal theorists of knowledge have shown us where a defense of T E 2 is to be f o u n d - in causal explanations for the truth of well confirmed sentences of the kind which a theory like T would provide. . Could T (if only we had it) be confirmed without begging n o n - H u m e a n skeptical questions? Part of T would be a body of a priori sentences of mathematics, logic, etc., which would be included in or presupposed by a systematically set out version of the theory. We can ignore whatever doubts there are about theses for they are not raised by n o n - H u m e a n skeptics. In addition to the a priori component, T would include procedures, practices, etc., for mobilizing, collecting, and assessing
TRADITIONAL
EPISTEMOLOGY
209
evidence, applying the theory of prediction and explanation, etc. These, if codified at all, would occur in a systematized version of T as rules which, like imperatives, are neither true nor false, confirmable or disconfirmable. We can also ignore these for now. For our purposes, the problematical parts of T would be sentences subj'ect to empirical testing, including formulations of theories of perception, biology, psychology, and physics upon which the naturalistic refutation of the Abnormality a r g u m e n t depends, as do the causal claims required to defend (TE 2), the no accident thesis. F r o m here on, except where noted or made obvious by context, ' T ' will be used to refer only to T ' s empirically testable sentences. I think the skeptic must grant that if we can tell that T is well confirmed, we are in a position to refute the Abnormality argument, and answer the Gettier question. But this gives rise to an objection from what I shall call "the Cartesian constraint": no empirically testable sentence should be accepted as part of a theory or used as a reason for accepting any part of a theory unless it is either a reliable observation sentence or is well confirmed by reliable observation sentences or sentences they confirm. 6 Thus an observation sentence is admissible only if it accurately reports what was perceived, and there is good reason to think the perception was not abnormal. A nonobservational, but testable sentence is acceptable only if it is well confirmed by admissible observation sentences or by other sentences which are well confirmed by perceptual evidence. Perhaps this constraint is too strong, but in order to do justice to the skeptic, I shall grant it for the sake of argument. Now, the objection from the Cartesian constraint is that in order to refute the Abnormality argument, the naturalist must show that certain observation sentences are not abnormal, and this requires an appeal to T. But according to the Cartesian constraint we cannot appeal to any sentences in T unless they are themselves reliable observation sentences, are well confirmed by reliable observation sentences, or are well confirmed by other well confirmed sentences. We have no good reason to think this of any sentence unless T is well confirmed. Thus, the objection continues, unless he begs the question by simply assuming that T is substantially correct and arguing from it that the observation sentences whose reliability he needs to establish are normal, the naturalist must violate the Cartesian constraint by using questionable observation sentences to show that T is well confirmed.
210
JAMES
BOGEN
Part of this argument is certainly correct. T h e naturalist must be able to show that T is well confirmed, and in order to show this, some of T's own sentences must be appealed to as reasons for thinking the confirming evidence is normal. T h e r e is undoubtedly a circularity of some kind in this. However, I do not think the circularity is vicious, or that it constitutes a genuine violation of the Cartesian constraint. T o see whether this is so, we shall have to consider what confirmation of T might look like, and how the possibility of abnormal observation sentences bears upon its confirmation. As is generally known, the bearing of one sentence, e, upon the confirmation or disconfirmation of a second sentence or group of sentences, s, typically depends upon further sentences, b, which I shall call "bridges" or "bridge sentences". Whether or not e confirms or disconfirms, or provides a good reason for believing or disbelieving s typically depends upon whether bridges, b, are true. For example the confirmation of a claim about temperature by sentences reporting mercury thermometer readings, depends on (b) a sentence or sentences establishing the relevance of the height of a column of mercury to the determination of the temperature of what is being measured. Clark Glymour argues persuasively that what I am calling "bridges" may, and often must be, empirically testable sentences which belong to the same theory as the sentences for whose confirmation they are required. 16 Instances of Glymour's point (in addition to those he offers in Theory and Evidence) are to be found in neuroanatomy, an important c o m p o n e n t of T. Without bridges belonging to a theory of the anatomy and physiology of nerve cells and the nervous system, it is impossible to determine the significance of some of the microscopic and clinical evidence required for the confirmation of further anatomical and physiological claims. More importantly, if features of what was observed and the conditions under which it was observed (e.g., the state of the observer and the lighting conditions, etc., and features of the perceptual experience he reports) are to indicate normality, there must be bridges to establish their relevance to normality. T h e confirmation of T may be expected to follow this much of Glymour's model and in view of the use which I would want to put T, there is nowhere else to look for the bridges required for the application of perceptual evidence to the confirmation of T than T's' own testable sentences. This is the source of the circularity in the naturalist's argument against the skeptic.
