Laurence BonJour, Knowledge, Justification, and Truth



5. The Nature of Coherence.

The outcome of the first chapter was the conclusion that the doctrine of Givenness is of no use in solving the problem of how empirical knowledge is ultimately justified. This essentially negative conclusion has the effect of returning us to the same dialectical position from which the essay began, with the only difference being that one of our three putatively possible solutions to the problem has been eliminated. It turns out, that is, to be impossible to justify a propositional or conceptual claim in terms of something non-propositional and non-conceptual; justification can therefore come only from within the conceptual framework. Thus the body of propositions in which human knowledge consists must be justified from within, presumably in terms of relationships among its component propositions, if it is to be justified at all, and hence if it is to be genuine knowledge at all. That is, the only alternative to skepticism is coherence.

To many philosophers such a conclusion will seem only slightly less paradoxical than would an unabashed skepticism. Of course, they might say, coherence is one important factor in justification and hence in knowledge. But how can it possibly be the only factor? How can mere internal relationships among a set of (believed) propositions qualify them as epistemically acceptable? Why couldn't one make up such a coherent set of propositions entirely arbitrarily? Doesn't such a view deprive justified belief of all but accidental relation to the world? Doesn't such a coherence account of justification completely eliminate the factual or empirical aspect of "empirical knowledge?" Doesn't it destroy the very notion of scientific objectivity and reduce us to the linguistic equivalent of idealism?

That this torrent of protest has genuine force can, I think, scarcely be doubted. Indeed, it was just such arguments and objections which historically overthrew the coherence conception of knowledge characteristic of nineteenth century idealism and launched Anglo-Saxon philosophy on the path of immediacy or Givenness. (Which is not, of course, to suggest that that path had never been trod before.) But if the argument of the first chapter is sound, the path of Givenness is ultimately a blind alley, which fact forces us to retrace our steps, to pursue the metaphor, and face up to the above objections. If they cannot be satisfactorily answered, if a coherence theory of justification cannot be made reasonably plausible from a broadly empiricist point of view, then skepticism will indeed be the only remaining alternative, and we shall have no choice but to embrace it cheerfully.

I believe that the objections in question can in fact be answered to an extent which greatly mitigates their force, and which makes a coherence theory of justification rationally defensible, and I shall attempt to provide such answers in the course of this chapter. (Though this is not to preclude the possibility that enough force may remain in the objections to compel us to accept a rather more modest conception of human empirical knowledge, and of the human cognitive enterprise generally, than is held by those who, like Lewis, believe in a foundation for knowledge which is certain, indubitable, and incorrigible. Such an increase in epistemic modesty would be roughly analogous to that which resulted from the earlier realization that empirical knowledge could never possess the certainty characteristic of mathematics and logic.) The essential core of such an answer is, I think, an account of the notion of observation which eliminates any dependence on Givenness, and harmonizes observation with a coherence theory of justification. To provide such an account of observation and coherence, along the general lines suggested by Sellars, {12} is the goal of the present chapter. The balance of the present section will attempt to prepare the ground by broadly sketching the relevant notion of coherence, while succeeding sections will deal with the notion of observation, and with the relation between observation and coherence.

The concept of coherence is a vast and complicated one, fully as broad as the notion of inference with which it is most intimately connected, and I cannot hope to provide anything like a comprehensive treatment of it within the compass of the present essay. As will soon become apparent, if it is not already, a full account of coherence would have to deal in detail with issues concerning, e.g., induction and probability, which are themselves subjects for entire books. All I can hope to do here is to sketch a clear enough picture of the general notion of coherence and of its relations to other epistemological concepts to provide a basis for the more detailed discussion of observation and coherence which is to follow. In any case the purpose of the chapter is not to fully state a coherence theory of justification, but only to show that such a theory represents a viable alternative to the Given.

Coherence has two related aspects or may be considered from two different standpoints, corresponding to two distinct modes or levels of justification. On the one hand, there is coherence as involved in the justification of a single propositional claim in terms of other propositional claims, in the context of a particular conceptual framework and of a relatively fixed body of propositional claims within that framework. On the other hand, there is coherence as involved in the justification of an entire body of propositional claims and, by implication, of the conceptual framework within which they occur. I shall refer to the former aspect of coherence as "coherence in the small" and to the latter as "coherence in the large." It is not, of course, to be expected that the accounts of these two will prove finally to be independent of one another.

The central notions in terms of which coherence in the small is to be understood are the related notions of inference and argument. In the most general terms, one justifies a particular propositional claim, that-P, by providing an acceptable argument which has that-P as its conclusion. To say that the argument is acceptable is to say that the premiss or premisses are known, and that the inference from those premisses to the conclusion that-P is of an acceptable sort, i.e. meets whatever standards are appropriate for the variety of inference of which it is an instance (that some standards or other are appropriate, even though we may not know precisely what they are, is, I take it, the cash-value of saying that the transition from the premisses in question to the conclusion that-P is an inference rather than, say, a "leap of faith").

This general characterization of coherence in the small requires several comments and qualifications. First. I have said and shall say very little here about the sorts of inference which might be involved in such a justification, except to remark that any epistemically respectable sort of inference will do. On the most standard view, this would include deductive inference plus whatever varieties there may be of inductive and/or probabilistic inference. Whether this taxonomy of inference is the best or most helpful one I shall not attempt to say. All that the present view demands is that one or more (roughly specifiable) ways of deriving one proposition from one or more further propositions be epistemically acceptable, and others not. Any mode of derivation thus acceptable will be what I mean here by an acceptable mode of inference.

Second. It is worth noting in passing, however, that some justifying inferences at least will depend on those patters of inference which, according to the account sketched in chapter one, constitute the very concepts involved in the premisses and conclusion. Thus that some potentially justifying inferences are available is a necessary condition for the very existence of a conceptual system and of a proposition to be justified.

Third. Clearly simple consistence seems also to be a requirement for coherence. This means that the proposition that-P and the proposition that-not-P cannot simultaneously be justified. Since it does not seem impossible that there could be prima facie acceptable arguments for both of two such contradictory propositions, it would seem that a coherence theory of justification must allow for some sort of principle or principles for resolving such cases of conflict, and perhaps also for preventing their occurrence. One principle of the latter sort is the familiar requirement of total evidence, which stipulates roughly that a proposition is to be regarded as probable only if it is probable on all the evidence; this principle thus prevents one from "gerrymandering" the evidence so as to produce probabilistic arguments for each of two or more conflicting propositions. The principles for deciding an actual case of conflict which is not prevented by the principle of total evidence are much less clear, though presumably considerations of relative probability and of "simplicity" play a major role. In any case, the only point which matters in the present context is that the justification afforded by a justifying argument is always to be regarded as only prima facie, subject to the possibility of such a conflict and thus to the relevant principles of adjudication, whatever they may be.

