(Originally presented in a 1979 student philosophy conference. Text footnotes are in square brackets.)
One of the central problems of epistemology is that of whether or not there are epistemically ultimate propositions: that is, propositions which:
Foundationalist theories of the structure of human knowledge affirm the existence and necessity of such propositions, while coherentist theories deny both. This is, in fact, the essential point of disagreement between the two types of theory. If coherentists can show that no proposition can be epistemically ultimate in the sense required for foundationalism, then they will for all practical purposes have proven their case.[1] Foundationalists will have deflected the coherentist attack if they can produce even a single example of such a proposition.
This paper will argue that epistemically ultimate propositions, in the sense required for foundationalism, are impossible. I will begin with an overview of the current state of the debate between foundationalists (such as C.I. Lewis) and coherentists (such as Nicholas Rescher). I will then show that both sides have missed the most important point of the issue, and explain what that point is. In so doing, I will offer ideas that can lead toward a coherentist theory of the structure and justification of human knowledge.
1. The Foundationalist Advance
C.I. Lewis, explaining the foundationalist theory, writes that:
Empirical truth cannot be known except, finally, through presentations of sense. ... Such justification involves a step or steps depending on logical truth. ... Our empirical knowledge rises as a structure of enormous complexity, most parts of which are stabilized in measure by their mutual support, but all of which rest, at bottom, on direct presentations of sense.[2]
What Lewis claims in this passage, essentially, is the following:
To these points, Lewis would add that such foundational propositions must be certain. Noting that some of our beliefs are merely probable, he asks the rhetorical question: "But can there be such probability which is genuine unless the basis of it may be assured as unqualifiedly true?"[3] He then concludes that:
Unless there is eventually some termination of this series of supporting grounds; in something which stands fast without support; will not the whole edifice of empirical knowledge come tumbling down? Is it not, then, required that there be ultimate data ... which themselves are certain? There are grounds of empirical knowledge which are thus certain, in given presentations of sense. [4]
There are two interwoven strands of argument here. First, Lewis argues that there must be epistemically ultimate foundational propositions if empirical knowledge is to be possible at all. Second, he argues that such foundational propositions cannot be merely probable but must be certain.
The argument that there must be foundational propositions is familiar and intuitively appealing. Suppose that S knows that P on ground Q, that Q on ground R, that R on ground T, and so on. Human beings’ mental capacity is finite, so S cannot carry around in his or her head an infinite chain of supporting reasons. Therefore, we seem to be faced with two possibilities.
The first possibility is that the final supporting ground in the chain, say Z, will not be justified at all. In this case, the whole chain of propositions justified by reference to it will ultimately be un-justified: and then, "the whole edifice of empirical knowledge will come tumbling down," unsupported at its very bottom.
On the other hand, the second possibility is that Z might be in some way self-justifying: because of the kind of statement it is, it might be legitimate to say, "Z because of Z." In this case, the chain of propositions justified by X would be secured at the bottom "in something which stands fast without support," that is, Z. If empirical knowledge is to stand, Lewis argues, we must opt for possibility #2, that there are self-justifying foundational propositions.
Lewis’s reasons for thinking that foundational propositions must be certain seem more complex and less clear[5]: all involve considerations about probabilities. The first line of thought he seems to have entertained is that because the probabilities of propositions are determined relative to the class of foundational propositions, and the probability of a proposition relative to itself is 1.0, then the probability of foundational propositions must be 1.0 -- i.e., they must be certain.[6]
Lewis’s second line of thought involves applying the probability calculus. He offers two strands of argument. Briefly, they go as follows:
1. The probability of a conjunction is determined by multiplying the individual probabilities of the conjuncts.
2. If the probabilities of the foundational propositions were significantly less than 1.0, then the probability of the foundational class of propositions, considered as a whole, would be a product of fractions and hence, quite low (i.e., they would be improbable).
3. Therefore, the probabilities of the foundational propositions must be 1.0 or close to it, because otherwise, we would be "justifying" empirical knowledge by reference to a class of foundational propositions which we were not justified in believing.