TRADITIONAL
EPISTEMOLOGY
211
With this in mind, I shall sketch a partial inventory o f kinds o f sentences which would be relevant to the confirmation of T and to the skeptic's claim that its confirmation must violate the Cartesian constraint. Along with m a n y others which I shall not consider, T may be expected to contain Sl, a sentence used to claim that abnormal observation sentences are unreliable, el, sentences (at least some of which are observation sentences) used to confirm Sl, and b~, bridge sentences required for the confirmation of sl by e~. In addition, T should include sz, a sentence which claims that positive results (call them " R + " ) of certain tests are good grounds for taking an observation sentence to be normal, and that negative results ( R - ) are good reasons for thinking it unreliable because abnormal e2 observation and other sentences used to confirm s2, and b2, bridges required for its confirmation, and s3, the testable sentences of T ' s perceptual, biological, etc., components. Included a m o n g these will be m e m b e r s of bl and b2, along with relatively nonobservational m e m b e r s of el and e2. Finally, T should include
b3, bridges required for the confirmation of s3, and e3, evidence used to confirm b3. Sentences are classed under these headings according to the use they are put. Because sentences may have m o r e than one use, a given sentence may fall under m o r e than one heading, and as a result, there will be an overlap between their memberships. Thus a sentence which belongs to the perceptual theory (and thus falls under heading s3) may itself be used to confirm a sentence falling under other headings; so used, it will belong to one of the ' e ' groups or ' b ' groups, as well as to one of the ' s ' groups. This is particularly to be expected of sentences used by T to explain how abnormal perceptions occur. For example, if T is any good at all, it should establish that, predict when, and explain why a cantaloupe can be hallucinated when no cantaloupe is present. This result, and sentences supporting it, belong to s3 as m e m b e r s of T ' s perceptual theory. But such sentences are also required for the
212
JAMES
BOGEN
confirmation of sl, for they will provide the best reasons for thinking that an observation sentence reporting the visual hallucination of a cantaloupe is not reliable evidence for the presence of a cantaloupe. As applied in the confirmation of sl, they will belong to el or bl. Now we can see a bit m o r e precisely what kind of circularity is involved in trying to convince a skeptic that the appeal to T does not violate the Cartesian constraint. A r g u m e n t s against the Abnormality and the Gettier arguments require sentences from s3, (e.g., the perceptual theory). T o show that an s3 sentence meets the constraint, we must show that it is well confirmed. This is to be established by appeal to e3, and to b3 - sentences which secure the application of the evidence to s3. T h e sentences f r o m e3 used for this purpose must be normal, and to establish their normality, we test for R + . But R + results are inconclusive unless s2 is well confirmed, and its confirmation will require sentences (not, it is to be hoped, the very one we started out originally to test) from s3. T h e first point to be noted concerning this is that (assuming there is confirmation for the required bridges, and that the evidence we get is what they require for confirmation of s group sentences) the circularity of the confirmation p r o c e d u r e guarantees the satisfaction of the C a r tesian constraint. If T has the proper structure and content, then unless things go badly and the observations disconfirm what we hoped to confirm, we can start anywhere in the circle, and follow it around until we c o m e to the sentences required for confirmation or for the application of observation sentences to confirmation of the sentence we started out to test. Thus we needn't worry about whether the constraint can be m e t if the evidence turns out to be what is required for confirmation. Instead, we need to worry about whether the satisfaction of the constraint is nontrivial and nonquestion-begging. T h e confirmation of T would be trivial if, e.g., one sentence (along with its logical consequences) were allowed to confirm itself, or if two or m o r e sentences confirmed each other in such a way that it m a d e no difference what observation sentences resulted from any attempt to test them. T h e question whether or not T ' s confirmation would turn out to be trivial in this way cannot be answered a priori, or before T is well developed and systematically set out. It depends upon the structure of the theory, the data to be treated by it, and upon what kinds of tests can be devised to test it. But this is no c o m f o r t to the skeptic, for there is nothing in the Cartesian constraint, the Gettier or the Abnormality
T RAD~tTJ O NAL
EPISTEMOLOGY
213