Fourth. To say that A's belief that-P is justified only if A actually provides an acceptable justifying argument with the conclusion that-P is to oversimplify in a way which leads inevitably to the regress problem of chapter one. For if a justifying argument must actually be provided by A, then it would be possible to order A's beliefs serially according to the order in which such justification was in fact provided for them. This would have the consequence that a particular belief in the resulting sequence, that-Pn could only have been justified in terms of inferences from one or more of the propositions that-Pn-1, that-Pn-2, . . . , that-P1, which precede it in the sequence, because only those propositions would have been justified and hence putatively known at the time at which justification for that-Pn was in fact provided. But this would mean that the first proposition in the sequence, that-P1, could not have been justified in terms of inferences from any other propositions at all, since none were available at the time at which it was justified, and hence that that-P1 could not have been justified at all, since we have already ruled out justification by something non-propositional and non-conceptual. And the further consequence would be, of course, that none of the propositions in the sequence was justified, since the sequence of justifications could never get started.

I do not believe that this problem is particularly difficult to solve. Indeed, the solution is worth mentioning only because the regress argument is, as we saw in chapter one, one major prop which has seemed to support the doctrine of the Given, and also seems to still turn up with some frequency.{13} That solution is simply to reject the idea that a proposition is justified only where justification in the form of a justifying argument has been explicitly provided. Clearly such a view is unacceptable anyway unless we are willing to say that we do not really know most of the things which we ordinarily think that we know. For we do not ordinarily provide justifying arguments in explicit form unless we have some specific reason for doing so, e. g. if one of our claims has been challenged. We should rather say that A's belief that-P is justified if and only if A could, if challenged, provide an acceptable justifying argument [cf. EPM 169]. (Exactly what force to give this use of 'could' is unclear, since most of us in fact could probably not provide explicit justification for many of the things we think we know, at least not without considerable thought. Perhaps one should speak of a strict sense of 'knowledge', in which one must actually be able to provide justification, and a derivative sense in which it is only required that the materials for such a justification be in some sense within one's grasp, even though one may be uncertain at the moment just how to deploy them.) Such an account makes it clear how a body of propositional claims can be self-supporting, each such claim being justifiable in terms of some of the others, without any of them being justifiable apart from the others.

The conception of coherence in the small just sketched avoids the regress problem of chapter one, but is immediately faced with a quite different objection. Even if it be granted, so the objection goes, that you have shown how the various component propositions, that-P1, that-P2, etc., of a particular framework of propositions F1 in a particular conceptual system S1 can be justified relative to one another and to F1, this does nothing to solve the problem of whether and how they are justified tout court. For there is absolutely no reason to suppose that the status just described of F1 and its component propositions is in any way unique or special. I.e., there is no reason to suppose that one could not find or devise an alternative framework of propositions F2, in S1, whose component propositions would be justified relative to one another and to F2, and in which some or all of the propositions of F1 would not be justified. And the same could equally well be true of yet further frameworks of propositions in S1. Moreover, the objection continues, there is no reason to think that the conceptual system S1 in which Fl, F2, etc., appear is itself unique. On the contrary, it seems quite reasonable to suppose that one could with ingenuity devise further conceptual systems S2, S3, S4, etc., each of which would contain yet further frameworks of propositions which were internally justified in precisely the way in which we have supposed F1 to be. Thus on your view of justification you will have to say that all such bodies of propositions, and hence all the individual propositions composing them, are justified. And since there is no reason to suppose that any self-consistent proposition, to whatever extent it makes sense to talk of individual propositions in isolation, could not appear in some such internally justified body of propositions, the implication seems to be that all self-consistent propositions are justified. And surely that conclusion constitutes a reductio ad absurdum of your coherence theory of justification. Thus the objection. Clearly what is needed by way of an answer is an account of what I have called "coherence in the large," i.e. coherence as involved in deciding between different whole frameworks of propositions, possibly in different conceptual systems. Again, as in the case of coherence in the small, I shall of necessity content myself with the broad outlines of an account.

It is well to begin here by exposing a serious and perennial mistake made in discussions of coherence and justification. It has too often been assumed that to say that a given body of propositions is coherent is simply to say that it is consistent and nothing more. Thus, e.g., Scheffler, in his recent book Science and Subjectivity, gives the following characterization of coherence in the large:

. . . Faced with a conflict between my observation reports and my theory, I may freely alter or discard the former or the latter or both, so long as I replace my initial inconsistent set of beliefs with one that is coherent. Clearly this much freedom is too much freedom. Constraints beyond that of consistency must be acknowledged. [My italics.]{14}

Clearly Scheffler simply equates coherence with consistency. But this is just a mistake, pure and simple. Coherence does not mean merely consistency, and two equally consistent (i.e. just plain consistent) bodies of propositions need not and usually will not be equally coherent. Consistency is a yes-or-no matter; a body of propositions is either consistent or it isn't. But coherence is a matter of degree. (Consider as an example here the following two bodies of propositions, (A) and (B). (A) contains (i) "This chair is brown."; (ii) "Electrons are negatively charged."; and (iii) "All mammals have hair." (B) contains (i) "All ravens are black."; (ii) "George is a raven."; and (iii) "George is black." Clearly though body (A) is perfectly consistent, it is nonetheless much less coherent than body (B).)

Intuitively, coherence is a matter of how well a body of propositions "hangs together," i.e. of how well its parts cohere with each other. But how exactly is this vague notion of "hanging together" to be understood? Part of the answer is already implicit in the above account of coherence in the small, and depends on the notion of inference. Thus the coherence of a body of propositions will vary, in part at least, with the extent to which its various component propositions are linked by inference relations and hence could be justified in terms of one another in the way just discussed. This means, since presumably not all such justifying inferences will be deductive in character, that coherence in the large will depend on the degree to which each individual proposition in the body of propositions in question is made probable by the other propositions in the body. That body of propositions whose internal probability relations are stronger in this way will be more coherent, and hence on the present account better justified.

This is not quite the whole story, however. To see why, consider the possible case of a body of propositions which consisted of several sub-bodies whose internal inference connections were as tight as you please, but which had little or no connection with each other. Intuitively, though the degree of coherence of each sub-body might be very high indeed, it would seem that the overall coherence of the body of propositions would be fairly low. Thus coherence in the large seems to require more than that inference and probability linkages, however tight, hold within small bodies of propositions. What is also required is that the total body of propositions form a unified structure, i.e. that there be basic laws and principles which underlie widely separated sub-bodies of propositions and hence provide some degree of inferential connection between them. One is reminded here of the familiar notion, so dear to the positivists of a "unified science," in which the laws and terms of all the disparate disciplines of science are reduced to the laws and terms of some one discipline, presumably physics.{15} Clearly one's overall scientific account will be more coherent to the extent to which such unification is obtained (though this, unfortunately, assumes that the laws of the unifying discipline are themselves unified or connected).{16}

A different, though related, approach to the same idea would invoke the familiar account of scientific explanation given by contemporary philosophy of science. On the most standard account of explanation, viz. that of Hempel,{17} one explains a particular fact by providing statements of other facts and of laws from which a statement of the explanandum fact can be inferred. Clearly such an explanation also provides inference linkages of the sort which can be employed in justification and which contribute to coherence in the large. This is especially clear when the various particular facts are already known and one provides an explanation of some of them by providing a law, of either universal or a probabilistic sort, which links these facts with others, thus allowing the former to be inferred from (or justified in terms of) the latter. The better explained (the fact stated by) any particular proposition in some body of propositions is, the more that proposition will cohere (in the small) with the rest of that body of propositions, and the more coherent (in the large) the overall body of propositions will be [cf. RLG 356].

