As we saw in the previous couple of chapters, Chisholmian internalism has impressive historical credentials. Coherentism, the subject of this chapter, has a lineage only slightly less impressive; both as a theory of truth and as a theory of justification it enjoyed enormous vogue during the second half of the nineteenth and first quarter of the twentieth centuries. Like music, coherentism had its three great B's: Francis Bradley, Bernard Bosanquet, and later (as one born out of due time) Brand Blanshard. During the next fifty years or so (with the rejection of absolute idealism and everything connected with it) coherentism fell on evil times. Now, however, it is flourishing once again and has such doughty advocates as Keith Lehrer¹ and Laurence BonJour,² who adopt coherentism as an account of warrant or justification, and Nicholas Rescher,³ who adopts it as an account both of warrant and of truth. Furthermore it has spawned a whole new variety or subspecies: probabilistic or Bayesian coherentism. In this chapter I shall ignore my own caveat (Preface, p. vii) and consider coherentism überhaupt; I shall argue that it does not afford the resources for a satisfying account of warrant or positive epistemic status. In chapter 5, I shall take a specific example, arguing the same conclusion with respect to the variety of coherentism admirably presented by BonJour in The Structure of Empirical Knowledge.⁴ Then in chapters 6 and 7 I shall turn to probabilistic or Bayesian coherentism, thought of first as an account of warrant and second as an account of rationality.

Our first problem, naturally enough, is to characterize coherentism. This problem is not wholly trivial: there is considerable confusion as to what coherentism is and no generally accepted account of the relevant coherence relations. It is worth noting initially, however, that coherentism is nicely thought of as a variety of the post-classical Chisholmian internalism of the preceding chapter. So thought of, the sort of intrinsic value appealed to is, naturally enough, coherence (however exactly the latter is to be thought of); and since coherence is a relation just among beliefs, the only relevant “purely psychological properties” would be beliefs.

Part of the motivation for coherentism, I think, has been similar to Locke's motivation for deontological internalism. Conceding that it might not be within our power to bring it about that all our beliefs are true, Locke claimed that, even so, we can at least make sure that we do our epistemic duty in forming our beliefs. Similarly, part of the attraction of coherentism is the thought that even if there is no guarantee that we will or can reach the truth, it is at least within our power to bring it about that our beliefs are coherent. Perhaps that is the best we can do, but perhaps that is also enough for mere mortals. Of course it isn't as easy as one might think to achieve coherence (as we shall see in the next chapter); in fact it isn't even easy to achieve logical consistency, and no amount of effort will guarantee it. A coherentist who recognizes this difficulty (and thus loses that motivation for coherentism) can nonetheless retreat to the Chisholmian, Brentanoesque stance: coherence is an intrinsically valuable state of affairs, and it is coherence that confers warrant, whether or not it is within our power to attain it.

The best way to understand coherentism, I think, is by contrasting it with foundationalism, whose structure is at least initially clearer. Foundationalism comes in many varieties, but perhaps the most famous and important brand of foundationalism goes back to Descartes and Locke. As we saw in chapter 1, Descartes and Locke were the fountainheads of the deontological tradition with respect to warrant and hence also the fountainheads of the internalist tradition with respect to it. Descartes and Locke are therefore the sources of classical deontologism and classical internalism; but they are also the sources of classical foundationalism. We shall get to classical foundationalism later (pp. 84–86); but let us begin here by thinking about foundationalism more broadly. Suppose we start with a partial account of foundationalism, focusing on those features that relevantly distinguish it from coherentism. And since the foundationalist sometimes accuses the coherentist of endorsing circular reasoning, we shall pay special attention to the foundationalist's rejection of such reasoning.

I. Ordinary Foundationalism

We form, discard, maintain, and modify beliefs. And a salient characteristic of our way of doing these things is that we sometimes do them on the basis of evidence. Taking the term ‘evidence’ in perhaps its most familiar sense (at any rate most familiar to philosophers), the evidence on the basis of which I form a given belief will be some other proposition or propositions I believe; in the preceding chapter I called such evidence ‘propositional’ evidence. There is, for example, the evidence for relativity theory, or for the proposition that the gospel of John was composed late in the first century A.D., or for the claim that life began on earth more than three billion years ago via the mechanisms suggested by A. G. Cairns-Smith.⁵ Of course this is just one use of the term: clearly there are other, analogically related uses, and in chapter 10 of Warrant and Proper Function I shall explore some of them.

Foundationalism—a family of views that has had an extraordinarily illustrious career in Western thought—takes its fundamental inspiration from the first kind of evidence, from propositional evidence. It starts from the apparent cleavage between those beliefs you accept on the evidential basis of other beliefs, and those you accept in the basic way—accept, but not on the evidential basis of other beliefs. One attributes to Aristotle the property of being the fountainhead of foundationalism (subject, as Quine says in another connection, to contradiction by scholars, such being the penalty for attributions to Aristotle). Aristotle and some of his medieval followers are classical foundationalists—ancient classical foundationalists, as I shall call them, to distinguish them from such modern classical foundationalists as, for example, Descartes, Malebranche, Locke, Leibniz, Berkeley, Hume, and a thousand lesser lights. Modern classical foundationalism, obviously enough, has been the dominant tradition in epistemology, in the West, since the seventeenth century.

The foundationalist, therefore—call him a generic foundationalist, since we are here concerned with what is common to all foundationalists—starts from the distinction between beliefs we accept in the basic way and those we accept on the evidential basis of other beliefs. Thus I believe 23×48 = 1,104 on the evidential basis of such propositions as 3×8 = 24, 3×4 = 12, 12 + 2 = 14, and the like. His idea is that every belief is either basic or accepted on the basis of other beliefs (and given the definition of basicality I just gave, here he will encounter no disagreement); he adds that in a correct or healthy human system of beliefs, there are basic beliefs, and every nonbasic belief will be accepted on the basis of other beliefs that offer evidential support for it, in such a way that every belief is supported, finally, by basic beliefs, beliefs in the foundations. These beliefs, of course, are not accepted on the basis of others; the basis relation is finite and terminates in the foundations.