argument, or the fact that the confirmation must be circular in the way I have indicated which show that reasonably strong standards for nontriviality could not be met in confirming T. 17 T h e question whether T's confirmation must beg questions against the skeptic may be answered by considering the fact that nothing is required for its confirmation which is not also required for the application of the Cartesian constraint, or for raising the questions posed by the Abnormality argument. T h e use of the Cartesian constraint to show that the naturalistic appeal to T against the abnormality argument begs the question depends upon the claim (sl) that observation sentences which report abnormal perceptions are unreliable. The use of T to rebut the Abnormality argument is perfectly in order unless there is reason to doubt that T is well confirmed, and the possibility that observation sentences used to confirm it are abnormal is no reason to doubt T's confirmation unless sl is true. Furthermore, the Cartesian constraint forbids appeal to Sl by the skeptic unless there is good reason to think sl is well confirmed. Because Sl is neither an a priori truth nor an observation sentence, its confirmation requires observation sentences. Because no observation sentences logically entail sl, its confirmation requires bridge sentences as well. What makes the confirmation of T appear to beg questions is that the confirmation of s2 (the claim that abnormality can be tested for) depends upon the confirmation of s3. But so does the confirmation of sl, (the claim that abnormal evidence is unreliable) and the assessment of the skeptic's claim that undetectable abnormality is possible. Without a theory of perception, there is no reason
to think abnormal perceptions of the kind the skeptic argues from are possible, and thus, unless s3 is confirmed, the skeptic's appeal to the possibility of undetectable abnormalities violates the Cartesian constraint. The constraint is just as much violated by claiming that abnormal evidence is unreliable without confirmation for that claim. Its confirmation involves showing that an abnormal perception of a physical thing is not indicative of that thing's presence; this in turn, requires bridge and theoretical sentences from s3, whose perceptual theory explains what abnormal perceptions are, and how they are produced, and thus why abnormal perceptions of a physical object need not be indicative of that object's presence, features, etc. If it were impossible to confirm s3 without question begging, the Cartesian constraint would prevent the skeptic from asking his question, just as much as it would prevent the
214
JAMES
BOGEN
naturalistic epistemologist from answering it, and illegitimate questions in this subject require no answers. If the skeptic could provide an argument whose conformity with the Cartesian constraint did not depend upon the confirmation of T, it would clearly beg the question to appeal to components of T to confirm its other components, and to use the results of this procedure as arguments against the skeptic. But there is no good reason for thinking abnormal observation sentences are really possible or that they are suspect for purposes of confirmation unless T's perceptual theory is well confirmed. In short, if the confirmation of T is disallowed, there is no skeptical question to beg. It might b e objected that confirming T without violating the Cartesian constraint is an impossible task. Suppose we start with sentences from s3, whose confirmation requires an observation sentence, o. In order to use o, the objection goes, we must already have shown that it is normal. But according to this objection, we can't possibly have shown it to be normal because the test for normality presupposes s3, whose sentences we are now trying to test. So we have a vicious cycle: in order to confirm sentences at any given time, t, we must have previously confirmed sentences whose confirmation depends upon the sentences to be tested at t. This objection requireg us to understand the Cartesian constraint as imposing a requirement of temporal order upon the confirmation procedure. So read, the constraint requires that where s is any testable sentence required for confirming a second sentence, s', (C1)
s must be confirmed (temporally)
before it is used to confirm
S r"
This version of the Cartesian constraint cannot be satisfied by the confirmation of T. Indeed, it is hard to think of any interesting empirically testable theories which could satisfy it. T h e confirmation of most if not all empirically testable theories of any explanatory power requires the use of bridges which are themselves subject to empirical testing. C1 is initially implausible, and should not be invoked Without convincing arguments in support of it; I do not know what those arguments would be. On the other hand, the Cartesian constraint is plausible if interpreted nontemporally as requiring merely that (72 confirmation must be provided for any testable sentence used in the confirmation of other sentences.
