Part VIII [32]-[38] : Does Empirical Knowledge Have a Foundation?
Section 32: Another incorrect, foundationalist account is described here. Sellars only disagrees with one bit of this story, though it turns out to be an important bit. Foundationalism is the claim that there is a structure of particular beliefs such that:
Each one is noninferentially arrived at.
The beliefs in (1) presuppose no other belief, either particular or general.
These noninferentially acquired beliefs constitute the ultimate court of appeal for all factual claims.
Sellars accepts (1) and (3), but denies (2). His project at this point is to show how a bit of knowledge (belief) can, and indeed how all of it does, presuppose other knowledge (belief), even though it is not inferred from that other knowledge or belief. This possibility was not seriously examined by the classical epistemological tradition. It is a certain hierarchical picture of understanding (at this level a necessary condition of believing) that Sellars rejects. He does not object to a hierarchical picture of justification, once that has been suitably disentangled from bad foundationalism concerning the nature and acquisition of belief.
For Sellars, there is no such thing as a noninferential belief, if by that one means a belief one could have without grasping its inferential connection to at least some other beliefs. For to understand a sentence, to grasp a propositional content (a necessary condition of having a belief) is to place it in the space of reasons, to assign it an inferential role in the game of giving and asking for reasons, as entailing some other contents and being incompatible with others. A noninferential report or belief can properly be called ‘noninferential’ only in the sense that the reporter’s commitment to an essentially inferentially articulated content is elicited noninferentially on this occasion—that is, that it is elicited as a response to some nonlinguistic, nonepistemic environing circumstance, rather than as a response to another belief or assertion. Noninferential beliefs do not form an autonomous discursive stratum: there could be no language game consisting entirely of noninferential reports. (Notice that this is a stronger claim than that made above in connection with ‘looks’-talk. For this claim concerns any kind of noninferential report, whether what they report is inner or outer, appearances or empirical realities.) For any sentence to have noninferential uses, some sentences must have inferential ones. For the conceptual content expressed by a sentence (what is believed) essentially involves its potential as a premise and as a conclusion of inferences. Unless one can employ noninferentially acquired beliefs as the premises of inferences leading to further beliefs, their acquisition does not qualify as acquiring beliefs (something propositionally contentful) at all. On this inferentialist picture of conceptual content, one cannot have one concept without having many inferentially interrelated ones. This does not mean that there could not be a language consisting only of expressions for observables, however. For the concepts of observables are concepts that have noninferential, reporting, uses. The requirement is only that the concepts that can be used to make noninferential reports must also be available to be applied inferentially, as the conclusions of inferences whose premises are the noninferential applications of other concepts.
Sellars begins by asking about the nature of the authority (a patently normative notion) of noninferential beliefs, that is, their capacity to justify other claims. A distinction is needed first between sentence types and sentence tokens: the type is repeatable and can be instantiated on different occasions, while the token is unrepeatable. It is the utterance or inscription of the sentence on a particular occasion. So if the distinction is applied to letters instead of sentences, the sequence ‘aeaaeea’ contains two letter types and seven letter tokens, four of one type and three of the other. Now it can be seen that it is sentence tokens whose justification is at issue. For while there are some sentences that are justified, if they are justified at all, whenever they are tokened, such as ‘2+2=4,’ and ‘Red is a color,’ there are others that can be justified (and true) on one occasion and not justified or true on another. Then only the tokens and not the types can be said to be justified. These are sentences like ‘That car is red,’ or ‘I'm hungry now,’ which contain words whose reference is determined by the actual circumstances in which the sentence is tokened. These are called “token reflexive” expressions. Many, though not all, of the noninferential beliefs putatively described by (1)-(3) above are token reflexive. Authority or credibility (positive justification status) is either extrinsic, coming from something else, in this case by inferential inheritance, or intrinsic. Intrinsic credibility may be associated with types, as in meaning-analytic statements such as ‘All bachelors are unmarried males,’ or with tokens, as in ‘This is red,’ (or, given Sellars' account of 'looks', ‘This looks red’).
Section 33: Sellars now considers a line of thought according to which intrinsically credible types and intrinsically credible tokens, analytic claims[1] and observation reports, are similar in that they are both types such that their being correctly tokened, that is tokened according to the rules for the use of all the component expressions, is a sufficient, not just a necessary condition of their being true and justified (hence not just believed but known). Sellars can swallow all of this except the bit about rules. The idea he will reject is that analytic statements are true by virtue of discursive definition (definition of a linguistic expression in terms of other linguistic expressions), while observation reports are true by virtue of ostensive definitions. Ostensive definitions are the only sort we can give of terms like 'red'. They consist of defining the expression by exhibiting samples of the things it applies to (pointing to red objects). The usual foundationalist infinite regress argument can be applied to show that not all expressions of the language can be discursively defined on pain of circularity or infinite regress (in either case no definition is achieved). So there must be ostensive definitions in the language. These definitions, just like the discursive ones, codify the rules of appropriate usage of the expressions they define. Just as following those rules is sufficient for the truth of analytic statements, so following the 'rules' of ostensive definition is to be sufficient for the truth of observation reports. (Such a rule might look like the definition in [13].) At this point Sellars disagrees. One can imagine following the rules for the use of 'this', 'is', and 'green' only if one has some idea of prelinguistic awareness of green -- the Myth. For Sellars it is incoherent to talk of ostensive definitions setting up rules for using 'green', for there is no language available in which such rules could be stated. Ostensive definitions establish practices; they are regular, but not rule governed.[2]
Section 34: The notion that the authority of noninferential reports rests on episodes of nonverbal hence nonconceptual awareness, which verbal performances express, is a version of the Myth. From Sellars’ point of view, such episodes are the tortoise underneath the elephant.