An immediate and important consequence of this fundamental idea is the rejection of circular reasoning: that is, the foundationalist finds fault with a system of beliefs in which a belief A₀ is accepted on the evidential basis of a belief A₁, which is accepted on the basis of A₂, which is accepted on the basis of A₃,…, which is based on A_n, which is based on A₀.⁶ Thus suppose I believe

(1) Life originated in tide pools during an n-year stretch some m years ago

on the basis of my beliefs that

(2) Conditions C then obtained

and

(3) The probability that life would arise on earth during an n-year stretch when condition C obtained is high.

Suppose I believe (3) on the evidential basis of my belief that

(4) The probability that life would arise in a single tide pool during a 100-year stretch under condition C is 1⁄p;

and suppose finally that I believe (4) on the basis of my belief that life did indeed arise in tide pools during an n-year stretch, which would be monumentally improbable on any assignment significantly less than 1⁄p to the probability that life would arise, under condition C, in a single tide pool during a 100-year stretch. That is, suppose I believe (4) on the basis, among other things, of (1). Then, says the foundationalist, my system of beliefs is defective by virtue of containing a circle in the basis relation.

Now it is often said that the central difference between foundationalism and coherentism lies just here: the coherentist does not object to circular reasoning, “provided the circle is big enough.”⁷ Indeed, so the suggestion goes, the coherentist goes further; he revels in circular reasoning, for it is precisely in such circular chains that he sees warrant as arising. He must therefore suppose, if this characterization is correct, that the basis relation in a noetic structure does not simply transfer warrant: it somehow generates it, at least if the chain involved is sufficiently long. On this view (so the story goes) every belief is accepted on the basis of other beliefs; none holds the privileged position of being basic or foundational; and apart from coherence considerations, none is more warranted than any other. Suppose we begin by looking into this alleged opposition. Can it really be that the coherentist does not reject circular reasoning, instead recommending it as the path of true philosophy? And why does the foundationalist reject circular reasoning? Why do we reject it? And how shall we characterize foundationalism?

A. Basis and Support

The ordinary foundationalist begins by observing that we accept some beliefs on the evidential basis of other beliefs. Thus I may learn from you that Paul was either at the party or at the bar, and from the bartender that he wasn't at the bar; I then form the belief that Paul was at the party on the basis of these other two beliefs. Of course it isn't entirely easy to say precisely what it is to believe B on the evidential basis of A. I don't believe B on the basis of A just because A is the or a cause of my believing B;⁸ but no doubt the relation in question involves a causal element of some sort. Further, believing B on the basis of A does not require that I explicitly infer B from A, although that would certainly be a significant special case of the relation. On the other hand, it does require that I believe both A and B, and that, at any rate, I have in the past believed both A and B occurrently, not just dispositionally. Perhaps the best the foundationalist can do here (in addition to correcting misapprehensions) is to give examples and hope for the best.

Now if I believe a proposition A but do not believe it on the evidential basis of other beliefs I hold, then A is basic for me. (It does not follow, of course, that A is certain, incorrigible, indubitable, unrevisable, maximally warranted, or believed more firmly than any belief that is not basic for me; many criticisms of foundationalism miss the mark just because they treat basicality as if it did entail one or more of those items.) Certain beliefs about my own immediate experience—the sort of belief I might express by such sentences as ‘I am feeling tired’ or ‘It seems to me that I see something red’ or even (in a Chisholmian vein) ‘I am being appeared to greenly’—are typically basic for me. It would be at best difficult for me to believe the proposition that I seem to see something red on the evidential basis of other propositions; such beliefs are ordinarily basic for me. Similarly, many beliefs that we accept a priori are also typically basic for us. I believe the corresponding conditional of modus ponens (as well as various instances of it) and a host of other obvious truths of mathematics and logic in this basic way; the same goes, I daresay, for you.

B. Proper Basicality

According to the modern classical foundationalist, beliefs of the previous two sorts are not only basic for us; they are also properly basic for us. Roughly speaking, a belief is properly basic for me if it is basic for me, and also meets some other condition C, differing choices for C leading to different varieties of foundationalism. We might say, for example, that a belief is properly basic for me in case it is basic and has a certain degree of warrant for me. How much warrant? Here again there are options. We could hold that a belief is properly basic for me only if it has so much warrant that it is certain for me (has the maximal degree of warrant for me); or we might say that it is properly basic for me if it has enough warrant so that I know it (or have some special type of knowledge of it). Alternatively, we could say that it is properly basic for me if I am not irrational in accepting it, am justified in accepting it, am within my epistemic rights in accepting it.⁹ Thus the evidential objector to theistic belief¹⁰ holds that one is not justified, not within her epistemic rights, in accepting belief in God without evidence (without proofs or arguments); and (so he says) hence belief in God is not properly basic in this sense. Aquinas, on the other hand, would hold that a person could certainly be within her epistemic rights in believing in God in the basic way, but belief in God taken in the basic way cannot constitute knowledge (that is, scientia) that there is such a person as God. Such modern classical foundationalists as John Locke (on still a third hand) proposes that a belief—any belief—is properly basic for me only if it is either self-evident or appropriately about my own immediate experience.

