Lecture Notes, UC Davis Philosophy 102, Theory of Knowledge
Circularity and Coherence
G. J. Mattey, Senior Lecturer
Text: Robert G. Meyers, The Likelihood of Knowledge, pp. 142-149 (1988)
Aristotle and the Pyrrhonian skeptics rejected circular justification, albeit with different results. Coherence accounts of justification allow circular justification, though usually under a less prejudicial name, such as "mutual support." Rejection of circular justification is the main reason that pure coherence accounts are shunned by most theorists of knowledge.
BonJour presented a coherence account that allows a relation of "reciprocal dependence within the system" of a person's beliefs. No belief is any more basic than any other. A particular belief is justified with respect to the system as a whole, and "the system is the primary unit of justification; particular beliefs are justified derivatively, by virtue of membership in such a system" (The Structure of Empirical Knowledge, p. 24). Coherence, for BonJour, is what makes a system justified.
We can contrast BonJour's account of justification, membership in a coherent system of beliefs, with that of Keith Lehrer, coherence with a system of beliefs. Lehrer proposes that "S is justified in accepting that p at t on the basis of system X of S at t if and only if p coheres with the system X of S at t" ("Knowledge Revisited," p. 321 of our text). He acknowledges that there is some circularity in justification but finds it benign (see pp. 322-325).
Robert G. Meyers, Professor at the State University of New York, Albany, evaluates the prospects for coherence accounts of justification that are open to the charge of circular justification. He argues, as we will see, that while there are some defenses against the charge, in the end it proves to be fatal to coherentism.
The Justification of Belief in Reliability
We have many beliefs that are directly about things, whether they be appearances or physical objects. Knowledge of such things is called by Meyers "first-level knowledge." Such knowledge, he goes on, requires higher-level knowledge, that is, knowledge of our reliability in making judgments at the first level. As Sellars puts it, to call something knowledge is to locate it in "the logical space of reasons, of justifying and and being able to justify what one says" ("Empiricism and the Philosophy of Mind, Section 36). I know that grass is green, then, only if I know that I am reliable in determining the color of grass and can use this knowledge in justifying my belief that grass is green.
The reason is that I am justified in believing that grass is green only if I am justified in believing that I am reliable in determining colors. (This requirement is common to many "internalist" accounts of justification, including coherence accounts, but is rejected by "externalist" accounts.)
How do we obtain knowledge of our reliability? Generally, we determine reliability of a process statistically, by observing its successes and failures, and then comparing them. A high ratio of success to failure is a strong indicator of reliability. It is not an infallible indicator, however, as it may have been mere chance that the observed cases came out mostly successfully. If we can go on to explain why the process is reliable in terms of an underlying mechanism, then (so long as the explanation is correct) we can show that success is no accident.
So if I wish to know whether I am reliable in determining the color of grass, I could observe my "track-record" (as Alston calls it) of successes and failures in identifying colors. Suppose I find that I am successful 98% of the time, and I conclude that I am reliable in what I believe about colors. In that case, the higher-level belief is justified by appeal to a range of first-level beliefs.
As already noted, all first-level knowledge are said to be justified on the basis higher-level knowledge, and according to the present method, higher-level knowledge is justified on the basis of first-level knowledge. If some of the first-level knowledge that justifies some higher-level knowedge that I am reliable are in turn justified (at least in part) by the higher-level knowledge, then there is circular justification.
As Meyers points out, it is at least logically possible that a higher-level belief that supports first-level beliefs is not justified by first-level beliefs it supports. In our example of color, the belief that I am reliable in determining colors may be justified by testimony of others about my reliability, and not by my assessment of my record of success.
But there are two reasons to think that this possibility does not hold in the case of justification of human belief. The first can be illustrated by the example just given. I would have to be justified in my beliefs that people have testified to my reliablity in determining colors. These beliefs ultimately would have to be justified by appeal to some perceptual beliefs, such as that I heard so-and-so say that I am a reliable judge of colors. So I would need a belief about my reliability in judging those perceptual beliefs.
