Foundationalism Has No Basis

The pursuit of the incorrigible is one of the most venerable bugbears in the history of philosophy. 104

A bugbear is something of which we are afraid for no reason. No one has ever actually been afraid of the incorrigible: the fear is that we might not know anything at all, since there might not be any incorrigible knowledge.

Knowledge has no foundation. In particular, empirical knowledge has no foundation and observational statements don't form or need a foundation. Despite what follows from the idea that observational statements about sense data form a foundation for knowledge of material objects, many observational statements about material objects are not based on evidence and need no verification, and thus are not based on the evidence of sense data. Conversely, observational statements about sense data can require evidence, and hence are not suitable to form an incorrigible foundation, ….

There is a picture of knowledge that goes back at least to Plato according to which most of whatever we know, which I will, somewhat arbitrarily take to be the statements we know, are based on other things we know ("by inference"), and so on, and since we know them, the process must stop (actually begin) somewhere, and so philosophers have been looking on and off for over 2000 years for the statements we know that we begin with, the ones that are not based on others. How can a statement be known if it is not based on others? It must be, in some sense, epistemically virtuous on its own. It might be self-checking (Descartes) or incorrigible (Ayer). I will sometimes call the bottom-level statements basic statements, since they are the ones that form the basis for the others, to avoid commitment to one version or another of what makes them suitable to serve as a basis (incorrigible, self-checking).

The picture is that knowledge is like a house that needs a foundation to rest on.

Descartes, for example, had two primary basic statements: "I think, therefore I am," and "God is not a deceiver." (That indicates the idea. I'm not making any claims about the appropriate interpretation of Descartes.) Carnap just says that in different fields of knowledge we may take different sets of sentences as basic, and his examples include reports of observations of medium-sized material objects and reports of experiences of sense data. Ayer shares Carnap's analysis of the main kinds of possibilities. Of course, in Foundations of Empirical Knowledge his goal is to show that sense data reports can serve as a basis for statements about material objects.

The problem is that the basic picture is wrong. There is no basic kind of sentence at all. What we know, and how we know it depends heavily on circumstances, and one and the same sentence can sometimes be known based on other sentences and sometimes be known in a way that is not based on sentences at all. In addition, the way sentences are known based on other sentences is rarely inferential in the sense in which mathematicians draw inferences from assumptions or axioms, and that limited kind of basing sentences on others is the only one Ayer and Austin, and many others for 2,000 years, consider.

A sentence is not the same the thing as a statement. Sentences are used to make statements, and one and the same sentence can be used to make different statements in different circumstances. The most trivial examples involve pronouns: no two uses of "I'm here now" make the same statement. What I said in the last paragraph needs some cleaning up: "one and the same sentence can sometimes be known based on other sentences and sometimes be known in a way that is not based on sentences at all" makes no sense, since, if, as is claimed, what is known is a sentence, either the sentence is known or not. What is actually known (in my version) is a statement, and so I should have said, "one and the same sentence can be used to make statements some of which are know based on other statements and some of which are know in a way that is not based on statements at all." It is nonetheless useful to talk about how sentences are known since one of the big mistakes Ayer, Carnap, Descartes, and the rest make is that one can tell from a sentence how it could be known. That isn't true. One also needs to know the particulars of how the sentence is being used, that is, what statement it is being used to make. One of the traps foundationalists fall into is considering uses of sentences only in very special circumstances when investigating how they are know, and presuming that there is only one way in which they could be known.

A story Suppose you want to know what is in a room, but you can't go in. There is a robot in the room that can transmit to you. It tells you that there is a red patch in such a position relative to it, that when it moves a certain way, it is blocked, and lots of stuff like that. You need to figure out what is there. That would require almost exactly the kind of inference Ayer claims we make in going from sense data language to material object language, but, of course, it is completely unlike anything we ordinarily do. We might do something like it in an extremely unfamiliar environment in which we have learned that we frequently get things wrong. That situation is the one the argument from illusion has convinced Ayer that we are normally in.

As Wittgenstein (Austin's teacher) emphasized, we learn to use most words by seeing examples of how they are used, and apply them in new cases by analogy. Many words do not have a core fixed meaning that can be formulated as a set of criteria, because they get stretched in various directions in ordinary use. Wittgenstein uses the analogy of family resemblance and the example of "game" (board games, solitaire, playing catch, war games).That means that using inference in a quasimathematical sense to characterize how we go from knowledge of some statements to knowledge of another is just not a rich enough picture of what happens. In addition, one and the same sentence may be appropriate or inappropriate for the same thing depending on circumstances. The characterizations of what statements are the basic statements always proceed by characterizing what sentences are used to make basic statements, as if sentences of a certain type can only be known in one way.

To say that sense data reports are incorrigible is crazy. We can come to know that a sense data report is true in many different ways. Ayer only argues that they are incorrigible when we come to know them in a very particular way.

If there is a foundation, then there must be a clear direction in which knowledge flows (from the bottom up). Of course, there are many different relevant senses of direction. Ayer, for example, takes it to flow from the particular to the general. But it goes both ways.

Austin gives the example of what pigs eat. You might know what pigs eat by knowing what some particular pigs eat. But you might equally well know what some particular pig eats by knowing what pigs eat.

It isn't clear that Austin's example works, because you might only be able to go down from a generalization after the generalization was acquired from particulars. The iguana example below shows that we do sometimes go down from generalizations not acquired from relevant particulars. Austin could make a case that certain terms are acquired with a particular use that includes general statements. We don't learn that bachelors are unmarried males by observing a lot of bachelors and generalizing, and there are many terms for which parts may be acquired by observation and others are just built in to the use. You can sit on chairs. (No observation is required.) Chairs have legs. (Observed.)