1. Any statement which is merely probable must be so relative to some supporting ground.
2. Suppose that a statement P is probable relative to a chain of supporting grounds G1..Gn.
3. If believing P is justified, then the conjunction of P and its supporting grounds (P & G1 & ... & Gn) must also be justified.
Here, the argument breaks into two distinct strands.
4a. Unless Gn is certain, then we will need to invoke additional supporting grounds until we reach a Gi which is certain (see premiss 1); otherwise, we will be involved in an infinite regress.
4b. Unless Gn is certain, then the conjunction of P and its supporting grounds will have a very low probability and we will not be justified in believing P.
5b. If Gn is not certain, from (3), (4), Modus Tollens, then P is unjustified and empirical knowledge fails.
6b. Therefore, if empirical knowledge is to be possible, Gn must be certain.
Other considerations aside, it seemed clear to Lewis that foundational propositions must be statements describing sense-experience, because statements about objective "things" could never be certain, while sense-experience statements could be.[7]
Thus, in summary, Lewis held that there must be foundational propositions which are absolutely certain and self-justifying reports of sense-experience.
2. The Coherentist Response
Nicholas Rescher at the University of Pittsburgh makes a very simple argument against the foundationalist position. Such views, he says, maintain that there must be a class of absolutely certain foundational propositions which serve to justify all non-foundational propositions. Members of this class must be certain in order not to require any verification beyond themselves; but on the other hand, they must say a great deal in order to be deductively strong enough to support all the rest of our knowledge. And, Rescher argues, these two requirements are inconsistent:
These two qualifications for the axiomatic role (content and security) clearly stand in mutual conflict with one another. This tension makes for a weak point ...[8]
The only empirical statements which can be certain are reports of sense-experience. If I say that "I see an apple," I could be mistaken: what I see might be a holographic image of an apple. On the other hand, if I say merely that "I seem to see an apple" -- reporting my sense-experience -- I can be certain that, regardless of how the perception was produced, my statement is true. But, maintains Rescher,
such assertions purchase security at the price of content. For no amount of claims in the language of appearance ... can ever issue in any theoretically guaranteeable result regarding what is actually the case in the world. While they themselves are safe enough, appearance-theses will fall short on the side of objective content. This dilemma of security vs. content represents the Achilles heel of foundationalist theory of factual knowledge.[9]
Rescher’s argument boils down to this. If we want our foundational propositions to be certain, then we must restrict their content so much that they are no longer an adequate basis for empirical knowledge. But on the other hand, if we want our foundational propositions to provide an adequate basis for empirical knowledge, then we must load them up with so much content that they can no longer be certain. Therefore, concludes Rescher, there cannot be foundational propositions in the sense required.
3. The Foundationalist Reply
Mark Pastin of Arizona State University answers Rescher’s "security vs. content" dilemma in two ways. First, he notes that if we dropped the requirement for foundational propositions to be certain, Rescher’s criticism would at once go wide of the mark. "Modest foundationalism," as Pastin calls it, does not require that foundational propositions be certain: merely that they be "self-warranted:"
A proposition which is self-warranted for a person at a time is a proposition which the person would be warranted to some degree, however slight, in believing at the time, even if he had no inductive evidential support for the proposition at the time.[10]
But what of the arguments made by Lewis and other foundationalists which purport to show that foundational propositions must be not merely self-warranted but also certain? In response to the first argument that
since the probability of propositions is determined relative to the foundational class, the members of that class must have a probability of 1.0,
Pastin simply takes this as a reductio ad absurdum of the view that epistemic probability is applicable to the foundational class itself:
This argument indicates that the foundationalist assumption that epistemic probabilities are probabilities relative to the foundational class should be restricted to non-foundational propositions.[11]
In response to the second argument, that
unless foundational propositions had a probability of 1.0 or close to it, then then probability of the whole class might be very low and hence unable to justify empirical knowledge,
Pastin asserts that
the argument assumes that foundational propositions are independent of one another and that the Multiplication Axiom is applicable in determining the epistemic probability of the foundational class given the probabilities of its members. Both assumptions are dubious.[12]