Now you could say that this new higher-level belief might be justified by some beliefs different once again from the perceptual beliefs that it justifies. But then note that these beliefs would have to be justified by reference to some other higher-level belief. Now a regress threatens: you would have to bring in fresh perceptual beliefs to justify any higher-level belief, a fresh higher-level belief to justify the perceptual beliefs, and so on.
To stop the process, you must either find a higher-level belief about reliability that is not justified by some perceptual belief, or find a perceptual belief that is not justified by some perceptual belief. Meyers sees no prospects for this. The only other alternative is circular justification.
Lehrer has proposed a more general argument to the same conclusion. In order to be justified in believing anything at all, he maintains, I must be justified in believing that I am reliable (or "trustworthy," in Lehrer's terms) in what I believe. But the belief that I am reliable in what I believe can only be justified in terms of other beliefs that I have. So reciprocal justification is unavoidable.
Sellars's Solution
One way to break the circle is not to regard justification in the manner of Aristotle as a temporal process that takes us from what we already know to some new knowledge. In justifying a general belief, such as a higher-level belief about my reliability, I rely on past observations of success or failure. If these observations had to be justified already, then circularity threatens. But suppose they were not justfied before. I use them as evidence for the belief about my reliability now, and that belief becomes knowledge. At the very same time, the original observational beliefs become justified as well. So rather than justifying each other, the beliefs at the different levels become justified together at once.
The issue this raises is how beliefs can be justified together in a clump, or holistically. It does not seem that Sellars embraces an extreme holistic view, according to which all beliefs must be justified together, so that no belief is justified unless all are justified. He seems instead to be concerned with justification of beliefs of certain types, such as beliefs about colors.
Narrowing the range of beliefs at issue does nothing to explain how (limited) holistic justification actually takes place. Sellars is silent on this point. Meyers argues that explaining holistic justification is a problem for any empiricist (one who holds that the primary means of knowledge is through sense-perception).
Problems with Holistic Justification
Meyers gives two reasons that this is the case. The first has to do with the empiricist account of the acquisition of color-concepts. The argument is that we must acquire them several at a time because we can understand single colors only through their contrast with other colors. So it may be that I could acquire concepts of green and red at once, on the basis of observations of green things and red things, but I could not acquire the concept of red from observing only red things.
A similar consideration arises from what Sellars calls "concept-empiricism." This is the traditional account of the acquisition of concepts from sense-perception. Concept-empiricism is an old doctrine, as it can be found in Aristotle.
So suppose that the first step in acquiring a concept of red is to have sense-perceptions of many red things. This is followed by a recognition of the resemblances between them, which in turn leads to the formation of the concept of red. Meyers claims that this process cannot take place, because one must already have the concept of red in order to have a sense-eperception of red. He says, "A sensation is an awareness of a certain sort of appearance. It is thus a cognitive state and can occur only if one already has the concept."
To save concept-empiricism, according to Meyers, the sense-perception (or "sensation") must be taken to be a physical state of the body, and not to be something "cognitive." In that case, we can say that the concept and the cognitive sense-perception occur all at once, after the the occurrence of the physical sense-perception. If these considerations are correct, then even a foundationalist (who embraces concept-empiricism) must explain some holistic processes.
Immanent Justification
Explaining holistic justification is not the greatest problem facing the coherence theorist. "The more basic problem" lies in the claim made by coherence theorists that justification depends only on beliefs. Justification is "immanent" in the sense that it is wholly internal to one's belief system.
Meyers maintains that the mere having of a belief does not increase the credibility of any other belief. He grants that having beliefs does increase credibility, but this is so only when those beliefs have some initial credibility of their own. In the language of justification, this would amount to the demand that directly justified beliefs are needed to justify other beliefs.
