Lecture Notes, UC Davis Philosophy 102, Theory of Knowledge

The Role of Coherence in Foundationalism

G. J. Mattey, Senior Lecturer

Robert Audi

Text: Robert Audi, "The Foundationalism-Coherentism Controversy," from The Structure of Justification, pp. 117-164 (1993)

Our final look at the fallout from Aristotle's regress problem is directed toward a defense of a modest form of foundationalism. The author is Robert Audi, long-time University of Nebraska professor who has recently taken a post at Notre Dame University.

Two Forms of the Regress Problem

As we have seen, ending the regress of grounds for knowledge or justification provides a menu of options, each of which has its disadvantages. Either the regress ends with a circle, or it proceeds infinitely, or it ends in items which are not known or justified, or it ends in items which are said to be known or justified directly. The last option is probably the most favored, but it has the drawback that it appears dogmatic.

Audi claims that there is not a single regress problem, but rather two. The first type he calls "the dialectical form of the regress problem," and the second its "structural form." In the dialectical form, grounds are cited as a response to the question, "How do you know?" Such questions might take the form of a skeptical challenge. In the structural form, we are asking how specifically an item of alleged knowledge or a person's body of knowledge is structured. The question here is, "How is it that you know?" While the dialectical regress does not presuppose that one knows anything, the structural regress does.

It matters how the regress problem is framed for four reasons.

  1. An answer to the dialectical question would be showing that one knows, but an answer to the structural question does not.
  2. An answer to the dialectical question requires claims about what one knows, but an answer to the structural question does not.
  3. The answer to the dialectical question requires giving a solution to the regress problem by mobilizing an argument, but an answer to the structural question requires only that there be a solution, which is cited in propositions that make up the answer.
  4. The answer to the dialectical question involves a process of justifying, while the answer to the structural question in most cases presupposes that there is a property of justification.
For example, suppose that the regress concerns the proposition that someone was injured in a car accident. A structural answer might involve citing the testimony of a witness, Janet. This might not satisfy a skeptic, who could ask for stronger grounds. These grounds would involve claims to know that they themselves are true, which requires knowledge about what one knows. The grounds would be cited as part of an argument to the conclusion that I know that someone was injured in the accident, which is a process of justifying.

On the other hand, the citation of the testimony as the structural basis of my knowledge involves no claims to knowledge that the testimony is true and no argument to the effect that this is a sufficient ground for knowledge. Moreover, the citation is intended to convey only that my belief has the property of being justified on the basis of the testimony.

Audi claims that the dialectical form of the problem favors coherentism. The basic reason is that it requires higher-level claims about what I know, and it is hard to see how these claims could ever be directly justified.

There are two senses in which a belief might be directly justified.

Higher-order knowledge and justification seem to be in their very nature the product of inference. They also seem to require inferential relations in order to have their positive epistemic status of being knowledge or being justified belief. The kind of grounds cited in answering the structural question, however, seem suitable for direct justification. That Janet told me someone was injured in the accident seems psychologically direct and, for the foundationalist, is epistemologically direct.

On the other hand, the structural question seems to favor foundationalist answers. There is no place here for an infinite regress of grounds. But it is not really prejudicial against coherentism. For it does not demand that the justification terminate in directly justified beliefs. It could still be that justification accrues from coherence.

There is another respect in which the structural question is more neutral than the dialectical. The latter is biased toward internalism, which is

roughly the view that what justifies a belief such as a visual impression, is internal in the sense that one can become (in some way) aware of it through reflection or introspection (internal processes).
Externalism, by contrast, denies that what justifies a belief need be accessible internally. To answer the dialectical question, "How do you know?" it seems that one must appeal to what is internally accessible.

Because he finds the structural question to be more neutral than the dialectical question concerning the kind of account of knowledge and justification its answers suggest, he will devote the rest of his account to that question.

The Argument for Foundationalism

Audi tries to show that answering the structural question generates a regress. To do so, he introduces the concept of an "epistemic chain," which is a connected series of items of knowledge, including the degenerate series with only one member. We can apply the regress argument to epistemic chains, so that there are four ways of ending a chain. The best way to do so is with direct knowledge. The conclusion is that if one has knowledge, then one has at least some direct knowledge.

This argument is supposed to be completely general, so that it applies to any item of knowledge whatsoever. A similar argument can be generated using the concept of justification rather than that of knowledge. Audi will shift his focus back and forth between knowledge and justification.