In response to the third argument about diminishing probability down a chain of supporting reasons to the foundational class, Pastin points out that the probabilities of propositions need be calculated only relative to the foundational class, so that the intervening chain of supporting reasons (with its effect on the probability calculation) drops out. Again, he argues, the foundational propositions need not be certain.[13]
In answer to the other half of Rescher’s dilemma, that sense-experience statements have inadequate content to support empirical knowledge, Pastin asserts that it "is simply a dogmatic denial of all views according to which a person can have warranted belief about physical reality on the basis of experience."[14] He concludes that Rescher’s argument fails to show the impossibility of a foundationalist theory of the structure of knowledge:
Rescher’s argument against foundationalism depends on one assumption which is applicable only to certain variants of foundationalism and another assumption which is unsupported. So the argument provides no reason for the non-skeptical epistemologists to embrace coherentism.[15]
4. Foundationalism’s Fatal Flaw:
The Impossibility of Epistemic Ultimates
Interesting as their arguments are, Lewis, Rescher, and Pastin have all missed the most important point in the dispute between foundationalism and coherentism. Whether foundational propositions are construed as certain or not is unimportant; I think that, if we grant foundationalism, Pastin makes a persuasive case that it should be "modest" in its claims for the certainty of foundational propositions.
The key point, however, is that foundational propositions are literally impossible, even in the sense needed for "modest" foundationalism.
In any version of foundationalism, foundational propositions are supposed to be epistemic ultimates, that is:
- They are self-warranted (however one glosses the term); and
- They are adequate in content to provide for the construction of empirical knowledge.
They must be self-warranted because if they depended on prior propositions for their justification, then they ipso facto would not be foundational. And they must be adequate in content to provide for the construction of empirical knowledge because if they required additional substantive propositions to do so, then they once again would not be foundational.
Both of these requirements come from the fact that in foundationalism, we are trying to construct our knowledge from the bottom up, so to speak. The foundational propositions are on the bottom: and if anything is required to be underneath them -- whether to provide additional justification or additional content -- then they cannot be foundational. If either of the two conditions for epistemic ultimacy were to fail, then it would show the impossibility of a foundationalist theory of the structure of human knowledge. As it turns out, they both fail. Let’s consider each of the requirements in turn.
Let’s consider a typical example of what Lewis and other foundationalists consider foundational propositions: "I see a red patch," interpreted as a report of sense-experience that makes no commitment about anything in the "real world" beyond the experience itself. This statement asserts:
"I," that is, a self
"see," that is, have a visual experience of
"a," that is, one member of a class
"red," that is, a color
"patch," that is, a spatial expanse
This statement is supposed to be self-warranted because it reflects a fact immediately given in experience, and because it is held that we cannot be (or at least, are very unlikely to be) mistaken about our own experience. But how much of this so-called foundational proposition is actually given in experience? Among the other propositions one would need to know (at least implicitly) in order to make or understand this statement, we find:
- There are selves and I am one of them.
- Selves can stand in certain relations to visual presentations.
- There are colors and red is one of them.
- There are spatial expanses and a patch is one kind.
Unless one were at least tacitly aware of these facts (which have been presented here only in minimal detail), one could not justifiably assert that he or she saw a red patch. What this shows is that sense-reports such as "I see a red patch," far from being self-warranted, depend for their warrant on already being in the context of a system of other propositions. Hence, as far as warrant goes, they are not "on the bottom" of knowledge and cannot be foundational.
It is important in following this argument to keep two relations distinct: the relations of truth-dependence and of warrant-dependence. If the foregoing argument is taken as referring to truth-dependence, then it involves an elementary logical mistake. The truth of "I see a red patch" entails the truth of the other propositions cited, but not (at least, so it would seem) vice-versa. On this interpretation, statements such as "I see a red patch" are logically prior to the other propositions and could well be foundational.