Lengths of iguanas. A team of biologists, studying iguanas in the Galapagos Islands, measured the lengths of many iguanas over a period of several years. It is hard to measure the length of an adult iguana (10 feet long, claws, and not cooperative). Sometimes one and the same adult iguana would be measured as having one length, sometimes another. It was just chalked up to inaccuracies of the measurement process, until someone noticed that the lengths were highly correlated with the amount of rainfall. The amount of rainfall influences the amount of food. Iguanas are now, as a result of the discovery of the correlation, the only vertebrates known to shrink in size as part of their normal growth patterns. The point for our purposes is that we only came to know the lengths of the iguanas in virtue of knowing a fairly obscure statistical correlation.

Both examples, and Austin's general point, are deeply related to Quine's argument against foundationalism. Quine notoriously argues that there is no distinction between analytic truths (truths by definition) and synthetic truths (empirical truths). Is it an analytic truth about pigs that they eat slops or is it an empirical truth?

Here is an example that is clearly correct, though too technical, due to Hartry Field:

Which of those is an empirical fact, and which a matter of definition? There is no answer to that question.

Ayer says that to be certain something is some kind of material object would require infinitely many verifications (which are, for him, always based on sense data). That is just wrong, and it shows his theory to be wrong. Austin uses the example of verifying that there is a telephone in the next room. How do you do it? You go in the next room, you look, you make a call, you have someone call you. You now know that there is a telephone in the room. No more is required. Telephones are not edible, but you don't have to fail to eat something to find out that it is a telephone. Indeed, Ayer's model of inference suggests that if you could eat something it wasn't a telephone, but the more flexible fact that terms are specified by family resemblance means that it is perfectly possible that someone could devise an edible phone. After all, there are plastic guns, and that would have seemed nearly as unlikely to Ayer or Austin.

In addition, we sometimes know that a statement is true, not based on other knowledge and without evidence: I live in Tucson. I'm typing on a computer. I haven't come to know either of those based on any evidence. (That is not to say that I cannot produce evidence that I live in Tucson.) Now, against Austin, it is very plausible to think that in cases like this I have something like evidence that was involved in how I came to know, that there is a reasonable way to stretch the word to apply to cases like that. Austin is happy to note that we ordinarily use words without clear definitions and that we adapt them to various situations, but when Ayer does exactly that, with, for example, "evidence," Austin pillories him for it. Is that legitimate?

David. No. Hypocritical.

Melissa. Maybe Austin thinks Ayer stretches words to excess in a counterintuitive way.

One would have to inquire into Ayer's reasons for stretching the word.

Dan. Yes, as a philosopher, Ayer should not be so sloppy.

Richard. Austin's case is that Ayer is trying to make it seem like these uses are the only appropriate ones, and so it is important that they are in fact stretched.

Ian. Ayer might say that it is not appropriate to say that we have evidence, but that doesn't show that we in fact do not have evidence.

So, Ian, what is the evidence on the basis of which you know [insert Ian's favorite observation statement here].

Whether or not these are the only appropriate uses, it is important that the reason we were interested in the first place is because we are interested in the relationship between perception, reality, and knowledge, and so stretching the words that are central to those ideas (evidence, verification, material object) threatens to change the subject.

In fact, as Austin points out, something has quite clearly gone wrong when Ayer claims that it requires infinitely many verifications to be certain of the truth of any sentence about material objects. In the ordinary way we use these words, that is just plain wrong.

Ayer says, as Austin quotes him, "Whereas the meaning of a sentence which refers to a sense-datum is precisely determined by the rule that correlates it with the sense-datum in question, such precision is not attainable in the case of a sentence which refers to a material thing."

Ayer's thought is connected with his claim that no sentence about a material thing can be conclusively verified, but sentences about sense data can be ("incorrigibly" verified).

Something has obviously gone deeply wrong. For one thing, it is easier to be certain about vague claims than it is about precise ones, and Ayer has things going in the opposite direction. Example: "I see something" (vague) vs "I see a book" (precise).

The example I just used makes the point that "vague" and "precise" are, at least often, relative terms. A term that is sufficiently precise in some uses may be vague in others. If one wants to know if a room is big enough for a meeting, "50 people will attend" is perfectly precise. If one wants to know whether one can, in speaking at the meeting, assume that the audience knows the basics of Peter van Inwagen's metaphysics, the same sentence is much too vague.

In addition, one can say vague or precise things about both medium-sized dry goods and about perceptual experiences:

So, what does Ayer in fact have in mind, that he is evidently misdescribing in terms of precision? It evidently has something to do with the idea that the perceptual experience of sense data is the only experience relevant to a report about the sense data. Ayer admits that we might lie in making a report or misuse the words in which we make the report (mistakenly think that cyclamen is a shade of green, or the like), but he thinks that we can't make any other kind of error. Austin points out (113n) that for that to be true, one would have to eliminate the possibility of a difference between the verbal report and what it is a report of by stipulation. It is then hard to see how the truth of a sense data report is more than a purely linguistic matter, which would amount to exactly the mistake Ayer claims Carnap makes, a mistake Ayer takes himself to be careful to avoid.

In addition, I can't resist pointing out that Austin claims the Ayer confused precise with exact, and he shows that the two are different with the following example: (paraphrased) If I measure a banana with a ruler, I may find that the banana is precisely 5 5/8 inches long. If I measure a ruler with a banana, I may find that the ruler is exactly five bananas long. The second is exact, but not precise. The example has taken on a life of its own, completely with references to banana metaphysics, and the like.

-- ShaughanLavine - 21 Apr 2003 - 12 Apr 2007