The option of an infinite epistemic chain seems psychologically unrealistic. An argument for this claim is that the human mind does not have the ability to grasp all the elements of an epistemic chain, such as the huge numbers that would be generated in the infinite task of doubling each number, starting with 1. Circular justification is ruled out from the start. Audi thinks that the most convincing forms of coherentism do not require arguing in a circle.

Although in general beliefs that do not amount to knowledge cannot terminate the chain, an exception might be made for beliefs that are justified, but not to the extent required for knowledge. We could in that case be "in the general vicinity of knowledge," which is close enough so that "Knowledge is not emerging from nothing, as it were." The foundation then might not be "bedrock, but perhaps ground that is nonetheless firm enough to yield a foundation we can build on."

The final alternative is the preferred one. Epistemic chains terminate in non-inferential knowledge. Audi endorses the four standard types of non-inferential knowledge that traditionally have been invoked: knowledge arising directly from perception, memory, introspection, reason. Appealing to these sources also allows us explain why the beliefs they generate are likely to be true. In the case of perception, the reality of music playing "explains both my perception and, by explaining that, indirectly explains my believing the proposition I know on the basis of this perception."

There are several modes in which chains grounded in non-inferential knowledge might differ.

Regarding the last mode, Audi claims that the non-inferential beliefs need not be infallible or even justified in a way that cannot be defeated. Introspective beliefs (illustrated by an earlier example: "It vaguely seems to me that I hear strains of music") are not justified this strongly. Other reasons for this claim will be given later.

Fallibilist Foundationalism

After giving an extensive example, Audi opts for a kind of foundationalism with respect to justification which he calls "fallibilist." Justified foundationalist beliefs may be false, the means of transmission are fallible as well: they could take us from true premises to a false conclusion. A further feature of the account is that it is possible to discover error or lack of justification at any point in the chain.

A fallibilist foundationalism could also apply to knowledge. The links in the epistemic chain are false, since they are knowledge. But the grounds for any given belief could be defeated, in which case one ceases or fails to be justified in the original belief, so that knowledge is lost. "We can lose knowledge when our grounds for it are defeated by counterevidence."

Coherence may play two important roles in fallibilist foundationalist justification. The negative role is that incoherence may defeat justification or knowledge, even with respect to what is non-inferential. The positive role is played by an "independence principle" that could be used in explaining justification. The principle is that "the greater the number of independent mutually coherent factors one believes to support the truth of a proposition, the better one's justification for believing it (other things being equal)." A further role might be that justification is enhanced by the greater explanatory power that coherence brings.

Holistic Coherentism

It is difficult to explain what coherence is. It is something more than mere consistency, and it involves an internal relation of beliefs to one another. It is enhanced, in some way, by close explanatory relations and by a belief's raising the probability of another. In further explaining coherentism, Audi takes it to be of the type championed by Lehrer: a belief is justified by virtue of its coherence with other beliefs one has.

The most plausible type of coherentism makes justification non-linear. Therefore, it is not circular, since circular justification requires that we move in an epistemic chain from a conclusion through premises and back to the conclusion as a premise. Non-linear coherentism is called "holistic" by Audi. The "whole" which defines holistic coherentism could be global, encompassing all of one's beliefs, or it could be as small as set of two beliefs.

A couple of examples are given to illustrate. After an initial ground is given for the belief that his daughter is home (he hears music), a number of other claims are made which are related, but not linearly, to the original belief. "Beliefs representing knowledge do not lie at one end of a grounded chain; they fit a coherent pattern and are justified through their fitting it in a coherent way." Another example is supposed to show that "arriving at a justified belief, on this view, is more like answering a question by looking up diverse information that suggests the answer than like deducing a theorem from axioms."

In neither case is a circle involved. Another point is that the beliefs which provide grounding through coherence may be psychologically direct, not based on inference, but not epistemically direct. The "wedge strategy" on behalf of coherentism is to separate the psychological from the epistemic grounds. A psychologically direct belief may not be epistemically direct, and justification may involve no psychologically direct beliefs at all. But it is open to the coherentist to allow a "psychological foundationalism." "Holistic coherentism may grant experience and reason the status of psychological foundations of our belief systems, but it denies that they are the basic sources of our justification or knowledge."

The Role of Coherence in Foundationalist Justification

A possible case seems to favor fallible foundationalism over holistic coherentism. If a person could have a single justified belief, then coherence is not needed for justification. A coherentist could object that this is impossible because we must at least have beliefs about the meanings of its terms before a belief can be justified. This, in turn, would require coherence. But Audi responds that the only thing required here is "conceptual coherence." The idea here is that one cannot acquire a single concept without acquiring a battery of concepts.

