Holistic vs. Linear Coherentism

I don’t understand holistic coherentism. Here’s what I know: Linear coherentism is the form of coherentism that endorses circular reasoning (provided the circle is big enough). Holistic coherentism is supposed to be an alternative form of coherentism, which avoids foundations, circularity, infinite regress, and skepticism. It makes justification ‘holistic’. (Sorry, I don’t know what that means.) I’ve heard that the holistic coherentist believes:

A belief is justified iff it is part of a coherent system of beliefs.

This view is obviously false, since it implies that either all one’s beliefs are justified, or none of them are. If one’s belief system is coherent, then all one’s beliefs are justified; if it isn’t, then none of them are. I’m sure that the above offset proposition is not what many coherentists genuinely believe.

Leave that aside. What I don’t see is what logically possible alternative there is to (a) foundations, (b) circularity, (c) infinite regress, and (d) skepticism. Stated more generally, suppose there is a relation R, and a thing X that something might stand in R to. Consider the set of things that stand in the ancestral of R to X. This set of things either

(a) contains at least one thing such that nothing stands in R to it;
(b) contains at least one thing that stands in the ancestral of R to itself;
(c) contains infinitely many things that something stands in R to; or
(d) is the empty set.

I take it that what I just asserted is simply a truth of logic, independent of what R is. When we let R be the “is a reason for” relation (or something like it, maybe “is the reason for”, or “is a thing that _____ depends for its justification on”), (a) is the foundationalist structure, (b) is the linear coherentist structure, (c) is the infinitist structure, and (d) is the skeptical “structure.” I don’t see any logical space for the holistic coherentist to be another position. Can someone help explain this to me?


Holistic vs. Linear Coherentism — 25 Comments

  1. Suppose a coherentist says the following. What justifies a belief is its place in a system. So no beliefs are justified except in virtue of its relationship to a system of some sort. Thus there is a relation R between a system and a belief such that standing in R to that system is necessary and sufficient for justification.

    As stated, there need not be any ancestral of R, since a system need not itself be a belief. So far, then, the logical facts above about R don’t tell us anything pro or con about holistic coherentism.

    Even so, I haven’t addressed your first point, the point about the incautious quoted remark (hope it wasn’t me you were quoting…). Anyway, what we want are two things: (i) a specification of the kind of system that is relevant and (ii) an account of what it takes for a belief to bear R to S. Presumably, R is the relation of coherence, but we don’t know how coherence is defined. One way to define it is subject to the problem you mention: that the system is the system of actual beliefs of a person, and if any belief bears R to S then every belief bears R to S. But the coherentist shouldn’t, and needn’t, say this. Here’s one proposal that doesn’t say it (though it doesn’t tell us what we want to know about coherence, either). Let’s assume that the system in question is a system of the propositional contents of beliefs and experiences (I’m going to assume that experiences have such content here, but don’t ask me to give an argument for this!). Then say that a proposition bears R to S just in case S contains INUS conditions for the justification of that claim (conditions that are insufficient but non-redundant for justification that are part of a larger condition which is itself unnecessary but sufficient for justification) and the remainder of what coherence involves is displayed by that proposition together with S. Perhaps, for example, (some of) S explains p or p explains (some of) S.

    Of course, we’ll need a slight modification when the proposition is already in the system, since we’ll have to construct a modified system purged of that proposition. The purging idea requires care in construction but I won’t go into that here.

    The part that is perplexing here is the idea of INUS conditions being part of the notion of coherence. Pretty much, this coherentist idea mimics the foundationalist’s notion of that which prima facie justifies a proposition. It’s just that the coherentist denies that anything short of an entire system conveys any degree of warrant or justification for a proposition, so the coherentist can substitute the notion of an INUS condition for warrant or justification in order to avoid this foundationalist idea.

    OK, much too long here. This view mimics the foundationalist’s relationship between experience and belief, and between some beliefs and other beliefs, so the responses to the regress argument go pretty much the same, substituting the language of INUS conditions for the notion of that which imparts justification to. As a result, it is a holistic theory in virtue of insisting that nothing short of a system imparts any degree of justification and it is coherentist because the relation involved is a coherence relation.

  2. Thanks, Jon. I’m familiar now with your papers on “Coherentists Distractions” and “Can a Coherence Theory Appeal to Appearance States”. I think your two main proposals are (a) that the coherentist should include appearances (or experiences), where these are not beliefs, as part of the ‘system’ that we want to be coherent, and (b) that the coherentist should hold experiences to be only INUS conditions, not sufficient conditions, on the justification of beliefs with the same contents as the experiences.

    I like both of those, actually (or I would if I were a coherentist). Yet I’m still confused about how there can be a fourth response to the regress problem. (Perhaps there isn’t– perhaps the holistic coherentist structure is a special case of the foundationalist structure.) I don’t think that those points actually address the problem.