However, the argument refers not to truth-dependence, but to warrant-dependence. Mathematics could be deduced from mathematical physics, but not (apparently) vice-versa. Nevertheless, one needs to understand mathematics before one can understand mathematical physics. What we are concerned with here is the order of epistemic priority: that unless one already knew the other propositions cited, one could not be warranted in asserting that he or she saw a red patch. The systematic context does not need to be explicitly present -- indeed, that is never the case -- but it must always be tacitly so. In the absence of such a context, a person might have a visual presentation of a red patch and say "unhhh," but that would not be an item of knowledge.
Before moving on, I should mention one other point. If, as I have maintained, sense-experience statements cannot be self-warranted and hence cannot be foundational, why do they seem, introspectively, to be self-warranted? Even a coherentist would never claim that before judging "I see a red patch," he or she sits down and says:
Let me see: there are selves, and I am one of them. Ah, yes, red is a color, isn’t it? I do believe that selves can see red patches. Yes, I am entitled to assert that "I see a red patch."
The systematic context, while it is used to make the judgment, remains tacit -- in the background, but present. "I see a red patch" seems to be self-warranted and immediate for the same reason that, on the common-sense level, "I see a cup of coffee" seems self-warranted and immediate. Both judgments use a systematic context and rely on it for justification, but do not explicitly invoke it. This conclusion is confirmed by the psychological finding that people can perform extremely complex mental operations without being aware that they are doing so.[17]
The same kind of consideration which applied to the question of foundational propositions’ warrant also applies to the question of their meaning. We want, it should be recalled, to be able to start just with foundational propositions and then construct the rest of empirical knowledge from them. But it should be patently obvious that, isolated from its systematic context, "I see a red patch" is completely devoid of meaning. In other words, remove from the statement all of its logical connections to other statements of the system, i.e., take it out of the system and look at it alone, and what remains?
Peel off its connections to "There are selves and I am one of them," and you peel off the content of "I." Peel off the connections to "Selves can stand in certain relations to visual presentations," and you peel off the content of "see." Peel off the next two sets of connections and you peel off the content of "red patch."
It appears that either a proposition exists in the context of a system of other propositions, or it doesn’t exist at all. When we have removed the systematic context from a statement, nothing is left. We might still say "I see a red patch," but from the standpoint of meaning, we might just as well have given an inarticulate grunt.
I conclude that because propositions have no meaning in isolation from a systematic context, it is impossible to "start" with a small class of them and proceed to construct that context.[18] Again, the reason why propositions can seem meaningful in isolation is that their systematic context remains tacit.
5. Conclusion
In the foregoing, I’ve shown that in the sense required for foundationalist theories of knowledge, epistemically ultimate foundational propositions are impossible. Isolated from a system of non-foundational propositions, so-called foundational propositions are neither justified nor meaningful. Because foundationalism and coherentism are the only two alternatives to explain the structure of human knowledge, this shows that coherentism is the correct theory.
Footnotes
1. That is, assuming that there is not an infinite regress and that empirical knowledge is possible.
2. C.I. Lewis, Analysis of Knowledge and Valuation, p. 171.
3. ibid, p. 333.
4. ibid, p. 333.
5. These arguments are discussed in "C.I. Lewis’s Radical Foundationalism" by Mark Pastin.
6. Pastin asserts that Lewis himself might not have advanced this argument.
7. Cf. Analysis of Knowledge and Valuation, Chapter VII.
8. Nicholas Rescher, "Foundationalism, Coherentism, and the Idea of Cognitive Systematization," p. 12.
9. ibid, p. 13.
10. Pastin, "Modest Foundationalism and Self-Warrant," p. 143.
11. Pastin, "Foundationalism Redux," p. 4.
12. ibid, p. 5.
13. ibid, p. 3.
14. ibid, p. 6.
15. ibid, p. 6.
16. This analysis draws on the ideas of my friend Brand Blanshard, The Nature of Thought, Vol. I, pp. 230ff.
17. See, for example, Ernest R. Hilgard, "Hypnosis and Consciousness," in Human Nature, January 1978, p. 45.
18. The question which then arises, "How do we acquire any empirical knowledge in the first place?" is addressed in my paper, "Knowledge, Experience, and Justification," Auslegung, Spring 1980. The paper is also reproduced on this Web site.