More importantly, however, one can also explain laws by inferring them from other, "higher-order" laws.{18} It is this sort of subsumption of many different laws under one more basic law which unifies a scientific account, and makes it more coherent. It is the search for such unifying principles which is the basic motivation for the postulation of theoretical entities and laws governing them. As Hempel remarks at the end of the chapter on theoretical explanation in his Philosophy of Natural Science:

. . . What scientific explanation, especially theoretical explanation, aims at is . . . an objective kind of insight that is achieved by a systematic unification, by exhibiting the phenomena as manifestations of common underlying structures and processes that conform to specific, testable, basic principles.{19}

I think that what Hempel here calls 'systematic unification' is exactly what I have called 'coherence in the large'.

This view of the relation between explanation and coherence also provides, I believe, a better understanding of the epistemic role of the systematic conception of explanation, as distinguished from what might be called the pragmatic conception of explanation. In the pragmatic sense, anything which increases one's understanding of some phenomena or makes the unfamiliar familiar can be called an explanation, but in the systematic sense only that which contributes to a more coherent overall body of knowledge can explain.{20} Since a systematic explanation thus increases the coherence of a body of propositions, and since a more coherent body of propositions is more justified and hence more acceptable, there is a clear epistemic motive for adding the best of competing explanations (where 'best' is presumably to be explicated in terms of the very sort of systematic relations which we have been considering) to a body of propositions containing the putative explanandum. This resembles what Harman has called "the inference to the best explanation."{21} (Though it seems clear that the sense of 'inference' in which one infers an explanation is not and cannot be the same sense of 'inference' in terms of which coherence in the small was explicated above; rather the former is to be explicated, in part, in terms of the latter. The fact that a particular proposition results in the best explanation of those available does not seem, by itself, to justify it as putative knowledge; if no justification in the sense explicated above, i.e. which involves the more basic sense of 'inference' is available, the explanatory proposition seems to be regarded as an hypothesis, rather than as known.)

If this account of coherence in the large and thereby of justification is generally on the right lines, it has also a further implication which is of the utmost importance. It is a familiar enough fact of the history of science that scientific progress often involves not only the acceptance or rejection of propositions within some conceptual system, but also the alteration of the very conceptual system in terms of which the scientific investigation was conducted, either through the addition of new concepts or through changes in old ones. The clearest case of the latter is of course the Einsteinian revolution in theoretical physics; it seems simply to be a fact, given only that the notion of "sameness of meaning" is transitive, that key terms like 'mass', 'simultaneity', etc., no longer have the same meaning that they had in pre-relativistic physics (which is not of course to say that the meanings are totally unrelated). But although the fact itself is familiar enough, it has seldom if ever been really assimilated into epistemological theory, where the tendency has been, on the contrary, to regard such cases as aberrations, to be explained away rather than explained. In the light of a coherence theory of justification such as here proposed, the other hand, such facts fit neatly into the epistemological picture in a manner which reveals them to be, not aberrations, but, in a way, the heart of the matter. For since coherence depends essentially on inference, clearly one way to arrive at a more coherent total account is to modify the inference patterns obtaining between the various propositions in one's conceptual system. But, if the account given in chapter one is correct, those inference patterns, or some of them at least, are constitutive of the very concepts employed in those propositions, so that to modify the inference patterns is to modify the concepts. (Thus, e.g., in relativistic physics one can no longer infer from the fact that event A is simultaneous with event B for some observer, and that event B is simultaneous with event C for some observer, that event A is simultaneous with event C for all observers; whereas such an inference was available in classical physics.) And, of course, to add new concepts is to add new inference patterns, connecting the new concepts with the old. Thus one would clearly expect alteration of one's conceptual system to constitute an integral part of that attempt to arrive at a more acceptable, i.e. more coherent, account of the way the world is, which is the process of scientific development [cf. RLG 356].

This account of coherence in the large provides at least a schematic answer to the earlier objection that on our account all propositions would turn out to be justified, by providing a basis for a choice between competing bodies of propositions, all of which are consistent and to some extent coherent. That body of propositions is to be preferred which is most coherent in the sense just explicated. This is, if I am right, the sole ultimate criterion in terms of which issues of justification and of epistemic acceptability are to be adjudicated. We may also modify our earlier account of coherence in the small to say that a single proposition is justified tout court if and only if it is justified relative to that body of propositions which constitutes the most coherent and hence most acceptable account of the world. (It should be obvious that all such issues of justification and acceptability are relative to a particular time.)

Our earlier critic is, however, likely to reply at this point that our account of coherence in the large at best only mitigates the force of his objection, rather than refuting it. For if coherence is the sole basis for choosing among alternative accounts of the world, then even if coherence admits of degrees, so that not all bodies of propositions which are coherent in the sense that their component propositions are in some degree justified relative to one another are equally acceptable, it may still very well be the case that we are unable to choose a very large number of equally coherent accounts of the world. And wouldn't that consequence be quite enough to rule out a coherence theory of justification? I am not at present in a position to answer this objection in anything more than a promissory fashion. Clearly, if it is to have any force, it must be taken to mean, not that a coherence view of justification might find it impossible at a given moment of scientific history to decide between two or more alternative accounts of the world; this practical undecidability seems to afflict any conceivable view of justification, and in any case is clearly a feature of the standards of justification which we actually employ, whatever they may be. Rather the objection must be taken to mean that it might prove impossible even in principle, at the "end of inquiry," to decide among two or more equally coherent accounts. To the objection construed in this way, the answer is that although such a possibility cannot be ruled out absolutely, it will seem much less likely once the complicated way in which observation operates through coherence to add new propositions to an account of the world is made clear. I will elaborate this answer somewhat at the end of the chapter.

The mention of observation, however, calls to mind a rather more serious objection to a coherence view of justification, which will force a considerable complication of our account. This objection is the one outlined in the second paragraph of this section, to the effect that a coherence theory of justification seems at least to deprive the notion of justification of all relation to the world. Why should the fact that a particular putative account of the world is the most coherent in the indicated sense provide any very strong reason for thinking it to be accurate or true? How can coherence by itself be an adequate criterion of justification? Does not such a view abandon all vestiges of scientific objectivity, and even of that very basic empiricism which has represented common ground for virtually all twentieth century Anglo-Saxon philosophy?

I have already indicated my belief that this objection, though formidable, can be answered, and have suggested that the answer ultimately turns on the notion of observation. A first step toward seeing how this is so can be made by realizing that the goal of the human cognitive endeavor is to describe the whole world, and that such a description must therefore include that very prominent feature of the world which is the human knower himself. Thus the coherent account of the world which results from the advance of science will include an account of the behavior, in a broad sense, of such human knowers in relation to the rest of the world, which sub-account must cohere with the overall account. This suggests that a solution to the problem of how observation and an empirical outlook generally fit into a coherence view of justification might be obtained by interpreting observation and other related activities of a broadly empirical sort as aspects of the behavior of human knowers. The account of these aspects of human behavior would then be a part of the total account of the world and could therefore, via the ties of coherence, impose a constraint from within upon that account, a constraint which would be at least broadly empirical in character.