Section 35: Here Sellars presents his alternative view. It begins with the observation that " ... a token of 'This is green' in the presence of a green item...expresses observational knowledge [only if] it is a manifestation of a tendency to produce tokens of 'This is green'-- given a certain 'set' (context of collateral commitments and circumstances) if and only if a green object is being looked at in standard conditions...". That is, it must be the expression of a reliable differential responsive disposition. But photocells and parrots could satisfy this condition, which shows that so far only the responsive dispositions part of the observation report has been specified. It remains to add conditions to capture the epistemic side, the dimension of endorsement, of undertaking inferentially articulated commitments, of producing a performance with a distinctive kind of authority.
To have the authority of knowledge, the report must not only be reliable, it must be taken to be reliable. In fact Sellars claims that it must be known by the reporter to be reliable (and in this he perhaps goes too far): "...the perceiver must know that tokens of 'This is green' are symptoms of the presence of green objects in conditions which are standard for visual perception." ‘Justification’ has the 'ing/ed' ambiguity (cf. [24]): justifying, a practical activity, or being justified, a normative status. Sellars claims that one cannot have the status except when it is possible to redeem that claim to authority and epistemic privilege by engaging in the activity of justifying it. This claim of the priority of practice over status is a specific variety of pragmatism, to which Sellars adheres. The difference between a noninferential reporter and a photocell, or a parrot trained to utter ‘It's getting warmer,’ as the temperature rises, does not lie in the reliability or range of their responsive dispositions. It lies in the capacity of the reporter to redeem the commitment undertaken, the authority claimed by the reporting, by justifying the claim (if challenged) by giving reasons for it. The by now familiar basic point is that in order to count as making a claim (expressing a belief) at all, the reporter must be "in the space of giving and asking for reasons", in addition to having the right responsive dispositions. The further claim being forwarded here is that for a noninferential report to express knowledge (or the belief it expresses to constitute knowledge), the reporter must be able to justify it, by exhibiting reasons for it. This is to say that the reporter must be able to exhibit it as the conclusion of an inference, even though that is not how the commitment originally came about.
The inference in question is what might be called a “reliability inference.” One justifies a noninferentially elicited report that something is red by noting that one was disposed noninferentially to apply the concept red to it, and pointing out that one is a reliable reporter of red things in these circumstances. To say that one is reliable is just to say that the inference from one’s being disposed to call something red to its actually being red is a good one. Thus the reliability of one’s differential responsive dispositions, together with the report’s being an exercise of those dispositions together justifies—offers good reasons for—the report. In insisting that in order properly to be credited with knowledge a reporter must be able to offer an inferential justification of the belief in question, Sellars is endorsing an epistemological internalism that puts him at odds with more recent reliabilist externalists in epistemology. Their claim is that the real function of the traditional justification condition on knowledge is to rule out accidentally true beliefs. If so, then the rationale for engaging in assessments of whether various beliefs qualify as knowledge is perfectly well-served by insisting only that candidate beliefs result from reliable belief-forming mechanisms—that is, mechanisms that are likely to lead to truths, whether or not the reporter knows that they are. Forming beliefs one can justify then appears as one reliable mechanism among others.
Of course, from Sellars’ point of view it would be a mistake to conclude from this line of thought that one could trade inferential justification for reliable belief-formation in a wholesale fashion. For that it is beliefs one is forming, that what one is doing is applying concepts, is a matter of their specifically inferential articulation—their role in the game of giving and asking for reasons, justifying and demanding justifications. Against that background of inferential practice, however, it is not obvious why Sellars should resist the reliabilist’s suggestion. Why isn’t it enough that the attributor of knowledge know that the reporter is reliable, that the attributor of knowledge endorse the inference from the reporter’s responsive disposition noninferentially to apply the concept red to the thing’s (probably) being red? Why should the reporter herself have to be able to offer the inferential justification for her noninferential report? (This is the thought behind the qualification offered parenthetically early in the second paragraph above.)