Of course many other sorts of belief are often taken as basic, and with respect to these the modern classical foundationalist will not be indulgent. Suppose I seem to see a tree: I have that characteristic sort of experience that goes with seeing a tree. I may then form the belief that I see a tree, or that there is a tree there (more realistically, something like that tree must be more than one hundred feet tall!). In the typical case, that belief will be basic for me; I will not ordinarily accept the proposition that I see a tree on the evidential basis of other beliefs I hold.¹¹ In particular, I will not ordinarily accept this proposition on the basis of the belief that I have that special seeming-to-see-a-tree or being-appeared-to-treely sort of experience, for I will not ordinarily believe this latter proposition at all. (I will not, ordinarily, be paying any attention to my experience; I will be concentrating on the tree.) And here the modern classical foundationalist will disapprove. She need not deny, of course, that we do in fact accept such perceptual beliefs in the basic way, but in her view such beliefs are not properly basic; for the modern classical foundationalist holds that only self-evident beliefs and (appropriate) beliefs about my own mental states are properly basic. Hence, she thinks, a perceptual belief is unwarranted if it is not accepted on the evidential basis of propositions that are either self-evident or appropriately about my own immediate experience.

Similarly, consider a memory belief—that you had breakfast this morning, for example. Once more, you do not believe this proposition on the basis of some belief about your experience—for example, the proposition that you seem to remember that you had breakfast this morning. Here again the modern classical foundationalist may concede that in fact we often do accept memory beliefs in the basic way; he will deplore this tendency on our parts, however, for in his view—his official view, at any rate—memory beliefs are not properly basic.

According to the foundationalist view some of my beliefs are basic for me: the rest, naturally enough, are nonbasic for me, held on the evidential basis of other beliefs I hold. If things are going properly, however, I will not (of course) believe a given proposition on the basis of just any proposition; I will instead believe A on the basis of B only if B evidentially supports A. Foundationalists of varying stripes, naturally enough, have made different suggestions as to the further characteristics of this relation. Descartes seemed to think that the only support worth its salt is deductive support. Locke added inductive evidence. (The facts that Feike is a seventeen-year-old Frisian and 19 out of 20 seventeen-year-old Frisians can swim evidentially support the proposition that Feike can swim.) Peirce added abductive evidence: the sort of evidence provided (for example) for special relativity by the null result of the Michelson-Morley experiment, muon decay phenomena, the Hafele-Keating experiment involving jet transport of cesium clocks, and so on.

Say that S's noetic structure is the set of propositions he believes, together with certain epistemic relations that hold among him and them. An account of S's noetic structure will specify, for example, which of his beliefs are basic and which nonbasic. It will also include something like an index of degree of belief, specifying, for each proposition he believes, how firmly he believes it. (We could add, if we like, that it specifies for each proposition to which he affords some degree of confidence, the degree of confidence he affords it.) Further, it will include, for any belief B that is a member of it, an account of which beliefs (and which sets of beliefs) support B, and the degree and kind of support they provide for it. It will also include an epistemic history of each belief, specifying the conditions under which the belief was formed and has been sustained, and a deontological history, which specifies for each belief whether it was formed and has been sustained in accord with epistemic duty. Finally, it will include a coherence index of the structure, which measures both the coherence of the structure as a whole and, for each of its members, its coherence with the rest of the structure. Still other properties of a noetic structure are relevant and important in some contexts, but I shall ignore them for now.

1. Six Foundationalist Theses

Now foundationalism is a normative thesis about noetic structures; more exactly, it is a connected group of such theses. It is a group of theses about how a system of beliefs ought to be structured, about the properties of a correct, or acceptable, or rightly structured system of beliefs. The normativity in question could be deontological: one who conforms to his intellectual duties, on this suggestion, will be such that his noetic structure satisfies the theses.¹² The normativity could be axiological: the state of affairs consisting in the existence of a noetic structure that satisfies the theses in question is intrinsically valuable.¹³ The normativity could be aretaic: there are (intrinsically or extrinsically) valuable noetic or intellectual states; there are also the corresponding intellectual virtues, the habits of acting that produce or promote or enhance those valuable states of affairs; and the noetic structure of a person with the appropriate intellectual virtues will satisfy the theses in question.¹⁴ The normativity in question could be understood in terms of what an idealized human being would be like, as in some of the Bayesian literature (see the next couple of chapters). Finally, the normativity could be ‘functional’ as we might call it: the sort involved when we say of a diseased heart or knee or immune system that it isn't functioning properly, isn't working the way it ought to work; and the claim would be that a properly functioning noetic structure (the noetic structure of a person whose epistemic faculties were functioning properly) would satisfy the theses in question. Foundationalism as such is neutral among these options, although—as I argue in Warrant and Proper Function—the last is most apposite if it is warrant that is at issue.

So foundationalism is a connected group of normative theses about noetic structures. Our present concern is with only some of these theses: those relevant to the alleged dispute between foundationalists and coherentists. And a noetic structure is what it should be, says the former, only if the beliefs it contains are either properly basic or believed on the basis of propositions that support them. Just to have a term, say that a noetic structure that is what it ought to be is proper. Then, according to the foundationalist,

(I) A proper noetic structure will have a foundation: a set of beliefs not accepted on the basis of other beliefs.

Next, the foundationalist accepts some theses about the relation of evidential support (call it the supports relation) and the believed-on-the-basis-of relation (the basis relation, for short):

(II) The supports relation is irreflexive.

Although of course every proposition entails itself, no proposition provides evidential support for itself, that is, no proposition is such that a properly functioning human being could believe it on the evidential basis of itself. From this perspective, the term ‘self-evident’ is something of a misnomer. A self-evident proposition is not one for which the evidential support is provided, oddly enough, by itself; it is rather a proposition that is evident in itself and thus requires no evidential support.

(III) The basis relation is irreflexive in a proper noetic structure.

It may be doubted that anyone is so benighted as to believe a proposition A on the evidential basis of A; but even if it could be done, it clearly should not be done.