I think that something very much like this suggestion is in fact correct. Clearly, however, it needs to be spelled out in much more detail before it can constitute an answer to the objection just considered. And the first requirement for such a spelling-out is an account of observation of the appropriate sort. I shall attempt to give such an account in the next section.

6. Observation Without the Given.

My aim in the present section is to give an account of the concept of observation which squares with the results of chapter one, i.e. which avoids construing observation in terms of Givenness. Although the final cause of my discussion is a unified account of observation, coherence, and justification, I shall ignore these more systematic concerns for the moment, and concentrate on observation for its own sake, leaving until the next section the job of fitting the resulting view of observation into a broader theory of justification.

In considering the concept of observation, it will help to have a simple and relatively non-controversial example before us. Thus consider the following case. Someone, for whatever reason, perhaps to test my color perception, asks me what color the pencil on the desk in some designated room is. I thereupon proceed to the desk in question, look at the pencil in normal lighting and atmospheric conditions, and observe (i.e., in this case, see) that it is red. What account are we to give of such a case?

In the first place, what is clearest is what results from such an observation, viz. my having the belief, which may or may not express in an overt utterance, that the pencil on the desk is red. Such a belief is clearly propositional and conceptual in character. According to the account of conceptual apprehension sketched in section three, it is to be viewed as constituted or determined by the various sorts of further inferences which one is prepared to make from it. Thus in the present case I would presumably be disposed in appropriate circumstances to infer from my initial belief, e.g.: that there is a red physical object on the desk; that the pencil on the desk is not green; that the pencil on the desk will, ceteris paribus, be red tomorrow; that the pencil on the desk would be approximately the same color as a ripe apple; that it would not be the same color as a dandelion; etc. To have such a belief is to be disposed to make such inferences (and also to be disposed to act in various ways, depending on one's desires and other beliefs); it is, in Sellars' metaphor, to be at a certain position in "inference space."

Secondly, it is a feature of observation that the belief in question is itself arrived at non-inferentially. Thus if my arriving at the belief that the pencil on my desk is red is to be a case of observation, it must not be the case that I arrive at it by remembering that I saw a red pencil there yesterday and inferring that probably the same one is still there; or by seeing a paper with red writing on it lying on the desk, and thereby inferring that the pencil on the desk is probably red, since pencils with red lead are generally also painted red; or by remembering that the owner of the desk always uses red pencils, and thus inferring that this one will very likely be red also. The fact that the belief in question is consciously mediated by any such inference is sufficient to rule it out as non-observational.

Thirdly, it seems to be a further feature of the concept of observation that the belief in question is causally produced in a certain way, viz. by the action of the environment on one or more of my sense organs. Thus if I genuinely observe that the pencil on the desk is red, as opposed to coming to believe it irrationally or through mystical insight, it seems to be required that the belief have resulted causally from an appropriate sort of sensory stimulation, in this case presumably from the action of light rays on my eyes. The fact that I was blind, or that the room was dark, would normally suffice to disprove my claim to have observed the color of the pencil, unless I could demonstrate that I possessed some alternative sense for discriminating colors. And similarly for other cases of observation: that a belief results from observation presupposes that the putative observer possessed a sense organ capable of detecting the sort of fact claimed by the belief. (Which presupposition is, of course, generally taken for granted in normal cases.)

The traditional empiricist epistemology was inclined to supplement these three factors in its account of observation with a fourth, namely a sense experience or sense impression, which was intended to provide a link between the sensory cause and the resulting belief, and was regarded as distinct from both. Thus, in the case of our example, the traditional view would say that the causal action of my environment results in a sense impression of a red pencil (or perhaps of a red object of a certain shape, or even of a red, two-dimensional patch -- here the accounts diverge), on the basis of which experience I come to believe that the pencil on the desk is red. It should be clear that this account is no longer available, or at least no longer plays any useful epistemological role,{22} once the Given is abandoned. The reason is that the sense experience in question must itself, in terms of the distinction of chapter one, be either conceptually or discriminatively apprehended. If the former, then the crucial phrase "on the basis of" in the above account can be glossed as referring to an inference from the conceptual apprehension of the experience to the belief about the pencil, but that apprehension seems to be or involve just a further belief, so that we have answered questions about the derivation and justification of one belief, that the pencil on the desk is red, only by introducing a further belief, that the sense experience in question is thus-and-so, to which the same questions apply. Moreover, this futile pushing one step back of the questions is accomplished at the serious cost of making our original putatively observational belief, that the pencil on the desk is red, not observational after all (because inferential), and at the further cost of introducing all the puzzles concerning "unconscious inference" and perhaps also beliefs which the average person is normally quite unaware of, and may even lack the vocabulary to express. Alternatively, if the apprehension of the sense experience is of a purely discriminative sort, then the key phrase "on the basis of" will have to be glossed in purely causal terms, since there will be no basis for anything like inference. And then the sense experience will be at best only another link in the causal chain from environment to belief, of no special epistemic importance. (No one ever thought that the causal chain stopped at the sensory surface.) Thus, once the Given has been abandoned, the traditional account of observation in terms of sense experiences or sense impressions becomes at best otiose. (In this paragraph I may seem to have been belaboring a point already made and remade many times in what has gone before; my justification is that the Given is one of those doctrines which has a way of popping up again and again even in the writings of those who have expressly repudiated it, and accounts of observation are one of the places where this phoenix-like quality has been most often manifested.)

(All this is not, of course, to deny the obvious fact that people do have sensory experiences, and that these experiences are involved in observation. It is only to deny that such experiences are a (non-conceptual) factor in observation distinct from the resulting conceptual apprehension. Presumably the upshot of the present account is that to have a sense experience is just to arrive at one or more, usually a large number of, observational beliefs, rather than something distinct from such beliefs which in some way either explains how they are arrived at or justifies them. But I shall not pursue these issues here.)

We seem to be left then with only the causal circumstances and the non-inferentially resulting belief as factors in the situation of observation. The obvious course of action is, I submit, to make a virtue of necessity, and construe observation as simply a state of belief, which is produced causally rather than inferred from other beliefs. On this view, to say that I observe that the pencil on the desk is red is simply to say (i) that I come to believe that the pencil on the desk is red, (ii) that this belief is non-inferentially arrived at, and (iii) that it is causally produced by the action of my environment on my sense organs. To observe is thus simply to occupy a position in "inference space" as a causal result of sensory stimulation, rather than via inference.

I think that such an account is very close to being correct. It needs, however, to be qualified in one respect, if it is not to include too much. We do not want to count as a case of observation every stray belief which is (in part) occasioned by sensory stimulation; there seems to be no reason why any given belief could not, in principle at least, be caused by any given stimulus, no matter how unrelated, on some occasion or other, if 'cause' is merely taken in the usual sense of 'necessary condition'. What rather happens in a case of observation is that the particular belief in question is, so to speak, "forced upon us" or "extracted from us" by the environment via the sensory stimulus. I.e., it is not an accident that particular belief is produced by that particular stimulus; which I take to mean not only that the stimulus is a necessary causal factor in producing the belief, but also that the particular sort of stimulus in question regularly produces the particular sort of belief in question. And if we suppose, as seems reasonable, that this regularity is not natural or innate but rather is acquired through training, it will seem appropriate to borrow a bit of psychological jargon and say that the belief is a conditioned response to the stimulus, and not merely caused by it [RLG 329-30, 333-34].