Section 36: The moral is that on the true view "one could not have observational knowledge of any fact unless one knew many other things as well". This is not to say that observation reports are somehow the results of inferences after all, but only that, though noninferential, they must be justifiable to be justified. The false view thinks it is supposed to give a causal description of how knowledge is possessed, but: "...in characterizing an episode or state as that of knowing we are not giving an empirical description of that episode or state: we are placing it in the logical space of reasons, or justifying and being able to justify what one says." Thus everything irrelevant to justification, either to knowing what would be a justification or to being entitled to produce one, is a noncognitive causal antecedent, perhaps a necessary condition of empirical knowledge, but not one that is constitutive of it. Nor is the general point is not specific to the normative, epistemic status of knowledge—though Sellars does not point this out. He could as well have said that in characterizing an episode or state as one of believing, or applying concepts, or grasping propositional contents we are not giving an empirical description of that episode or state but placing it in the logical space of reasons, or justifying and being able to justify what one says. For only what is inferentially articulated is conceptually contentful (and hence qualifies as a believable or claimable) at all. As we saw in the previous section, however, Sellars does wants to insist further that one cannot know noninferentially that something is green unless one also knows that one is a reliable reporter of green things under the prevailing circumstances.
Section 37: This view—the one Sellars endorses—seems to involve an infinite regress. For how could we have acquired knowledge that tokens of ‘This is green,’ are reliable symptoms... unless we had had knowledge of such facts as ‘This is green,’ and "This is a token of 'this is green'" beforehand?[3] Sellars' answer is that we can acquire knowledge of facts of these three types simultaneously, but that we can know facts that bear on events that occurred before we acquired any of this knowledge. Thus: "...it requires only that it is correct to say that Jones now knows, thus remembers, that these particular facts did obtain. It does not require that it be correct to say that at the time these facts did obtain he then knew them to obtain. And the regress disappears." Thus children at the age of six can know that at four they saw—in the sense of reliably responded to—a fire, although at the age of four all they could do was say 'fire' parrot-fashion, without knowing there was a fire.[4] The important difference is not one of responsive disposition but of capacity to endorse. The six year old has moved into the space of giving and asking for reasons; he can commit himself to a claim and be treated as authoritative; he is responsible for the claim he undertakes. For this he must at least be able to tell what he is thereby committing himself to and what evidence would entitle him to it, that is, he must understand his claim. But even that is not sufficient. For this new normative status is socially conferred. No nonepistemic description of the candidate reporter suffices for the conferral of this status, unless and only insofar as the community conferring that status, treating the individual as responsible, reliable, and so on, takes it to be sufficient. Compare achieving one's majority and being able for the first time to undertake contractual obligations. This status consists in the community's recognition of it. Some minors are more reliable at carrying out the commitments they undertake than many over the age of 21, but this fact does not make their signature mean that they have entered into a contract. This is how “the light dawns slowly over the whole”: at some point one masters the moves, inferential and noninferential, sufficiently that one’s noises come to be taken by one’s community as having the significance of making claims, undertaking commitments, giving reasons.
Section 38: The only sense in which there is no foundation for empirical knowledge is the sense in which the observation reports, which in a certain sense are its foundation, themselves rest (not inferentially, but in the order of understanding and sometimes of justification) on other sorts of knowledge. Observation reports, whether of inner episodes or outer happenings, do not constitute an autonomous stratum of the language—a game one could master though one had as yet not mastered the inferential use of any expressions. That is, Sellars rejects only claim (2) of the three foundationalist theses considered in [32]. But there is no need for a foundation in this sense: "Empirical knowledge is rational, not because it has a foundation but because it is a self-correcting enterprise which can put any claim in jeopardy, though not all at once."
(The book can be bought here.)
[1] In “Two Dogmas of Empiricism” Quine objects to the notion of meaning-analytic claims (claims true in virtue solely of the meanings of their words) on the broadly pragmatist grounds that there is no practically discernible status corresponding to this supposed category. Claims taken to be analytic, such as “All bachelors are unmarried males,” are not immune from revision, known a priori, or otherwise distinguished from statements of very general fact, such as “There have been black dogs.” Sellars accepts analyticity, which he associates with the practical status of counterfactual robustness. This line of thought ties our concepts to what we take to be laws of nature. (See Sellars’ “Concepts as Involving Laws, and Inconceivable Without Them,” in PPPW, and “Counterfactuals, Dispositions, and Causal Modalities” [ref.].) So conceived, analytic claims are neither immune from revision, nor known a priori.
[2] This distinction, and the need for some sense in which a practice (paradigmatically, a linguistic practice) can be governed by norms even though its practitioners cannot be said to be following rules, is Sellars’ topic in his important essay “Some Reflections on Language Games”.
[3] Notice that this is a problem Sellars need not have faced, had he endorsed the modified externalism offered to him in [[35]] above.
[4] Commenting on this point for the 1963 edition, Sellars said that his thought was that one could have direct (in the sense of noninferential) knowledge of a past fact which one could not conceptualize at the time that it occurred.