The basis relation (in a proper noetic structure) and the supports relation coincide on irreflexivity; they diverge on asymmetry. For:

(IV) The supports relation is not asymmetrical

and

(V) The basis relation is asymmetrical in a proper noetic structure.

The supports relation, clearly enough, is not asymmetrical. Special relativity provides evidential support for muon decay phenomena, and muon decay phenomena also provide evidential support for relativity theory. A person could sensibly accept relativity theory on the evidential basis of muon decay phenomena, but it is also true that a person could sensibly accept muon decay on the basis of relativity. For one who is convinced of the Axiom of Choice, that axiom could serve as her evidence for the Hausdorff Maximal Principle; for the former entails the latter. But someone else already convinced of the latter could properly use it as his evidence for the former; for the latter entails the former. A person might find it obvious that there are no nonexistent objects and use this truth (as I see it) as her evidence for his view that proper names in fiction (‘Captain Marvel’, ‘Hester Prynne’) that do not name existent objects do not name any objects at all. On the other hand, someone who held that such names name nothing could properly take that fact as part of his evidence for the claim that there are no nonexistent objects.

But even if the supports relation is not asymmetrical, the basis relation, in a proper noetic structure, is asymmetrical. If my belief that A is accepted on the evidential basis of my belief that B, then my belief that B must not be based on my belief that A. More exactly, suppose N is a proper noetic structure. Then if the belief that A (in N) is based upon B₁… B_n, none of the B_i will be based upon A. If my belief that life arose in antediluvian tide pools is based on, among others, my belief that the probability that life would arise in a given tide pool in a hundred-year period (under the conditions that then obtained) is 1⁄n, then (if my noetic structure is proper) my belief that that probability is 1⁄n will not be based on the proposition that life arose in this way, and there were n tide pool/100-year pairs available. If my evidence for special relativity is, say, the muon decay phenomena and the Hafele-Keating experiment, then it will not also be the case that my evidence for muon decay is special relativity.

So the basis relation, in a proper noetic structure, is asymmetrical. More generally,

(VI) The basis relation, in a proper noetic structure, is noncircular.

In a proper noetic structure it will not be the case that a belief A₀ is accepted on the evidential basis of a belief A₁, which is accepted on the basis of A₂, which is accepted on the basis of A₃,…, which is based on A_n, which is accepted on the basis of A₀.

2. Against Circularity

Now why, exactly, does the foundationalist object to circular reasoning? What exactly is objectionable about it? The answer is not trivially obvious. Of course, if the basis relation (in a proper noetic structure) were transitive, then we could see why circular reasoning would be objectionable. For suppose this relation were transitive: then if there were a circular path in my noetic structure, there would be some proposition A that I believe on the evidential basis of A itself; and this, according to the foundationalist, is either impossible or at any rate not a way to achieve warrant. The fact is, however, that the basis relation need not be transitive in a proper noetic structure; I can quite properly believe A on the basis of B and B on the basis of C without believing A on the basis of C. Perhaps things are different for a suitably ideal reasoner; for perhaps a person who is aware of believing A on the basis of B and B on the basis of C (and who has intellectual powers of an appropriate order) will also believe A on the basis of C. But there is no reason to think transitivity thus required for us ordinary mortals. So this isn't the answer.

But now recall (by way of making progress toward the answer) that warrant comes in degrees; some of my beliefs, obviously, have more warrant than others. (The belief that it looks from here as if that peak is triangular will have a higher degree of warrant, for me, than the proposition that it really is triangular.)

According to the typical foundationalist, a proposition can have or acquire warrant in at least two ways. On the one hand, it can be properly basic, can achieve warrant just by being accepted in the right circumstances; on the other, it can acquire warrant by virtue of warrant transfer, by virtue of being believed on the basis of some other proposition that already has warrant.¹⁵ The degree of warrant enjoyed by a nonbasic belief will depend on at least two factors: the degree of warrant enjoyed by the propositions on the basis of which it is believed, and the strength of the supports relation holding between it and them. In the extreme case, a proposition B believed on the basis of a proposition A may have as much warrant as A itself—perhaps where A obviously and self-evidently entails B. (If you tell me that you are thirty-five years old, my belief that you are over thirty, even if based only on my belief that you are thirty-five, may enjoy as much warrant, for me, as does the belief that you are thirty-five.) In other cases the warranted proposition may display much less warrant than the warrant-conferring proposition; but in no case will the warrantee enjoy greater warrant than the warrantor. If my warrant for a proposition A arises from my believing it on the basis of another proposition B, then my warrant for A cannot exceed my warrant for B.

Here two caveats are necessary. First, a proposition believed on the basis of another may have a higher degree of warrant than that other by virtue of deriving some of its warrant from other sources. Suppose I have a relatively dim or vague memory of having seen Paul at the New Year's Eve party two months ago; you tell me that you distinctly remember seeing Eleanor at the party, and I know that Eleanor seldom attends parties without Paul. Then my belief that Paul was at the party is based partly on my beliefs that you saw Eleanor there and that she seldom attends parties without Paul; for my warrant for believing the proposition in question is greater than it would have been had I had only my memory to go on. It therefore receives part of its warrant from being believed on the basis of the propositions that Eleanor was there and that she never goes to parties without Paul. Nevertheless, it may have a higher degree of warrant, for me, than is enjoyed by either of those propositions; for it also receives some warrant, for me, from my memory of having seen Paul there, dim and vague as that memory may be. The foundationalist need not hold, therefore, that if A is believed on the basis of B, then the warrant of A cannot exceed that of B; he holds instead that, if A's warrant is derived entirely from its being believed on the basis of B, then its warrant cannot exceed B's.