TRADITIONAL
EPISTEMOLOGY
215
It is in keeping with C2 to think of confirmation as something undertaken once a theory has been set out in finished form, along with the procedures for testing it and the observation sentences obtained from its testing. F o r a theory set out in complete and canonical form, (72 requires confirmation for its testable sentences, including those which serve as bridges, but it does not require that this be done in any particular temporal order, and it does not require that a sentence be confirmed at some time before it is placed into the body of sentences which constitute the canonical version of the theory to be tested. A sentence, s, need not be confirmed (temporally) before it is used to confirm another sentence, s'. All that is required is that confirmation can at some point be provided for s'. Thus there is no a priori reason to think T could not be confirmed according to (72. If C1 sounds plausible, I suspect that is because of the supposition that confirmation is a necessary part of the discovery or development of a theory, and in particular, that in developing a theory no sentence is to be used until after it has been confirmed. But the history of science argues that in real cases, crucial steps are taken in the development, elaboration, and provisional testing of a theory before confirmation is possible for the steps they depend upon, and that in many cases the need for confirmation through certain kinds of tests does not arise until a theory has been well developed. Unless Baconian methodology undergoes a revival, there is little in the philosophy of science to r e c o m m e n d Ca as a requirement for theory development or testing. Thus the confirmation of T is impossible only under C~, and this is the most implausible interpretation of the Cartesian constraint. Finally, it might seem troubling to think that unless s3 is true, the use of s2 to test for abnormality is highly questionable. It might be objected that the truth of a great deal of T is presupposed by the legitimacy of its confirmation, and thus that a skeptic who doubted T should doubt its confirmation. In a way, this is correct. Regardless of how convincing the evidence for s2 seems to be, positive (R+) results of tests for normality should not be good evidence for normality of s3 is false. But this is just to say that the confirmation of T should be unsuccessful if the theory contains enough false sentences, and that is as it should be; we should expect it to be very difficult to confirm false theories, and much easier to confirm true ones. T o ask for assurance that a theory is true before the sentences which confirm it are accepted as good evidence would be like arguing that because the premisses of a valid deductive
216
JAMES
BOGEN
argument are false if its conclusion is false, we should never accept a deductive argument without an independent proof of its conclusion. . T h e program for a naturalistic reply to the Abnormality and the Gettier arguments differs from traditional epistemology in two respects which should be borne in mind. First, I do not suppose that observation sentences are limited to reports of sensations, sense data, or any other versions of L o c k e a n or Berkleyan "ideas". Indeed, I see no reason why an epistemologist should believe there are such things to be reported. Second, and more importantly, I do not assume, as did Descartes, Berkeley, C. I. Lewis, Price, and many others, that epistemology is a nonempirical discipline, or that only a priori truths are admissible as premisses for arguments against skepticism. Quite the contrary, I think that nothing less than an empirically well confirmed theory can provide answers to the skeptical arguments I have been considering. But despite this, naturalism (as I think of it) and traditional epistemology both hold that the crucial difference between beliefs which do, and beliefs which do not further our goals is to be explained in terms of the truth values of the sentences which express them. Both views are committed to (TE 1) and (TE 2). And both are committed to a realist conception of truth. Naturalist and traditional theories hold that it is reasonable to believe and act on what is well confirmed by perceptual evidence because perceptually well confirmed sentences tend to be true, and this is taken to mean that things in the world tend to be as well confirmed sentences say they are. Thus even if the strategy of section (7) provides answers to the Gettier and the Abnormality argument, it would be a serious objection to show that these results do not support realism. I want now to consider an objection of this kind, the claim that in order to reply to skeptical arguments considered so far in the way I have been suggesting one must adopt a position which is better understood as Pyrrhonist or Protagorean, than as a realist view. T h e source of this sort of objection is Carnap's paper, "Empiricism, Semantics, and Ontology" (1956). Suppose T can be confirmed without question begging triviality, or violation of the Cartesian constraint. Suppose also that T is powerful enough to show that we often have what can easily be recognized as reliable evidence for all sorts of
TRADITIONAL
EPISTEMOLOGY
217
sentences which claim or suppose that specific sorts of physical objects exist, and that contrary to the Gettier skeptic, these sentences are not true just by happy accident. Even so, according to the objection, the sentences which claim physical objects exist are nothing more than answers to what Carnap called "internal" or "external" existence questions. If this is so, naturalism parts company with traditional epistemology on the question of realism (TE 1), for Carnapian internal and external existence claims are in effect nothing more than claims about the usefulness of the language of a theory, or about that theory's standards of warrantability.