So take the case Audi gives, that of "S's having, if just momentarily, only a single belief which is nonetheless justified, say that there is music playing." Does it derive any of its justification from coherence? We might think that it does, due to the fact that it would not be justified if it failed to cohere with another belief, say that S is looking at a compact disk that is not turning in the CD player.

This kind of coherence is only negative, and it is compatible with foundationalism. The second belief would function as a defeater. But because we are dealing with fallibilist foundationalism, it is already built into the justification of a belief that it is not defeated by any other belief. Incoherence is merely a constraint on justification.

The foundationalist could go on to claim that the belief that would serve as a defeater ordinarily is the type which is itself foundationally justified, say by means of perception. Further, the epistemic principles which spell out what a defeater must be could be "a kind of foundational justification for principles to the effect that certain kinds of evidences or beliefs override certain other kinds." If these were not foundationally justified, then we would have to allow that a belief can be justified by coherence even if "one had no justified principles by which one could, for instance, inferentially connect the justified belief that p with others that cohere with it."

The point of these arguments is that the fact that a belief depends negatively on its not being incoherent with other beliefs "does not imply that as coherentists hold, its justification positively depends on coherence." Negative dependence is determined by the absence of defeaters, but positive dependence is determined by justifiers. So it remains open what role fallibilist foundationalism might assign to coherence.

The fact that appeal is made to other beliefs, and to higher-order beliefs, in order to show one's knowledge does not imply that these beliefs play any role in having justification. A small child can have a good character without being able to show that he has one. Showing justification through the mobilization of higher-order beliefs is a sophisticated process. It may seem to be required by the meaning of 'justified,' but this is due to an ambiguity in that term.

At this point, it appears that there is no positive role for coherence in perceptual justification. Audi considers the oddity in the assertion that I am justified in a belief, even I cannot justify it. But this is odd only because it is an assertion that one is justified. This calls for a showing. If one simply is justified, without asserting that one is, then this need disappears.

Audi makes two further points. The first is that it is easier to give a justification than to show that one is justified. In the case of perceptual justification, the former task can be accomplished simply by describing one's perceptual situation. The latter requires this and more: the connection between what has been given and how they meet the condition of being justified. Yet this is what skeptics and most coherentists (probably) are looking for. The second point is that giving a justification stops a regress, but the regress cannot be stopped in the process of showing that one is justified.

There is a temporal dimension at work here. One may be able to give a justification only after the point where one is justified. When asked, for example, one might be form new beliefs about one's perceptual situation, based on a disposition to do so. But this is not the same as having that other belief at the time, which is what the coherentist demands. When we do reach this point, we cease to offer anything further, which is another point in favor of foundationalism.

Foundationalism is Very Different from Coherentism

At this point, Audi has portrayed foundationalism and coherentism in ways that bring them together in certain respects. Fallibilist foundationalism requires the truth of conceptual coherentism, while coherentism is consistent with psychological foundationalism. But there remain deep differences between the two.

First, there are differences over what justification is. Coherentism requires that coherence is something cognitive (where what coheres has propositional content), while foundationalism allows that what gives positive justification may be non-cognitive, such as a perceptual experience. Coherence is said to come from relations of beliefs to one another, but perceptual experience does not stand in such a relation. The relations that make for coherence are accessible, but those which confer justification from experience need not be.

The key issue concerns whether coherence is a basic source of justification, whether it is sufficient to confer justification by itself. Audi cites the typical objection to this claim: that our beliefs may be ever so tightly organized but do not confer justification unless they originate in a proper way.

A coherentist response is to say that one can find coherence in any justified belief. Moreover, the coherence we find is of the kind that actually does account for the justification involved. An objection to this is that coherence requires us to have more beliefs than we have at any given time, but this can be overcome by allowing that all that is needed is the right dispositions to believe. So is coherence at least a necessary condition for justification?

Audi responds that it might be, in a way, but in a more important way it is not. The idea is that coherence might always accompany justification, in which case it is a "consequential necessary condition." But it is not a "constitutive necessary condition," as it is not what makes the belief justified. So coherence might be like the heat that is produced by friction: "a necessary product of it, but not part of what constitutes it."

This is supported by the fact that it is difficult to find cases where there is justification through coherence without the usual sources in experience and reason. It is hard to find real cases because our beliefs are generated in this way. So we would have to imagine some other creatures whose highly-coherent beliefs are not. They would lack justification.