Thus I submit that observation is best regarded as non-inferential belief which is a conditioned response to sensory stimuli. The term traditionally used to refer to the verbal expression of such a belief is 'report'; it will be convenient in what follows to extend this usage, by employing the term 'report' or 'observation report' to encapsulate this notion of observation, and thus to refer indifferently to either the observational belief or the overt expression thereof, if such exists.

One point to be carefully noted is that this notion of observation so far says nothing about whether or not the report in question is true, or even generally reliable. This is quite deliberate. I do not want to deny that in the most ordinary sense 'observation' is an "achievement word," so that to speak in this sense of a mistaken observation would be a contradiction in terms. I would only want to claim that there is also a sense of observation in which this is not so, in which to say that I observed that the pencil on the desk is red does not entail that the pencil on the desk is really red, or even that there really is a pencil on the desk. (And in any case, whether or not such a sense of observation actually exists in the vernacular, it seems quite legitimate to invent one for philosophical purposes; perhaps one should speak here of "putative observation.") The important point is that only such a restricted sense of 'observation', in which observation and error are compatible, is useful for epistemological purposes, since the more ordinary sense conflates considerations of justification and truth in a potentially misleading way. And therefore it is in this restricted sense, whether ordinary or invented, that 'observation' will be used in the balance of this essay. (I do not mean to suggest here that justification and truth can be kept totally separate either. It is hard to see how the putative observation reports of a given group of people could fail to be at least generally reliable; but this does not preclude the possibility that some narrow class of reports, specified either as to content or as to observer, could be generally unreliable.)

Our account of observation is still only approximate, and would have to be qualified in at least the following respects in order to be fully accurate. In the first place, one would have to introduce qualifications to deal with the case where one already has the belief in question, and thus where it cannot be produced by the stimulus in question, though it may be confirmed or strengthened. Secondly, there is also the case where one is for some reason suspicious of the circumstances of observation or of one's own sensory equipment, and this results in one not coming to have a particular belief in circumstances which would ordinarily suffice to produce it. Presumably both of these cases are to be handled in terms of dispositions or inclinations to believe. Then there is also the complicated issue of how to fit the notion of "standard conditions" into the account of observation. An ideal account of observation would of course treat all such matters at length. But I do not think that the details of such a treatment matter very much for our present purpose of fitting the basic notion of observation into a coherence theory of justification. For these purposes, we may assume that we always have confidence in the conditions and in our sense organs, that the conditions are always standard, and that we are always dealing with a newly acquired belief. Our problem will still remain.

This account of observation is an expansion and development of the answer which was briefly adumbrated at the end of chapter one to the question of how the concepts of language and thought come to be applied to the world if the Given is indeed a myth. On the present view, concepts are applied to the world by means of a purely discriminative, conditioned response to sensory stimuli. One does not first recognize that the stimuli are thus-and-so, and then decide on that basis to apply the appropriate concepts, because to do anything reasonably characterizable as recognizing that something is the case or deciding one must already have applied concepts. Rather the application of the concept, the attaining of the appropriate belief, is a direct, i.e. cognitively unmediated, response to the stimuli in the same way in which withdrawing one's hand is a direct response to heat, or ducking to an approaching object. Such application does not result from a judgment, though it is of course itself, unlike other sorts of conditioned responses, a judgment [RLG 334]. (Presumably the disposition to such responses is acquired during the general process of conditioning via which one learns a language [RLG 333].)

One implication of such a view of observation is that the actual event of arriving at an observational belief, the occupation of a position in inference space, is essentially an a-rational (though not ir-rational) process, one for which it is neither possible nor appropriate to offer justification. Though such an event is, in the classical terminology, an "act," it is not an "action" because not voluntary or chosen [EPM 166-67]. (Though it can, of course, be resisted, as in the fishy circumstances, just as one can resist one's impulse to duck.) But, and this is the key point, the fact that the "act" of arriving at the belief neither allows nor admits of justification does not mean that the same is true of the state or belief thus arrived at, and hence does not mean that such a belief does not constitute knowledge.

Exactly what form such justification might take, and just how this conception of observation fits into a coherence theory of justification to preserve something like an empiricist view of knowledge, will be my topics in the next section. In the meantime, I want to devote the balance of the present section to drawing out some of the implications of the present view of observation for two of the issues in the dialectical vicinity of the concept of observation, viz. (a) the problem of observational error or illusion, and (b) the status of the observational/theoretical distinction in the philosophy of science.

(a) Given this account of observation, I submit that the problem of observational or perceptual error ceases to be very troublesome. In brief, what happens in a case of such error or illusion, assuming that one's basic learning of the concepts in question is not at fault (here it does not seem that we would speak of observational error), is one of two things: either (i) the stimulus is in some way non-standard due to something abnormal or unusual about the conditions, so that the normal conditioned response which would be correct and appropriate (i.e. true) in normal or standard conditions is incorrect in the actual situation; or (ii) something about the sensory apparatus and/or the mind-brain of the observer is abnormal, so that a normal stimulus results in a deviant and incorrect response. But in both sorts of case, the familiar causal explanations of the resulting error can be invoked. The important point is that since observation is not a cognitive process, though it has a cognitive result, only a causal explanation is necessary; there is no need to postulate anything beyond the sensory process and the conceptual response (e.g. a "sense-datum"), so that the traditional "argument from illusion" collapses.

(b) The basic idea underlying the observational/theoretical distinction is the idea that it is possible to draw a sharp distinction between two classes of scientific terms (or, alternatively, concepts): (i) observational terms, which can be applied to the world "directly" via observation, and (ii) theoretical terms which cannot be thus "directly" applied, and hence which can be applied to the world only via some form of inference. The conception of such a distinction, given roughly verificationist views about meaning, quickly leads to doubts about the meaning rulness of theoretical terms and about the reality of the various sorts of esoteric entities to which such terms purport to refer. And once such doubts have arisen, the step to various views of a broadly "fictionalist" sort, i.e. to views like instrumentalism which deny that theoretical entities are real in the full-blooded sense in which ordinary macro objects are real, is almost immediate.

I want to argue that once the Given is abandoned, the idea of a firm observational/theoretical distinction, which is clearly the foundation of all such fictionalist views, must be abandoned too. On the account of observation developed above, the distinction between what one can observe and what one cannot, i.e. between those terms which can appear in observation reports and those which cannot, will depend on what one has been trained to report, and thus will vary quite radically from person to person and from time to time. Such extreme relativity and variance would seem to completely destroy any ontological significance which the distinction might otherwise be thought to have. Surely what is real and what is only fictional does not depend on what people have been trained to report.

A fictionalist might want to accept this argument but still attempt to salvage his position by reformulating the observational/theoretical distinction as a distinction between the sorts of terms which people can in principle be trained to apply observationally and those which they cannot be trained to thus apply, while conceding that few if any people are actually trained to apply all the terms which they could be trained to apply. Thus the claim would be that although there are wide variances with respect to what particular people can actually observe, there are still some things which they could not observe, no matter what training they might be given. On our account of observation, the reason for this (physical) impossibility would be simply that some sorts of things, e.g. magnetic fields, simply have no direct effects on human sense organs which could provide the causal basis for an observation report. And thus the rationalist worries would still apply to these things, which turn out in fact to be roughly the same class of things about which he was worried all along.