Second, things are a bit more complicated when we turn to the case where my warrant for a proposition A arises from my believing it on the basis of several propositions—B₁…, B_n, say. The foundationalist will not hold that my warrant for A cannot exceed my warrant for the conjunction of the B_j. Of course for some propositions believed on the basis of others—conjunctions of those others, for example—this would be no more than the sober truth; but for others—disjunctions of the B_i for example—it is plainly false. Nor will he want to hold that A's warrant cannot exceed the warrant of the least warranted of the B_i: my warrant for a disjunction B₁ ∨ B₂ may derive entirely from its being believed on the basis of B₁ and B₂, but its warrant may nonetheless exceed theirs.¹⁶ About all the foundationalist can say generally here is that if the warrant of a proposition A derives entirely from its being believed on the basis of the B_is, then A's warrant will be a function of the warrant of the B_is, together with the degree to which they support A; and in no case can A have more warrant than that of the disjunction of the B_is.

In sum, when a belief B enjoys an increase or access of warrant by being believed on the basis of other propositions, we have warrant transfer from a belief that already has it to another. And here we meet a seventh foundationalist thesis, one crucial to the whole foundationalist picture:

(VII) Warrant does not increase just by virtue of warrant transfer.

We are now in a position to see more clearly why circular reasoning is objectionable from a foundationalist point of view. For the sake of simplicity, suppose we confine our attention to the special case where A₀ is believed solely on the basis of A₁, A₁ solely on the basis of A₂,…, A_{n − 1} on the basis of A_n, and A_n on the basis of A₀; and let us add that none of the A_i receives any warrant from any source other than its being believed on the basis of A_{i − 1}. (The application to the more general case is easy enough to make.) Say that B is directly warranted by A if B is believed on the basis of A and gets all of its warrant by virtue of being believed on the basis of A. Then what we have here is a circular chain (a chain circular with respect to the directly warrants relation): a finite set of propositions A₀ to A_n (ordered by the directly warrants relation) such that for any A_i (i ≠ n), A_i is directly warranted by A_{i + 1}, and A_n is directly warranted by A₀. Say further that, for any member A and B of the chain, A gets warranted by some proposition that gets all its warrant from B. It is clear, first of all, that if a proposition gets all its warrant from itself, then it gets no warrant at all; a proposition that gets all its warrant from itself has no warrant.

It is clear, second, that this relation—the gets-all-its-warrant-from relation—is transitive. For suppose C gets all its warrant from B and B gets all its warrant from A. If C is directly warranted by B, then it follows immediately that C gets all its warrant from A; so suppose C is not directly warranted by B. Consider the segment of the chain from B to C (inclusive). Clearly B + 1, the proposition that is directly warranted by B, gets all its warrant from A; but then the same will go for B + 2, the proposition directly warranted by B + 1, and so on all the way to C. Thus C gets all its warrant from A. This relation, therefore, is transitive. But then it follows that, in the circle in question, A₀ will get all its warrant from itself. As we have seen, in that case A₀ has no warrant at all; and the same will go for each of the other members of the circular chain. It is for these reasons, therefore, that the foundationalist rejects circular reasoning. As he sees it, a noetic structure that displays a circle in its basis relation displays a defect—a warrant defect.

II. Coherentism

According to foundationalism, then, a proper noetic structure will not contain a circle in its basis relation; and this is because warrant does not arise just by way of warrant transfer. Current lore has it, however, that the coherentist does not object to circular reasoning at all, provided the circle is large enough. What the coherentist claims (so we are told) is that in a proper noetic structure, every belief is accepted on the evidential basis of other beliefs; since the number of beliefs any of us has is finite, it follows that in such a structure there will be circles in the basis relationship; that, however, is no barrier to their having warrant, if the resulting circles are sufficiently large; and indeed it is in just such circles that warrant arises.

Now this is not at all easy to believe. The intuition relied upon by the foundationalist here—that circular reasoning cannot produce warrant—is very strong.¹⁷ First, it seems wholly obvious that even if a person could believe a proposition on the evidential basis of itself, this maneuver would confer no warrant whatever upon that proposition. Suppose I believe that my dog is an alien from outer space, and suppose I could manage, somehow, to start believing this proposition on the basis—the immediate evidential basis—of itself. Surely this belief would not thereupon acquire a greater degree of warrant for me than it had before I executed this dubious maneuver. So if the belief in all its warrant from B if and only if A is directly warranted by B or A is directly question has no warrant for me apart from what accrues to it by virtue of standing in the believed-on-the-basis-of relationship, then in this case it has no warrant at all. Say that a circular chain of the sort under consideration is of unit circumference if the set of beliefs involved is a unit set; and say more generally that it is of circumference n if the set of beliefs involved is n-membered. Then clearly a circle of unit circumference confers no warrant upon its member. But surely the same goes for a circle of circumference 2. If at first I believe both A and B and then manage to believe each on the basis of the other, I am no better off, epistemically speaking, than I was at first. (If I accept special relativity solely on the basis of the muon decay phenomena and believe in muon decay solely on the basis of special relativity, then neither has any warrant for me at all.) And how could it help to increase the size of the circle? If a circle of circumference n does not produce warrant, surely the same will go for a circle of circumference n + 1.¹⁸ Warrant cannot magically arise just by virtue of a large evidential circle. (If I go around the circle twice, do I get twice as much warrant?) If the coherentist really holds that circular reasoning is a source of warrant, then his views are unlikely indeed.