.
Carnap's characterization of internal and external existence claims applies most clearly, and is most easily understood in connection with mathematics, logic, and other systems with effective proof procedures. For example, consider the sentences: (1)
there is a prime number greater than 100, and
(2)
there are numbers.
Carnap would say these sentences answer questions about languages, and are true or false only in languages containing numbers, certain predicates ('odd', 'even', 'prime', 'greater than', etc.), a general term, 'number', variables for which numerals can be substituted, along with the rules of arithmetic, and deductive logic. Where L is such a language, the questions 'is there a prime number greater than 100T and 'are there numbers?' are to be analyzed as asking no more than whether sentences (1) and (2) are provable in L. So understood (1) and (2) do not claim that numbers, or a prime number greater than 100 are to be found among the furniture of the universe. Instead, they reflect a fact about language L, namely that (1) and (2) are well formed and provable in L. So analyzed, the question answered by (1) and (2) are called 'internal existence questions', and I will call their answers 'internal existence sentences'. 8 Alternatively, the questions 'are there numbers?', etc., can be read as asking whether we should speak a language in which (1) and (2) are well formed and provable. Here, what look like ontological claims are really claims about the advisability of using a language of a certain kind for
218
JAMES BOGEN
whatever theoretical or practical purposes are pragmatically relevant. T h e purposes for which the language is to be used . . . determine which factors are relevant for the decision. The efficiency, fruitfulness, and simplicity of t h e . . , language may be among the decisive factors. (Carnap, 1956, p. 208)
T h e world's containing or not containing entities called numbers is not among the decisive factors (unless 'there are numbers' is used as a claim about what is provable in L, or a reflection of L's pragmatic virtues). Carnap (1956, p. 207) says something like the same thing about the question whether there really are physical objects. Suppose L is a 'thing language', a language containing such terms as 'table', 'cantaloupe', 'raccoon', and other physical object words. Once we have accepted the thing l a n g u a g e . . , we can raise and answer internal questions, e.g., 'is there a piece of paper on my desk?', 'Did King Arthur actually l i v e ? ' . . , and the like. These questions are to be answered by empirical investigation. Results of observations are evaluated according to certain rules as confirming or disconfirming evidence for possible answers . . . . The concept of reality occurring in these internal questions is an empirical, scientific, non-metaphysical concept.
Carnap does not say explicitly what this proposal amounts to, but he seems and is usually taken to mean that such questions as 'is there a piece of paper on my desk?' are to be understood as asking with regard to a language containing terms like 'paper', whether such sentences as (3)
there is a piece of paper on Carnap's desk
are well formed and confirmable. If the language in question contains variables for which physical object words can be substituted, and the general term 'physical object', the sentence (4)
there are physical objects
construed as an internal claim, follows from the confirmation of (3), just as (2) can be deduced from the proof of (1). Alternatively, just as 'there are numbers' (if it is not read as the existential generalization of (1)) can be construed as an external sentence, so, Carnap (1956, pp. 207-208) appears to think, (4) can be read as an external sentence having to do with the advisability of using a language which includes physical object words.
TRADITIONAL
EPISTEMOLOGY
219
If someone decides to accept the thing language, there is no objection against saying that he has accepted the world of things. But this must not be interpreted as if it meant his acceptance of a belief in the reality of the thing world. T o ' a c c e p t the thing world' is merely to accept a language in which sentences like (3) are well formed and confirmable. T o do that is to decide on pragmatic grounds to use a language; it is not to m a k e the ontological c o m m i t m e n t required by the traditional epistemologist. Let L be the language of T, the theory used to reply to the skeptic, and let R be a well formed sentence of the form 'there is a raccoon at place p at time t'. L e t 'R~' name the sentence which gives the internal existence version of the claim made in L by R. R~ will be true if (5)
R is well confirmed by observation sentences in L.