There are, however, two related objections to such a revised observational/theoretical distinction and to the revised fictionalism which goes with it. In the first place, such a distinction clearly depends on the more or less accidental fact that human beings have just the sense organs which they do. If we had fewer or additional or simply different sense organs, the revised observational/theoretical distinction would have to be drawn at a different, in many cases markedly different, place, and hence different classes of things would be held to be real on the one hand and fictional on the other. But surely such an alteration in our sensory equipment, which undoubtedly has occurred in the past and may very well occur again in the future, through the vicissitudes of evolution, has no genuine ontological significance, and thus a view which would attribute such significance to it must be wrong. Things do not pop into and out of existence simply because human sensory faculties change. (If everyone became, through some mutation, congenitally blind, would colors thereby cease to exist?)

In the second place, it is arguable that such modifications in human sensory capacities have already occurred in our lifetime, and are likely to occur again in the near future. The sort of thing which I have in mind is what are often described as "instrumental observations," such as e.g. the case of the trained geologist who holds a rock up to a geiger counter, hears a distinctive clicking sound, and says, "Lo, radioactivity." Once one realizes that conscious inference need not be, though it no doubt often is, present in such a case, one cannot fail to be struck by the fact that those cases where inference is not present exactly satisfy our characterization of an observation report as a non-inferential belief causally conditioned to a sensory stimulus. Why then not count such a report as observational in exactly the same sense as reports of a more mundane sort? And such a move would of course immediately grant full-blooded reality to many entities upon whose fictional status the fictionalist has wanted to insist, plus making the fictional status of the rest totally dependent on the advances of technology, surely an inadequate basis for an ontological claim of this sort.

It would seem that the only way out for the fictionalist is to insist that all cases of instrumental observation are, despite appearances to the contrary, inferential. But it is hard to see what reasons could be given for such a view, without either begging the question at issue or appealing to the Given. It is still true, of course, that in the ordinary case the use of the geiger counter depends on or is causally mediated by the sense of hearing, but I cannot see that this fact is either very important or very inescapable. There is no reason to doubt that even this causal, though not cognitive, dependence could not in principle be avoided by hooking the geiger counter directly up to the central nervous system in some way, thus resulting in what might be called an artificial sense organ. To deny this possibility or to deny that what resulted from it would count as observation, would seem once again to place an unbearable weight on the mere fact that human beings happen to have just the sense organs which they do. (Note too that if the recipient of such an "instrument transplant" could be trained to reliably report radioactivity, it would seem to make absolutely no difference whether or not he was introspectively aware of any distinctive "experience" when he made such reports.)

I conclude that the conception of observation which results from the abandonment of the Given destroys the possibility of an observational/theoretical distinction with any degree of ontological force, and thus opens the door to a scientific realism in which theoretical entities like electrons are regarded as being at least as real as ordinary entities like tables. (Somewhat more will be said about the exact nature of the claim made by such a scientific realism at the end of the third chapter.)

7. Observation and Justification

The account of observation which emerged from the last section is as follows. Observation involves a conditioned, non-inferential conceptual response to sensory stimuli. Such a response, which we have agreed to call a 'report' whether or not it is overtly expressed, though it is an act, is not an action; it is no more voluntary than any other conditioned response. It is mediated causally, but not cognitively or epistemically. To say that it is conceptual is to characterize it as involving dispositions to make various inferences of an appropriate sort.

In the present section, I want to attempt to fit this view of observation into the coherence theory of justification whose general outlines were sketched in section one. This will involve two distinct though related tasks: (a) that of showing how the propositional claim which results from an act of observation can be justified without an appeal to the Given, and (b) that of showing how observation, thus construed and justified, can provide an answer to the most basic objection to a coherence theory of justification, viz. the objection that such a view completely destroys all connection between justified belief and the way the world really is and completely eliminates any vestige of empiricism.

(a) How then is an observational claim to be justified? Consider for this purpose an even more simple example than that of the last section: I look at the pencil on the desk and simply report, "This is red." Clearly on any reasonable epistemological theory it must turn out, barring abnormal circumstances of some sort, that I know that the object thus referred to is red; and this, according to the criteria for knowledge given in the first section of chapter one, means that the claim that this is red must be justified. But how?

On the view of section one, this is a case of coherence in the small, so that what we want is an acceptable argument having the conclusion that this is red. It is interesting to ask in this connection whether someone else, a color-blind person say, could come to know that this is red by hearing my report, and if so how his knowledge would be justified. The following argument (A) suggests itself:

(i) BonJour reported that this is red.
(ii) Observation reports are usually reliable.
Therefore (probably):
(iii) This is red.

Note that the probability qualification is to be taken as qualifying the inference, rather than the conclusion. What one comes to know as the result of such an inference is not that this is probably red, but simply that this is red. I submit that this inference is an example of the general way in which observation claims can be justified and can thus be regarded as knowledge. For clearly this inference is available not only to our color-blind onlooker, but also to me, i.e. to the person who makes the report. So that I can also thus justify my own observation claim. And, in the absence of the Given, some such inference seems to be the only available mode of justification.

Several comments are necessary by way of qualification and elaboration. First. Note that premiss (ii) can be made more specific in ways which may increase the probability of the inference. We have seen that there is no inconsistency, on our present notion of observation, in supposing that any particular observation report or even any whole class of observation reports may be mistaken or unreliable. This means that if (ii) is replaced by a premiss which deals with a narrower class of observation reports, the probability implicitly represented by the term 'usually' may be increased. Thus instead of (ii), we might have:

(ii') Observation reports of the color of physical objects are usually reliable.

(ii'') BonJour's observation reports are usually reliable.

(ii''') BonJour's observation reports of the color of physical objects are usually reliable.

Presumably any of these would give a higher probability to the inference to (iii) than does (ii).

Second. If argument (A) is to provide genuine justification, the premisses themselves must be known, and hence justified, i.e. justifying arguments for them must be available. The arguments for (ii) and its variants would presumably be essentially inductive in character, appealing to known instances of correct observation reports of the sort specified. (And of course further arguments would have to be available to justify the claim that those other reports were in fact true.) The argument for (i) in the key case in which argument (A) is intended to supply my own justification for my observational claim would presumably appeal to my further (introspective) belief that the original belief and/or utterance was indeed a report, as in the following argument (B):

(iv) BonJour believes that he reported that this is red.
(v) BonJour's beliefs about his observation reports are usually reliable.

Therefore (probably):
(i) BonJour reported that this is red.

And of course (iv) and (v) would also have to be known, and hence justified, (v) again in an inductive manner, and (iv) with reference to a further introspective belief. Etc.

Third. Note carefully that it is not sufficient for an observational claim to represent knowledge that some premiss like (ii) in fact be true. The putative observer must also know that it true, for otherwise he would be unable to supply the requisite justifying argument, and hence his observational belief would be, for him, unjustified. And this means that a person cannot have any genuine observational knowledge until and unless he has a very great deal indeed [EPM 167-69]. This means that another person could conceivably come to know something on the strength of my observation report, even though the very same claim did not count as knowledge for me, because he knew the truth of an appropriate premiss of the same general form as (ii), while I did not.