A. Coherentism Characterized

But why saddle him with anything so miserably implausible? There is a much more charitable way to construe his characteristic claim. He should not be seen as endorsing circular reasoning or making an implausible remark about the properties of the basis relation; he isn't really claiming that the basis relation is a source of warrant. Nor does he hold that the basis relation in a rational noetic structure can sometimes be circular. His suggestion, instead, is that coherence is the sole source of warrant. He is instead pointing to a condition under which a belief is properly basic—a condition under which a belief acquires warrant without being accepted on the evidential basis of other beliefs. On his view, a belief B is properly basic for a person S if and only if B appropriately coheres with the rest of S's noetic structure (or with some part of it, or with an appropriately purified version of it, or some part of that).¹⁹ If a proposition B coheres with my noetic structure, then B is warranted for me; its warrant does not arise, however, by virtue of my believing it on the basis of the rest of my noetic structure, so that those other propositions are my evidence—deductive, inductive or abductive—for B. Indeed, if, for any proposition A I believe, I accept A on the basis of the rest of what I believe, then my noetic structure would contain a host of tight basis circles; for then I would believe A on the basis of B₁,…, B_n and each of the B_i on the basis of A together with the rest of the B_i. (Of course, if we added a dimension and spoke not of circular but of cylindrical reasoning, then the cylinders in question could be of considerable height, even if only of circumference 2.)

So the coherentist does not really tout circular (or cylindrical) reasoning. What he does instead is suggest an unusual condition for proper basicality, a new source of warrant: he holds that a belief is properly basic for me if and only if it appropriately coheres with the rest of my noetic structure. A pure coherentist resolutely rejects warrant transmission altogether; for her, all propositions that enjoy warrant in a noetic structure are basic in that structure. Deduction, induction, and abduction may indeed figure, in one way or another, as elements in the coherence relation, but warrant does not get transmitted by the basis relation from one proposition to another. Of course, a coherentist need not be a pure coherentist; she can instead embrace an impure or mixed variety. She could hold, for example, that the source of warrant is coherence but add that warrant is sometimes transferred via the basis relation. Thus I may be warranted in my belief that some horses have quirky personalities by virtue of believing that proposition on the basis of my belief that Clyde is a horse and has a quirky personality; and I may be warranted in that latter belief by virtue of its coherence with the rest of my noetic structure. Global coherentism is compatible with local foundationalism; the view that coherence alone is the source of warrant is compatible with the view that warrant is sometimes transmitted. What is really characteristic of coherentism is not a view about the transmission of warrant but a view about its source. Seen from the present perspective, therefore, the coherentist reveals her true colors as a nonstandard foundationalist with unusual views about what is properly basic.²⁰

The pure coherentist holds that all warranted propositions in a noetic structure are basic in that structure; no warrant gets transmitted. The impure coherentist holds that some propositions may get their warrant by virtue of being believed on the basis of others; but the ultimate source of the warrant in question is coherence. Both accept the view that coherence is the only source of warrant; and this is the central coherentist claim. According to the ancient or medieval foundationalist, perception and self-evidence or reason are sources of warrant. The modern classical foundationalist replaces perception by introspection; Reid restores perception and adds testimony, memory, sympathy, induction, and others. And the coherentist, by contrast, casts her lot with coherence. She holds that coherence alone is a source of warrant.

This is the source of a fateful consequence: on the coherentist view, a belief acquires no warrant by virtue of its relation to experience. The fact that I am indeed being appeared to redly confers no warrant, either on my belief that I am perceiving something red or even my belief that I am being appeared to redly; the fact that the corresponding conditional of modus ponens seems to me self-evident confers no warrant, for me, on my belief that that proposition is true; the fact that I find myself powerfully impelled to believe that I had an orange for breakfast, that this memory belief seems right—that fact confers no warrant, for me, on that belief. What confers warrant on these beliefs, if indeed they are warranted for me, is no more and no less than their coherence with the appropriate body of beliefs.

B. Coherentism Rejected

Coherentism, therefore, is a special case (a very special case) of foundationalism: the variety according to which the only source of warrant is coherence. Hence the characteristic coherentist claim that all beliefs are on an epistemic par; all stand equally before the bar of coherence; and, in case of failure of coherence, all are equally liable to revision. The ordinary foundationalist, of course, balks at this excess of egalitarian fervor (as he sees it); relative to a given set of circumstances, he says, some beliefs are privileged, acquiring warrant just by virtue of being formed or sustained in those circumstances. Thus a perceptual belief—the belief that I see a tree, for example—may have warrant for me, and get it from circumstances having little to do with coherence; it is, we may say, a starting point for thought. (It does not follow that it is unrevisable, incorrigible, indefeasible, or certain; what follows is only that it gets at least some warrant in some fashion other than by way of coherence.)

Coherentism is clearly mistaken. Note first that coherentist theories are what Pollock calls “doxastic” theories; they hold that the warrant or positive epistemic status of a belief is determined solely by the relations that belief bears to other beliefs. If you hold precisely the same beliefs in two different circumstances, then any belief you hold in both circumstances will enjoy the same degree of warrant in each circumstance; the nondoxastic circumstances do not matter.²¹ The coherentist claims that coherence is sufficient for warrant, and necessary for it, in that a proposition has warrant for me if and only if it is coherent with my noetic structure or appropriately (deductively, inductively, or abductively) follows from propositions that are coherent with my noetic structure.

I think we can see that she is mistaken on both counts: coherence is neither necessary nor sufficient for warrant. As to sufficiency: it seems wholly clear that a person's noetic structure might be thoroughly coherent even though some of her beliefs have no warrant at all. Oliver Sacks recounts the case of the Lost Mariner, who suffered from Korsakov's syndrome, a profound and permanent devastation of memory caused by alcoholic destruction of the mammillary bodies of the brain. He completely forgot a thirty-year stretch of his life, believing that he was 19 years old when in fact he was 49; he believed it was 1945 when in fact it was 1975.²² His beliefs (we may stipulate) were coherent; but many of them, due to this devastating pathology, had little or no warrant.