The conclusion is that coherence is based on justification, not vice-versa. And the cases already considered can be explained by hypothesizing that both are the result of beliefs arising from basic sources. The explanation is to be found in conceptual coherentism and negative coherentism.

A final possible way that coherence might be basic is conditional. Suppose we grant that some beliefs are justified to some extent independently of coherence. What coherence can do is send this justification over the top, in a way that cannot be done "additively" by the provision of new evidence.

This suggestion is rejected by Audi. First, he claims that the picture of foundational degrees of belief being added together is misleading. This is not how it is done in probability calculations. And the rules of probability themselves seem to be justified a priori, which is to say, foundationally. Yet rules such as this seem to be presupposed by any account of how coherence can increase justification.

The second reason for rejecting conditional basicality for coherence is that coherence gives no special means of increasing the degree of justification. Suppose I hear confirming testimony on six different occasions. Then contrast this with hearing the testimony on one occasion, where I put it all together into a coherent account. This can be explained simply by the addition of a new belief and the kind of inferential relations open to foundationalists. No special relation of coherence need be involved.


The many distinctions made thus far are useful in defusing the charge that foundationalism is dogmatic, in the sense that what does the justifying (e.g., perceptual experience) is not itself justified, which makes the justification it confers dogmatic.

Dogmatism here is to be understood epistemically. It need not be a psychological trait of stubbornness or psychological certainty. One can adhere stubbornly to what is self-evident. Nor need it be closed-mindedness, which is a psychological trait that may have no epistemological consequences. The same holds for intellectual pugnacity or willingness to defend one's views vigorously.

A key ingredient in dogmatism is a second-order attitude or a disposition to have one. "Typically, a dogmatically held belief is maintained with a conviction (often unjustified) to the effect that one is right, e.g., that one knows, is amply justified, is properly certain, or can just see the truth of the proposition in question." This is not sufficient for dogmatism, though, since one can rightly have such a belief or disposition to believe in cases such as those of beliefs about simple logical truths.

Summing it all up, then, there are four conditions which are "highly characteristic of dogmatism."

Only the last lays claim to being necessary for dogmatism, but they probably but not self-evidently are jointly sufficient.

Given all of this, is belief, as characterized by the foundationalist, dogmatic? It seems to be so if the question is the dialectical one. The process of showing that one knows comes to a halt abruptly with the assertion that a given belief simply is justified. It is not generally a good strategy for foundationalists to say that these assertions are foundational. One reason is that one would have to be justified in believing that there are no defeaters.

On the other hand, answering the structural question does not make foundationalism look dogmatic. There is no presumption of second-level belief, which seems to be a necessary condition for dogmatism. And even if one has the second-order belief that one can assert something without offering evidence, the person may not meet the other conditions of dogmatism.


The counterpart to the charge that foundationalism encourages dogmatism is that coherentism encourages tolerance. This is, however, another false stereotype, and it reveals the fatal flaw in coherentism. If it really is tolerant, then it does not require for justification something like observational input or the goal of obtaining truth and avoiding error. But this is too much tolerance. The level of tolerance must be reduced if requirements like these are added.

Indeed, coherentism may foster dogmatism, because the lack of foundational appeals makes it difficult to settle debates. One tends to be dogmatic if one's opinions are not subject to checks. Moreover, the concept of coherence is vague, which allows one to understand it to one's own advantage, again fostering dogmatism. The extent of dogmatism seems to depend more on temperament than on one's account of justification.

A final consideration is that it looks like foundationalism is more dogmatic because it dismisses the skeptic. This happens, however, only in the context of answering the dialectical question. Given the structural question, there is no dogmatism, since the claim is conditional: if there is knowledge, then it has this structure.


It may be that the appeal of foundationalism and coherentism is based on the way they answer different questions. Fallible foundationalism seems to be superior. It can avoid objections and can give an account of the role of coherence in justification. It solves the regress problem better than the alternatives. If is in accordance with "reflective common-sense" by counting as justified what we commonly hold to be justified. It is also psychologically plausible. Finally, it allows integration of the account of justification with psychology and biology. The beliefs it takes as basic are vital for survival.

A final virtue is that fallible foundationalism can account for disagreement. People's experience is different, which will yield different foundational beliefs. And the inferences they draw from these beliefs will be different as well, especially when it comes to giving explanations. So it is hardly dogmatic. And it has the main virtues that have been thought to be the exclusive property of coherence theories.

[ Course Home Page | Menu of Lectures ]