The problem is re-stated in terms specific to coherentism. If beliefs get their warrant simply from cohering with other beliefs, then any set of coherent beliefs would confer warrant, "and there is no reason to think one set is preferable to any other no matter how bizarre or fabulous." Meyers quotes C. I. Lewis to back up this charge. There must be some propositions that are "categorically assertable," and not assertable merely on their relation to other beliefs, if there is to be justification.
The problem is not that there are some coherent sub-sets of our beliefs which are bizarre or fabulous. These could be rejected as fictitious on the basis of our other beliefs. Instead the problem is for the belief-system as a whole. "The entire set of our beliefs might be like a well-written novel, if there is no independent source of warrant. Coherence with this set then does not make it knowledge."
Lehrer, it should be noted, accepts this conclusion but denies that it threatens the coherence account of justification. That is, he admits that coherence within a belief system is not sufficient for knowledge, which requires "complete" justification, which in turn requires correspondence between the belief system and reality. But that this fact does not have any bearing on whether beliefs are "personally" justified by coherence.
BonJour's Solution
We have seen that BonJour attaches to his account of coherence an "Observation Requirement," according to which there is no coherence without a stream of fresh input which acts as a check on existing beliefs. In practice, this means that justification requires continual interrogation of the world as a means of testing the beliefs that we have.
The mere presence of fresh input does not by itself guarantee that one's belief system will be any more faithful to the world than a system without such input. BonJour must further suppose that the influence of the input will eventually shake out the beliefs that do not conform to reality. It would break down the coherence initially found in the bizarre and fabulous system.
The problem with this approach is that it is directed against the wrong target. It is not that coherence "leaves our theory of the world underdetermined, even in the long run." This underdetermination is a feature of any account of justification, even foundationalist accounts. As Lewis said, the problem is that coherence accounts give us no external guidance as to what to believe.
Suppose I receive some new input that conflicts with my earlier beliefs. Say, it appears to me that it is snowing in Davis, California, in July. How could this threaten my previous belief that it never snows in Davis in July? Only if I assign the appearance some credibility on its own. Otherwise, I could just dismiss it.
You might ask, though, what would happen if it appeared to snow every day in July. Wouldn't I then be forced to give up my old belief? Only if repetition of similar appearances confers credibility on them. But why must this be? Suppose have a coherent belief system, according to which there is a plot to steal my money by sending me phony utility bills. I receive a real utility bill and toss it on my desk. Then I receive a few more, each with a more urgent message. Must the fact that the pile keeps getting bigger have any effect on my belief that the bills are a hoax?
So unless we depart from the coherence account and give the fresh input some special status that existing coherent beliefs do not have, there is no reason to think that the Observation Requirement would make any difference in winnowing out bizarre and fabulous belief systems. BonJour thought that the additional input would threaten the coherence of a bizarre system. But it would do so only if it is given credibility. And it may be that the system is set up in such a way as to confer no credibility on the input.
BonJour might object on the grounds that Observation Requirement implies that coherence of a belief system requires that one take the input seriously as a test of one's beliefs. He writes, "The Observation Requirement should . . . be understood to include the requirement, common to all adequate theories of knowledge, that a user of the system must make a reasonable effort to seek out relevant, possibly conflicting observations, if his beliefs are to be justified" (The Structure of Empirical Justification, p. 142).
Still, even if a person tries to find relevant observations, the question of whether they are conflicting remains in question. A central tenet of Lehrer's coherentism is that observations by themselves do not play a role in justification until they are interpreted. One can interpret any observation in a way that it does not conflict with the existing belief system.
What BonJour needs is to give observations a special role in justification, something along the lines of taking them at face-value. But if he does this, then adding the Observation Requirement pushes BonJour's account of justification away from being purely a coherence account. Coherence is not, for him, merely a relation among beliefs, none of which is prior to the others. Certain beliefs based on observation are given a privileged role.