Or consider someone S who, by virtue of some noetic malfunction (due to tumor, Cartesian demon, or Alpha Centaurian) of which he is not aware, believes, whenever he is appeared to redly, that no one else is ever appeared to redly. (Perhaps you think this state of affairs a bit bizarre and unlikely; but what is required is only that it be possible in the broadly logical sense.) Add that S's beliefs, on these occasions, are coherent—coherent in any reasonable sense a coherentist might propose. Nevertheless his noetic structure is grossly defective and his belief lacks warrant. S may be deontologically justified in this belief; he may be within his rights in accepting it (perhaps it is not even within his power to reject or withhold it): it may be that there are no noetic duties he has flouted. This belief may also be rational in the Foley sense (see chapter 7, p. 132): perhaps the disorder is so deep-seated that even if he reflected long and hard, he would retain his peculiar beliefs, becoming ever more confirmed in them. Nevertheless the belief in question has little or no warrant for him.

Examples can be multiplied. Timothy is a young artist from Firth, Nebraska, with an intense (indeed, pathologically inordinate) admiration of Picasso. Waiting at a supermarket checkout, he idly picks up a copy of the National Enquirer, reading therein that Picasso, contrary to what most of us have always thought, was really an alien from outer space. As a result of his overwhelming and diseased veneration of Picasso, Timothy forms the belief that he, too, is really an alien from outer space, having been deserted by his Alpha Centaurian parents on an exploratory field trip to Nebraska. The rest of his beliefs fall into a coherent pattern with this one. His belief that he is an alien from outer space, however, clearly has little or no warrant for him—and even if it happens, by some enormous coincidence, that in fact he is an alien, he certainly doesn't know that he is.

Finally, consider the Case of the Epistemically Inflexible Climber. Ric is climbing Guide's Wall, on Storm Point in the Grand Tetons; having just led the difficult next to last pitch, he is seated on a comfortable ledge, bringing his partner up. He believes that Cascade Canyon is down to his left, that the cliffs of Mount Owen are directly in front of him, that there is a hawk gliding in lazy circles 200 feet below him, that he is wearing his new Fire rock shoes, and so on. His beliefs, we may stipulate, are coherent. Now add that Ric is struck by a wayward burst of high-energy cosmic radiation. This induces a cognitive malfunction; his beliefs become fixed, no longer responsive to changes in experience. No matter what his experience, his beliefs remain the same. At the cost of considerable effort his partner gets him down and, in a desperate last-ditch attempt at therapy, takes him to the opera in nearby Jackson, where the New York Metropolitan Opera on tour is performing La Traviata. Ric is appeared to in the same way as everyone else there; he is inundated by wave after wave of golden sound. Sadly enough, the effort at therapy fails; Ric's beliefs remain fixed and wholly unresponsive to his experience; he still believes that he is on the belay ledge at the top of the next to last pitch of Guide's Wall, that Cascade Canyon is down to his left, that there is a hawk sailing in lazy circles 200 feet below him, that he is wearing his new Fire rock shoes, and so on. Furthermore, since he believes the very same things he believed when seated on the ledge, his beliefs are coherent. But surely they have little or no warrant for him. The reason is cognitive malfunction; his beliefs are not appropriately responsive to his experience. Again, he may be deontologically justified in accepting those beliefs; they may also have Foley justification. But they have no warrant for him. Clearly, then, coherence is not sufficient for positive epistemic status.

But neither is it necessary. It is entirely possible that one of my beliefs should have considerable warrant, even if it is neither coherent with the rest of my noetic structure nor appropriately follows from ones that are coherent with it. A necessary falsehood is presumably incoherent with the relevant body of beliefs; but couldn't a necessary falsehood have warrant for me? You are a habitually authoritative mathematician; you tell me T is a theorem, and produce for it a ‘proof’ so subtle that you yourself cannot see its fallacy; the proof is fallacious, however, and in fact T is a necessary falsehood; but doesn't T have warrant for both of us? (Didn't Frege's axioms for set theory have warrant for him before he received Russell's fateful letter?) Perhaps you are suspicious of this example: you think little should rest on examples resting on necessary falsehood. Very well, there are plenty of examples that do not involve necessary falsehoods. You are an eminent but idiosyncratic Oxford epistemologist; I an unduly impressionable undergraduate. You offer me a battery of complex and subtly powerful arguments for the conclusion that no one is ever appeared to redly. I am unable to withstand the force of your argumentation and am utterly convinced. The next day I am walking along High Street, reflecting on the significance of what you have proved to me, when suddenly a great large double-deck red bus runs up on the sidewalk just behind me. I turn around in terror, see the bus, and am (violently) appeared to redly; since I have been reflecting about these matters, I notice (that is, believe) that I am thus appeared to. Unless my noetic structure undergoes instant metamorphosis (and we can stipulate that it does not), my belief that I am appeared to redly will be incoherent with my noetic structure; nevertheless it will have a considerable degree of warrant.

Another example: I am an arboreal expert giving a lecture on trees; among other things I claim that no oak trees grow in the state of Washington. Naturally I believe that I have never seen an oak tree in any part of that state. I suddenly notice you in the audience. Seeing you jogs my memory; I seem to remember an occasion on which you and I noticed a particularly luxuriant oak flourishing on the campus of Western Washington University in Bellingham. At the moment when it seems to me that I do so remember, the proposition that I have seen an oak tree in Washington has warrant for me, despite the fact that it does not then cohere with my noetic structure. I will find myself believing that I have indeed seen an oak tree in Washington; if for some reason my other beliefs do not alter, the belief in question will not be coherent with my noetic structure. Nevertheless it then has warrant for me, despite the fact that it does not thus cohere. And the change that is called for, of course, is not that of rejecting or trying to reject the memory belief in question; what is called for is revising the rest of my noetic structure in such a way that it is coherent with the belief in question. Coherence, therefore, is neither necessary nor sufficient for warrant.²³