Fourth. This account of the justification of observation also renders unnecessary a view which has achieved rather wide currency in recent years, to the effect that observation claims somehow do not require justification in order to constitute knowledge, or alternatively (which seems to me to be exactly the same thing) that they are somehow to be regarded as justified even though no evidence or justification even could be explicitly presented. Thus Goodman speaks of some statements having "initial credibility,"{23} and Danto of some claims to know being "justified though he who makes them is in possession of no evidence in their support."{24}

Such views seem to me to be clearly and obviously mistaken, though the seriousness of the mistake depends on exactly how they are construed. Taken literally they amount to a proposal either to so modify our criteria of knowledge as to sometimes count as knowledge propositional claims which lack justification altogether, or else to construe the notion of justification so as to count as justified propositional claims for which no evidence even could be offered by the claimant. It seems clear, however, that this can scarcely be what is intend by Goodman, Danto, and others. For to adopt such a view would immediately invite the obvious question: why accord such favored status to this propositional claim for which no justification can be offered and not to that one? This question cannot go unanswered, unless one is willing to grant the same favored status to all such propositional claims, thus in effect eliminating justification altogether as a criterion of knowledge and reducing the notion of knowledge to the notion of true belief. Thus the proponent of such a view must provide some criterion for separating out those propositional claims which cannot be explicitly justified but which are nonetheless to be taken as either justified or as not requiring justification, from those others which are not to be thus accepted. But clearly such a criterion cannot be purely arbitrary; there must, that is, be some reason for thinking that propositional claims which meet it are more worthy of acceptance than those which do not. And if such a reason is to be relevant to the question of knowledge (rather than religious faith, patriotic commitment, etc.), it seems that it would have to be a reason for thinking that propositions which meet the criterion in question are more likely to be true than those which do not. And this suggests that the fact that a particular propositional claim, that-P, satisfied an appropriate criterion, together with the reason which makes that criterion an epistemically relevant and acceptable criterion, might be better construed simply as providing a justification for the claim that-P, rather than as providing a reason for accepting the claim that-P even though it cannot be justified.

What I am suggesting is that the all-too-common view that some propositional claims are or have to be accepted even though the claimant can provide no justification for them even if challenged, where it does not amount to the extreme view that such claims are to be accepted for no reason at all, rests either on (i) the mistaken view, already discussed, that a propositional claim is justified only if its justification has actually been (rather than merely could be) explicitly provided, which leads to the regress argument; or (ii) on an inadequate conception of justification, probably a carry-over from the Given. For once it is realized that for a propositional claim to be justified it is not necessary that explicit justification actually be provided, and also that any acceptable argument from known premisses having the conclusion that-P constitutes at least a prima facie justification of the propositional claim that-P, it is hard to see how one could have a good reason for accepting the claim that-P without explicit justification, unless he also has the materials available for constructing a justifying argument for the claim that-P. And this is in fact the strategy of the above account of the justification of observation. Thus the paradoxical view that observation claims do not require justification seems both mistaken and unnecessary.

(b) I have now provided, I hope, at least a schematic idea of how an observation claim can be justified by considerations of coherence. My remaining task in this section is to give some account of how observation, thus construed, can provide an empirical constraint upon justified belief, thus showing that a coherence theory of justification need not deprive justified belief of all relation to the world, as the earlier objection claimed. The way to do this, I think, is to consider an empirical hypothesis and see just how it could be confirmed or refuted by observation, within a coherence framework. Thus consider the hypothesis that there is a red pencil on the desk. How could this be tested observationally? The answer turns on two preliminary considerations.

In the first place, it must be noted that the overall coherent account of the world will, as we noted earlier, involve an account of the behavior of human knowers like myself, including in part an account of how their observation reports vary with their environment. Thus, e.g., in the present case the total account would very likely include some such proposition as the following:


If there is a red pencil on the desk, and if BonJour is standing before the desk, and if his eyes are open, and if his "mental set" is of the right sort, and if there is a red pencil on the desk, then BonJour will report that there is a red pencil on the desk.

This is, of course, a highly complex law, presumably derived from various simpler laws about my behavior and about the behavior of human observers generally.

Secondly, it must be understood that not all human conceptual activity is narrowly cognitive in character, and in particular that some is practical, i.e. concerned with desires, intentions, and actions. Moreover, such practical reasoning will obviously link up with the results of more theoretical reasoning; the latter will provide premisses for the former. Thus, in the present case, a desire on my part to test the hypothesis that there is a red pencil on the desk would presumably link up with my knowledge of generalization (beta) to produce an intention to bring it about that I am in fact standing before the desk, etc. And this intention would, ceteris paribus, result in my actually acting so as to bring about this result. (Sellars' account of intentional action, which I cannot discuss in detail here, is that the action is a conditioned response to the intention, in a way exactly analogous to the way in which an observation report is a conditioned response to a sensory stimulus. To learn observation language is, in part, to be disposed to make reports; to learn the language of intention is, in part, to be conditioned to act in accord with one's intentions.{25} And of course I can also know that I have thus acted, either by knowing that I usually do what I intend to do, or by means or ordinary observation, or both.

Thus my desire to test the hypothesis that there is a red pencil on the desk, combined with my knowledge of generalization (beta), will normally result in my in fact coming to satisfy all of the antecedent of (beta) but the part which specifies that there really is a red pencil on the desk. At that point one of two things will occur. Either (i) I will actually report that there is a red pencil on the desk, or (ii) I will report that there is not a red pencil on the desk. In either case, I will also have an argument available, along the lines suggested in the first half of the present section, to justify the resulting propositional claim. In case (i), I thus come to know that there is a red pencil on the desk, and the original hypothesis is confirmed. Case (ii) is rather more interesting, for here I have a conflict. Either the claim that there is a red pencil on the desk, or the claim that there is not must be abandoned, in the interests of consistency. To abandon the latter would require some modification in at least the probabilities attached to one or more of the generalizations which support it. If the original hypothesis is also supported by further arguments and generalization, abandoning it will force similar sorts of modifications. It is, of course, quite impossible to say what actually would or should be done in the absence of many further details about the situation, though in most cases the observational claim would presumably win out. Clearly the criteria for deciding which revisions are to be made will involve considerations of coherence and simplicity. The way in which such considerations operate is, or course, not at all well understood, and I do not pretend to have clarified it to any very substantial degree. My point is only to show how something very much like empirical testing and confirmation can be accommodated within the broad framework of a coherence theory of justification. The key point is that such empirical testing operates through the ties of coherence, rather than being a factor distinct from them.

Thus, in more general terms, empirical hypotheses are tested by deriving consequences from them which, when combined with established generalizations relating to human action and behavior generally, have implications in terms of human observation reports. Such implications can be checked by putting oneself in a position to make such reports and then seeing if they actually ensue, There is, of course, no certainty at any point, and in the case of conflict one may doubt either the hypothesis being tested, or something about one's own actions or reports, or the laws linking the two. But such testing still provides a clear empirical constraint upon the total body of one's beliefs, and in most cases considerations of coherence will dictate pretty clearly just what sort of revision in one's total account of the world is indicated to satisfy that constraint. Thus the view that our coherence theory of justification would have to mean that one could make up any arbitrary system of beliefs without any possibility of empirical testing is to be rejected.