Similarly, the sentence which gives the internal existence claim interpretation of (6)
there are raccoons
will be true if (7)
R is well confirmed by observation sentences in L and R ' s
existential generalization follows from R in L. Finally, the external claim versions of R ('there is a raccoon at p at t') and 'there are raccoons' reflect the decision to use the language L. It is easy to see that if instances of 'there is a raccoon at p at t' and if 'there are raccoons' had to be internal or external sentences, the position I have taken in reply to the skeptic would be compatible with the denial of (TE 1). T h e truth of R merely warrants speakers of L to say or write R and to use it in practical and theoretical calculations carried out in that language. It does not warrant any beliefs about the existence of the extra linguistic objects for which the abnormality a r g u m e n t says we can h a v e no recognizably good evidence. A n d of course if 'there are raccoons' (or 'there are physical objects') are external existence claims, then they are nothing like the claims to which the skeptic objects. But it is easy to see that R is not an internal existence sentence if (as one would expect) T includes or presupposes a characterization of the existence of raccoons which is distinct f r o m its observation sentences, its rules of evidence, or any combination of them. T h e characterization of the existence of raccoons would be drawn from a theory of space and time, an account of what it is for a place to be filled by a physical object
220
JAMES
BOGEN
at a time, and a classification scheme which determines how physical objects are sorted and classified under natural kinds. R is true only if a thing with the properties T requires of raccoons occupies what is, according to T, place, p, at what is, according to T, time t. Although T should contain bridge sentences according to which a set of observation sentences, o, constitute good evidence for R, the properties which a thing must have in order to be a raccoon in a given place at a specified time are not themselves all observational. For example, nothing is a raccoon unless it has a certain genetic makeup. T h e presence of the biological features which constitute this (e.g., features of the animal's DNA) are not themselves directly observable. T h e r e are observation sentences which, together with the relevant bridges make it unreasonable to deny R, but, as skeptics rightly point out, the denial of R is logically consistent with the truth of the bridge and observation sentences. Thus R is not, and is not logically equivalent to the claim that R is well confirmed. But what confirms R makes it reasonable to believe (and warrants the assertion of) R, and not just Ri. While Ri says merely that a certain sentence (R) is well confirmed in L, R itself says what is false unless places and times actually (and not just apparently) are what T thinks they are, and unless a certain place is filled at time t by an object which actually meets T's conditions for raccoonhood. Thus R is not logically equivalent to, and afortiori, does not have the same meaning as Ri; it is not an internal sentence. Nor is R the answer to, or equivalent to the answer to an external existence question. T h e reason R is not an external sentence is simply that the world could be quite different from what the sentences of T say it is even if L, the language of T were overwhelmingly satisfactory on pragmatic grounds. Suppose for example that Berkeley was right in thinking the world contains nothing but ideas and minds, none of which fit T's characterizations of physical objects, and suppose sentences in L which contain such physical object terms as 'raccoon', 'piece of paper', etc., cannot be translated into a language whose nonlogical terms mention only Berkleyian ideas and minds. This would make R, and many other sentences of T false. But all the same (as Berkeley himself seems sometimes to have thought) we might be unable to conduct our theoretical or practical affairs in a language without physical thing terms. T h e falsity of T is thus perfectly consistent with L's turning out to be pragmatically indispensible. What this shows is that the external sentence reading of 'there are raccoons' could be true even if R were
TRADITIONAL
EPISTEMOLOGY
221
false, and thus that R is not and is not logically equivalent to an external existence sentence. T h e moral of all of this is that if what I have said is correct, we can, contrary to the skeptic, have recognizably good evidence for the truth of sentences which - unlike Carnapian internal and external existence sentences - make full-blooded claims about the world. In other words, nothing I have considered in this paper prevents naturalism from being realist enough for traditional epistemological purposes. 