So coherentism is false; coherence is neither necessary nor sufficient for warrant, and there are sources of warrant in addition to coherence. It does not follow, of course, that coherence is not a source of warrant; what follows is only that it is not the sole source of warrant. A foundationalist can accept coherence as a source of warrant—one source among others. A perceptual belief that doesn't fit with the rest of what I believe may be quite properly rejected. Climbing Guide's Wall again, Ric seems to see what looks like a cow on a two-foot ledge 200 feet away and 300 feet off the ground. Wise as he is in the ways of cattle, he realizes that there could hardly be a cow there, rejects the testimony of his senses on the grounds of coherence considerations, and concludes that the light or the angle of vision must be deceiving. Here the foundationalist could gracefully concede that the beliefs there is no cow on that ledge and appearances are deceptive get warrant by virtue of coherence with the appropriate body of Ric's beliefs; and here coherence considerations overcome or outweigh the impulse to the perceptual belief in question. Of course if he climbed over to the ledge and from a distance of 6 feet, say, it Still appeared to be a cow, he would have to draw a different conclusion—perhaps that the local climbers are unusually playful. Here coherence considerations are outweighed by the powerful impulse toward believing the proposition there is a cow on that ledge provided by the sort of experience that goes with clearly seeing a large cow from a distance of 6 feet.

Coherence considerations obviously have their own force and sometimes produce warrant: at any rate it is no part of foundationalism as such to deny this suggestion. The foundationalist denies that coherence is the only source of warrant, but (at any rate qua foundationalist) he has no stake in maintaining that it isn't a source of warrant at all.²⁴

III. Classical Foundationalism

Coherentism is therefore to be rejected: coherence is not the only source of warrant. But what are the other sources? According to modern classical foundationalism (an extraordinarily influential picture dominating Western epistemological thought for nearly three centuries), they are reason and experience—but then both reason and experience narrowly construed. On this view a proposition is properly basic if and only if it is either self-evident or else appropriately about one's own immediate experience—specifying how one is appeared to, for example. Any other propositions that are acceptable for you must be ones that are appropriately supported by propositions of these kinds.²⁵

Classical foundationalism has fallen on evil days; and rightly so. As Reid saw and argued, the whole development of modern philosophy from Descartes to Hume shows that classical foundationalism ‘taken to its logical conclusion’, as they say, yields the consequence that very little, far less than we would ordinarily think, is epistemically acceptable for us. None of the propositions we believe about ordinary material objects, or the past, or other persons—none of these propositions seems to be appropriately supported by propositions that are properly basic according to the classical foundationalist's standards for proper basicality; the latter offer precious little by way of evidence for the former. But these propositions certainly seem to be acceptable for us: why, then should we accept classical foundationalism? If there were powerful and compelling arguments for it, then perhaps we should have to grit our teeth and accept it; but the powerful arguments are not forthcoming. So classical foundationalism has fallen into disrepute if not desuetude. I don't propose to add my voice to that of the howling mob, except to say that many forms of classical foundationalism look to be self-referentially incoherent. According to these forms, a proposition A is acceptable for me if and only if it is either properly basic or believed on the evidential basis of propositions that are (1) properly basic, and (2) support A. But this proposition itself is not properly basic by this criterion: it is neither self-evident nor appropriately about someone's immediate experience, and (subject to the indeterminateness of what is to count as support here) it is certainly hard to see that it is appropriately supported by propositions that do meet that condition.²⁶

So classical foundationalism fails. This fact has been widely celebrated (sometimes with a sort of foolish extravagance); it has also been widely hailed as requiring rejection of all of epistemology, or even all of traditional philosophy, or even the very idea of truth itself. In a moment of anguish, Dostoyevski blurted that if God does not exist, everything is possible. Richard Rorty and his friends go him one (or more) better and without the anguish: if classical foundationalism is wrong, there is no such thing as truth. These intemperate reactions to the demise of classical foundationalism betray agreement with it at a deep level: agreement that the only security or warrant for our beliefs must arise by way of evidential relationship to beliefs that are certain: self-evident or about our own mental states. But why think a thing like that? And why follow these enthusiasts into that grand confusion between metaphysics and epistemology, confusing truth with our access to it, announcing the demise of the latter as a consequence of the failure of classical foundationalism? Here we have confusion twice confounded: first, confusion of truth with our access to it and, second, confusion of knowledge with Cartesian certainty. But as to the first, truth owes nothing to our access to it; and as to the second, Cartesian certainty is indeed a will-o'-the-wisp, but nothing follows for knowledge.

If classical foundationalism is to be rejected, however, if there are sources of warrant in addition to reason and experience (construed thus narrowly), if it is not the case that the only propositions properly basic for me are those that are either self-evident or about my own immediate experience, then what other sorts of propositions are properly basic? According to Thomas Reid, there is nothing but an arbitrary partiality in awarding this status only to propositions of those two sorts; he proposed that certain beliefs acquired by way of perception are also properly basic, as are beliefs acquired by way of memory, or by way of induction, as well as beliefs about the mental states of others acquired by way of what he calls “sympathy,” and still others. In Warrant and Proper Function I argue that these and other sorts of beliefs are properly basic. There are still other candidates: perhaps moral or ethical propositions are properly basic in this way. According to John Calvin, as I understand him, certain beliefs about God are also properly basic; the sensus divinitatis takes its place along with perception, reason, memory, sympathy, and induction as a source of warrant.²⁷

There is therefore no shortage of candidates. In Warrant and Proper Function I shall examine some of the candidates; for now, however, we must continue our scrutiny of coherentism. We therefore turn, in the next chapter, to the impressive work of Laurence BonJour.