I am now also in a position to say a bit more in answer to the objection, discussed toward the end of section five, that a coherence theory of justification could not choose between at the very least a quite large number of alternative coherent accounts of the world. To begin with, notice that the coherent accounts in question must, to be candidates for human knowledge, satisfy the following conditions. (i) They must contain sub-accounts of the behavior of human observers, including generalizations about the reliability of various varieties of observation, of the sort cited in the preceding discussion of observation. (ii) Such accounts must credit at least some modes of observation with sufficient reliability so as to not be easily overridden by conflicting claims. (iii) Such knowledge about the behavior of human observers must link up with the concepts of intentional action in such a way as to make possible the sort of quasi-empirical testing just discussed. Now any coherent account of the world which satisfies these three conditions will have the following interesting feature: if believed and applied it will tend to generate further propositional claims via the mechanism of observation already described. And only a body of propositional claims which not only starts out coherent, but also remains coherent when these new propositions are added (while continuing also to satisfy (i), (ii), and ( iii)), will be finally justified from our coherence point of view.

Thus the answer to the objection is, first, that there is no very good reason for thinking that there will be very many such coherent systems, or even that in the long run, for a given person, there will be more than one. Certainly a coherent account of the world constructed arbitrarily by a science-fiction writer, the example most commonly employed in connection with this objection, will be most unlikely to satisfy (i)-(iii), or to remain coherent when applied if it does satisfy them or is expanded so as to come to satisfy them. And, secondly, if there were in fact more than one such account which survived even in the long run, it is not at all clear why the same alternative account could not survive on any theory of justification.

One point which emerges from the last two paragraphs is that my talk throughout most of this chapter of justifying bodies of propositions is a serious, though perhaps initially pardonable, oversimplification. For if the answer just sketched to the objection just sketched about alternative coherent accounts is to work, it is necessary that the accounts of the world in question be actually believed and applied. And this means that it is really bodies of beliefs, rather than bodies of propositions, which are finally justified.{26} In this connection it is interesting to note that our conditions (i)-(iii) could be plausibly viewed as necessary conditions for an account of the world to be believable, i.e. for anything to count as believing it. If they are not satisfied, it is hard to see that there would be any significant difference between believing such an account and merely approving of it in some non-epistemic sense. This is to say that belief must have some potential influence on action.

Are conditions (i)-(iii) then to be regarded as further criteria of justification, thus contradicting my earlier claim that coherence is the sole such criterion? I cannot see that very much hangs on this question. I am inclined, however, to say that they are at least not criteria on the same footing as coherence. Rather it seems plausible to regard them, not as criteria for justification, but as necessary conditions for being the sort of account of the world for which the question of justification might become appropriate; the question would in fact become appropriate when such an account came to be believed. In this connection it is interesting to note that (i), (ii), and (iii), taken together, bear an interesting resemblance to the old verifiability criterion of empirical meaningfulness; in light of our construal of empirical testing in a coherence theory of justification, they amount to the requirement that any candidate for being an empirical account of the world must be capable of being empirically tested. And the justification for such a requirement would then be simply that only such an account could be believed, in any interesting sense of 'believe'.

It is worthwhile noticing how similar this account of justification is to more traditional views, and how little is really lost when the Given is abandoned. For although on the present account it is no longer possible to decisively confirm or refute an empirical hypothesis by confrontation with "experience," because all that observation can ever do is force one to make one or another of a quite large number of alternative possible modifications in one's account of the world, something very much like this was also true on the traditional view, even in Lewis' account, for any proposition not couched in expressive language. And propositions in expressive language are not the ones we are ordinarily concerned about.

Thus we may sum up our account of justification as follows. The only mode of justification for human knowledge is coherence. Individual beliefs are justified if it is possible to construct an acceptable argument for them from known premisses, subject to the demand for consistency. Whole frameworks of beliefs are justified by their degree of internal or explanatory coherence. Observation is indeed, as we suggested in the first section of chapter one, the key connection between language or thought and the world, but this fact does not mean that the justification afforded by coherence is to be supplemented by Givenness. Rather observation claims are themselves justified by considerations of coherence, and provide an empirical constraint upon the overall structure of knowledge only by operating through the bonds of coherence. This means that justification is ultimately holistic and temporally relative in character. A framework of beliefs comprising an account of the world is justified (for a given person) at a given time if and only if it is the most coherent total account of the world available (to that person) at that time. A particular belief, that-P, is justified at a given( time if and only if it is a part of the most coherent total account of the world available at that time. But there is no reason to suppose at any such time that a more coherent account than any then available cannot be found, and in the search for such an account nothing is inviolable, including most emphatically the very conceptual system itself.

[Table of Contents] - - [Go to Chapter 3]


{12| Especially in EPM and RLG. Cf. especially pp. 167-170 of EPM, and pp. 328-330, 333-335, and 356 of RLG. [Back]

{13} Cf., as a recent example, A. C. Danto, Analytical Philosophy of Knowledge (Cambridge, England: Cambridge University Press, 1968), pp. 26-28. [Back]

{14} Israel Scheffler, Science and Subjectivity (New York: Bobbs-Merrill, 1967), p. 92. The assumption that coherence in the large demands only consistency seems to have been held by both Neurath and Schlick, thus undermining their famous debate. Cf. Neurath's "Protocol Sentences" and Schlick's "The Foundations of Knowledge," both reprinted in Logical Positivism, ed. A. J. Ayer (New York: Free Press, 1959). [Back]

{15} Cf., e.g., Paul Oppenheim and Hilary Putnam, "Unity of Science as a Working Hypothesis," in Minnesota Studies in the Philosophy of Science, vol II, ed. Herbert Feigl, Michael Scriven, and Grover Maxwell (Minneapolis: University of Minnesota Press, 1958). [Back]

{16} Ibid., p. 4. [Back]

{17} Cf., e.g., the title essay in Carl G. Hempel, Aspects of Scientific Explanation (New York: Free Press, 1965). [Back]

{18} Ibid. [Back]

{19} Carl G. Hempel, Philosophy of Natural Science (Englewood Cliffs, New Jersey: Prentice Hall, 1967), p. 83. [Back]

{20} Cf. Ibid., pp. 83-4. [Back]

{21} Gilbert H. Harman, "The Inference to the Best Explanation," Philosophical Review 75 (1966), pp. 241-47. [Back]

{22} Sellars, in chapter one of SM, seems to be arguing that sense impressions are still present, even if they play no epistemological role. [Back]

{23} Nelson Goodman, "Sense and Certainty," Philosophical Review 49 (1952), pp. 162-63. [Back]

{24} Danto, op. cit., p. 28. [Back]

{25} See RLG 329-330, 350-51. Also "Thought and Action," in Freedom and Determinism, ed. Keith Lehrer (New York: Random House, 1967). [Back]

{26} I owe this point to Gilbert Harman (though he may not agree with what I say about it). [Back]

[Table of Contents] - - [Go to Chapter 3]