10. Now for the bad news. T h e reason n o n H u m e a n skeptical arguments are subject to naturalist refutations is that they depend upon empirically testable assumptions. A skeptical argument which made no empirically testable assumptions would be immune to naturalism, and as far as I can see, there are H u m e a n skeptical arguments of this kind. T h e y concern components of a theory of evidence which are different from the bridge sentences we had to consider in connection with the Abnormality argument. These are "confirmation policies" - rules, practices, standards, etc., which determine how evidence (whose relevance to confirmation is established by bridges) is to be weighed, and what degree of confirmation it bestows upon the sentences it tests. Here is an abstract, schematic representation of one kind of Humean argument. Let s be any sentence or collection of sentences in any theory, T, whose confirmation depends upon an observational generalization, G, which claims for a domain including more instance than have or can be observed, that when observational feature F1 is present, another observational feature, F2 is always or usually present. Suppose bridges establish the relevance of G to the confirmation of s. Suppose that we have no worries about determining the reliability of observation sentences reporting occurrences of F1 and F 2 needed to confirm G. Suppose there is no epistemically good reason to believe s unless G is well confirmed. Finally, suppose the best evidence for G consists of observation sentences reporting that F 2 was present when F1 was present in n of m cases, where m is greater than n, and n/m is greater than one-half. Is G well enough confirmed to provide good reason to believe s? That depends upon what confirmation policy we accept. Under one policy, the probability that G (however "probability" is interpreted) is n/m (or " h i g h " if no precise metric applies). Under
222
JAMES BOGEN
another policy, it might be calculated as 1 - n/m (or "low"). We can imagine still m o r e policies. U n d e r some, G (hence s) is confirmed to varying degrees, while under other, G is disconfirmed to different degrees. As long as neither G nor its negation are entailed by the observation sentences and bridges we can expect there to be alternative confirmation policies conferring different degrees of confirmation and disconfirmation upon G, each of which is incompatible with the others, each of which is compatible with the observation and bridge sentences. 23 Suppose there are such alternative confirmation policies. A H u m e a n skeptic argues that for all we know (where all we know is the observation and bridge sentences, and what degrees of confirmation they confer upon G under each of the confirmation policies) there are i - 1 chances that any given policy is incorrect, (assuming that one and only one alternative is correct). Accordingly, no policy has a better chance of being correct than any other, and none has a high chance of being correct. Therefore, we have no good reason to accept a policy under which G and s are well confirmed o v e r a policy under which they are not. Thus, in spite of observational evidence which looked good for G, we have no good reason to think G or s are well confirmed. If the confirmation o f T depends upon using evidence in enough cases which fit this schema, we cannot tell if T is well confirmed by even the most extensive observational evidence - unless we can find a reason to accept a confirmation policy which is favorable to T. In order to see how serious a p r o b l e m this kind of a r g u m e n t really poses, we should have to do a great deal, including looking in detail at a well set out version of a theory to see how m a n y of its sentences must actually be confirmed in a way which requires the confirmation of a generalization like G, whose confirmation involves policies subject to the H u m e a n objection. Without considering any of this, it seems plausible to say that the H u m e a n ' s reasons for thinking that G and s cannot be confirmed are not susceptible to empirical testing. T h e H u m e a n a r g u m e n t apparently stems from claims about the logical possibility of alternative confirmation policies, and calculations of the different degrees of confirmation or disconfirmation they assign on the basis of available evidence. Perhaps empirically testable assumptions lurk behind the H u m e a n premisses, but if there are any, it is hard to see what they could be. This is by no means to say that the H u m e a n argument is immune to nonempirical, a priori objections. T h e point is that unless appearances
TRADITIONAL
EPISTEMOLOGY
223