Lehrer's Solution
As has been noted above, Lehrer's account of justification is different from BonJour's in the sense that justification is coherence with a system of beliefs, and the coherence of the system as a whole plays no role in justification. Another difference is that for BonJour, an individual belief is justified because there is an argument from members of the belief system as premises to it as a conclusion. These premises are not really more basic or "prior" to the justified beliefs, but are only "contextually basic" (The Structure of Empirical Justification, p. 142). For Lehrer, justification of individual beliefs does not require an argument at all.
Eliminating the role of argument in justification effectively eliminates the regress of reasons, so that no circular reasoning is involved in justification. Higher-level beliefs about reliability of beliefs about color can justify beliefs about particular colors. Beliefs about particular colors can justify higher-level beliefs about the reliability of colors. But the two do not come together in a single circular argument.
Meyers argues that there is still a residual circularity involved in the fact that the only credibility any belief has lies in its relation to other beliefs. No belief has any warrant on its own, in which case "there is no fresh warrant, i.e., no epistemic input." Any new belief that occurs in one's mind would be justified only insofar as it coheres with the existing beliefs, and so it would make no independent contribution.
This criticism does not take into account Lehrer's distinction between belief and acceptance. Beliefs are involuntary and have no significance for justification in themselves. In contrast, acceptance is the outcome of the disinterested quest to attain as many truths and avoid as many falsehoods as one can. One might adopt a strategy regarding acceptance to accord a certain weight to various kinds of inputs, such as the deliverances of sense-perception.
What determines justification is what is accepted. And justification is a function of coherence. But what determines acceptance need not be dictated by what is accepted. It may be, rather, a decision as to what kinds of inputs get what kinds of credibility. Meyers would probably respond that one can make bizarre and fabulous decisions as to what to accept, in which case one would be no better off in the face of the main criticism of coherence theories.
At any rate, Meyers concludes that Lehrer would probably allow that there is a circularity in his account of justification, but not a vicious circularity. This is in line with Sellars's claim that beliefs can be justified all at once, and it may be what BonJour has in mind when he says that justification is "non-linear." (BonJour actually says that to be non-linear is not to accord any epistemic priority to one belief over any other.)
Vicious Circularity
So is the circularity of the coherence theory vicious? Meyers thinks "a good case can be made for thinking it is." An argument is viciously circular when the intended ground of the conclusion cannot be known unless the conclusion itself is known. This description can be carried over beyond inference. "The point is that p cannot provide a ground for q if it cannot be known or warranted without q, no matter how strong the coherence relation between them." There may be circles where q depends on p logically, but not in the order of knowledge or justification.
An example is given of a vicious circle. The Bible states that an event occurred that proves its reliability. But the reliability of this claim that the Bible is reliable depends on the claim from the Bible that the event occurred.
By analogy, the reliability of the mechanisms of sense-perception must be decided on the basis of sense-perception. So if one is to know that sense perception is reliable, one must antecedently know that sense-perception is reliable. We can use any innocent-sounding phrases to describe this situation, but it must be shown why this situation is any different from the biblical case.
A potential solution to this criticism is to deny that in realistic cases, p can provide a ground for q because p can be warranted without knowledge of q. Suppose that q is the proposition that I see a clump of green grass. Part of the justification for q is p, that I am reliable in identifying colors. It seems that although q may play a role in strengthening the justification of q, it is hardly essential to that justification. I have a huge reservoir of evidence for p independently of q.
The kind of cases we have been discussing, then, does not much resemble the Bible case. If so, then there may be no vicious circularity in this kind of case. Meyers could press the issue of vicious circularity by claiming that although no single one of my beliefs based on sense-perception are required to justify my belief in the reliability of my sense-perception, if you took away all perceptual beliefs, then belief in the reliablity of sense perception would not be justified.
A final point is that the coherentist cannot escape vicious circularity by appeal to external factors, such as the truth of the beliefs that make up the system. Coherence accounts make justification internal and immanent. The best that can be said is that the truth of the beliefs coheres with the belief system, and this tells us nothing about whether they really are true.