are highly deceiving, it is immune to the naturalist approach I advocate for dealing with nonHumean skeptical arguments. Once again, this is no reason to give up naturalistic objections to the Abnormality and the Gettier arguments. Instead it is a reason to be pessimistic about naturalism as a strategy for dealing with Humean skepticism. NOTES l T h e ancestor of sections 8 and 9 of this paper was a discussion of c o m m e n t s m a d e by Nicholas Rescher, Mark Pastin, Robert Kraut, William Lycan, and others at a session of the Oberlin Philosophy Colloquium at which I read a draft of the B o g e n - B e c k n e r paper on skepticism which is m e n t i o n e d below. T h e need to distinguish H u m e a n from n o n H u m e a n skeptical a r g u m e n t s (section 10) was b r o u g h t painfully to m y attention by T e d d y Seidenfeld in discussion during the Pittsburgh conference at which this paper was first read. I wish he h a d n ' t done it. Section 7 is heavily influenced by Clark G l y m o u r ' s bootstrap theory of confirmation, although I have no idea whether he would approve of the use to which I have put it. I am also greatly indebted to Morton O. B e c k n e r and Peter M a c h a m e r , with w h o m I have talked about skepticism every time I got a chance over the past ten years. W h e t h e r or not a sentence counts as an observation sentence depends upon the use to which it is put u n d e r the theory to which it belongs. T h u s what is an observation sentence under one theory, need not be under another, and within the same theory, what is sometimes used as an observation sentence m a y be used on other occasions for other purposes, cf. Jeffrey (1973). 3 A report which is abnormal and therefore unreliable with regard to one sentence m a y be abnormal but reliable with regard to another sentence. T h u s the report of a hallucinatory seeing of an elephant is in most cases unreliable evidence for the claim that an elephant was present, but would usually c o u n t as reliable evidence for the claim that elephants can be hallucinated. Furthermore, what is abnormal and therefore unreliable evidence for a sentence in one context m a y be abnormal but reliable evidence for the same sentence in another context. A d r e a m sighting of a road runner is usually unreliable evidence for the presence of road runners. But if Jones's d r e a m sightings are always caused by road runners pecking him on the head, abnormal observation reports of his sightings m a y be reliable evidence for claims about the presence of road runners. In what follows, all such complications will be ignored. 4 All but a very few of the m o d e s and tropes of Sextus Empiricus are n o n - H u m e a n arguments, as are Descartes' D r e a m arguments, and m a n y other stock skeptical a r g u m e n t s which derive from Sextus. 5 This view is n a m e d after Protagoras who held, according to Plato, that for any observer, X, and any object, Y, a sentence of the form ' Y is F - f o r - X ' is true just in case Y appears to X to be F. If appearances are considered warrants, then whatever X is warranted to assert is true, and so interpreted, Protagoras' is the first warrantability a c c o u n t of truth. 6 Descartes' own version of this requirement is apparently m u c h stronger than what I am calling the Cartesian Constraint. 7 A serious objection to m y reply to the n o n - H u m e a n skeptic would be to produce a
224
JAMES
BOGEN
theory, T' according to which abnormal evidence is both unreliable and indistinguishable from normal evidence, which can be confirmed as well as T, and whose confirmation is non-trivial in the sense that T's may be. Whether there could be such a theory is an open question. None of the skeptical arguments considered here (and no skeptical arguments that I am aware of) provide any reason to think there might be such a theory. Section 6 of Bogen and Beckner gives plausibility arguments against the possibility. Thus, although the development of confirmation of T' would vitiate the naturalistic argument against skepticism, I know of no reason to think such a theory could be developed or nontrivially confirmed. The mere fact that such a theory would be an objection is harmless unless T' can be developed and confirmed; I think it can't. 8 Carnap himself would have said that 'there are numbers' is always (if meaningful) an external, rather than an internal existence claim (Quine, 1966, p. 130 f.).
REFERENCES Austin, John L.: 1946, 'Other Minds', Proceedings of the Aristotelian Society Supplementary Volume 20, 148-187. Bogen, James and Beckner, Morton O.: 1979, 'An Empirical Refutation of Cartesian Skepticism', Mind 351, 351-369. Carnap, Rudolph: 1956, 'Empiricism, Semantics, and Ontology', reprinted in Carnap, Meaning and Necessity, University of Chicago, pp. 205-221. Gettier, Edmund L.: 1963, 'Is Justified True Belief Knowledge?', Analytics 23, 121-123. Glymour, Clark: 1980, Theory and Evidence, Princeton University Press, Chapter V. Goldman, Alvin: 1967, ' A Causal Theory of Knowing', Journal of Philosophy 12, 357-372. Goldman, Alvin: 1976, 'Discrimination and Perceptual Knowledge', Journal of Philosophy 62, 771-791. Jeffrey, Richard C.: 1973, 'Review of Putnam', reprinted in Grandy, Theories and Observations in Science, Englewood Cliffs, N.J. Plato: Theaetetus, 152A-E, 159B-161B, 166A-168C. Price, H. H.: 1961, Perception, London, Methuen, p. 2. Putnam, Hilary: 1978, Meaning and the Moral Sciences, Routledge and Kegan Paul, London. Quine, W. V. O.: 1966, 'Carnap's Views on Ontology', reprinted in W. V. O. Quine, The Ways of Paradox, Random House, pp. 130 ff. Sober, EUiott: 1981, 'Revisability, A Priori Truth, and Evolution', Australasian Journal of Philosophy 59, 68-85. Swain, Marshall: 1978, 'Reasons, Causes, and Knowledge', Journal of Philosophy 65, 5. Pitzer College Philosophy Fieldgroup Claremont, C A 91711 U.S.A.