    Let R be the “is a reason for” (or something like that) relation, because that is the relation that is discussed in the regress argument. (Never mind the coherence relation.) I assume the holistic coherentist does not reject that things stand in this relation, nor that there is an ancestral of this relation–he’s not denying that we sometimes believe things for reasons. And I assume (I think you said this in one of your papers) that the coherentist also does not want to take the implausible view that every belief in your entire belief system is a reason for each of your beliefs. So the holistic coherentist should be able to answer this: Assuming that X is a belief you have a reason for, when you follow it back to the reasons for it, and the reasons (if any) for those things, and so on, what ultimately is the structure of that series of reasons? Do you ever come back to X? Do you ever come to something that there is no reason for?

    On the view you sketched, I’m not sure what the answer is. At some point, as I understand it, you would come back to experiences, which do not themselves have reasons for them, so they act like ‘foundational’ items. However, you also have further beliefs that have to go together with the experiences to produce justification (perhaps beliefs like “my senses are reliable” and “viewing conditions are normal”). That’s okay; but what happens when you follow back those further beliefs to the reasons for them, etc.? Again, do you ever come back again to the original belief (or to any other belief in the series)? If so, then you have circularity too. (The worst of both worlds!) You see my problem?

  3. How about the following view:

    S’s belief that p is justified iff the system of beliefs that it is part of is more coherent when p is included than when p is not.

    Put another way: p is justified relative to some system of beliefs {B} iff {B + p} is more coherent than {B – p}. Then, to judge whether a belief is justified simpliciter, you simply assess whether the proposition is justified relative to (the rest of) the agent’s belief set.

    This strikes me as a plausible enough view, and worthy of the name “holistic coherentism.” I’m no epistemologist though, so do correct me if I’ve said something stupid.

  4. Mike, yes, you’re right that the view I sketched doesn’t provide a different solution to the regress problem from one having a foundationalist structure. The key element to avoid is that, using your terminology, the “is a reason for” relation can’t be interpreted in terms of providing any degree of warrant or justification, by itself, for belief. That’s why I used the language of INUS conditions for warrant. Your terminology will work just as well, as long as reasons aren’t taken to be justifiers.

    In this way coherentist opposition to foundationalism is maintained, even though the response to the regress problem has the same structure as a foundationalist response.

  5. One other thought, Mike–I think you’re right that some typical holistic coherentists have thought that the holistic element itself simply bypasses the regress issue. You’re absolutely right that this won’t work, and Pollock is right that such a move will have disastrous consequences for the basing relation. Holism by itself doesn’t tell us about basing nor does it tell us what to say about the regress problem.

  6. Richard – I think coherentists want something like that. They probably want to require also that {B} itself be coherent, so you have p adding to the coherence of a coherent system. I still think the view is false, of course.

    Jon – Okay, so we have the foundationalist structure but not the doctrine of foundationalism. I am still concerned about whether you also have circularity. Suppose that, in order for sensory experience to (help) justify a belief of mine, I also need, say, the belief that my senses are reliable. Then the question is what happens when you trace back the reasons for this belief (that my senses are reliable). I’m concerned that this goes in a circle – doesn’t it? If it eventually gets back to more sensory experience, then those sensory experiences are going to themselves require this same belief (that my senses are reliable) in order to provide justification for anything, so at that point we’ve gone in a circle.

    That was just an example – perhaps you would say that “my senses are reliable” is not one of the beliefs I need in order for my sensory experience to justify particular beliefs about the external world. Perhaps you would identify some other beliefs as necessary. But the same worry would arise about them. Whatever auxiliary beliefs you need to go along with your experiences, when you trace back the reasons behind these auxiliaries, aren’t you going to have to either go in a circle, or have infinitely many distinct beliefs?

  7. Mike, maybe I’m missing something, but I don’t think there’s circularity. The INUS conditions will trace back to experience, let’s say. Then justification is a matter of cohering with a system, where coherence itself requires the INUS conditions displaying the foundationalist structure. Every belief will be justified in precisely this way (though maybe not all trace back to experience). Maybe the belief that my senses are reliable is in the system of beliefs, but it needn’t be an INUS condition for justification as far as I can tell. And only INUS conditions enter into the foundationalist structure in question.

    The coherentist may also have a way of restricting the class of INUS conditions further, so that some but not all enter into the foundationalist structure in question. The systemic elements beyond whatever elements enter into the foundationalist structure will contribute to whatever coherence involves (or be irrelevant to it), but won’t be the right kind of INUS conditions for them to count as reasons for the belief in question. The coherentist will deny that whatever contributes to coherence of a system is itself a reason for belief, in the sense at issue for participation in the foundationalist structure.

  8. So, Michael, does the sort of account I sketched above involve any foundations, regresses, or circularity? It isn’t skeptical, at least. So doesn’t it satisfy your four conditions?

  9. Jon,
    Okay, but now I’m not sure why this form of coherentism is incompatible with foundationalism. Foundationalists do not say that, when I have a reason for a belief, p, my having that reason is sufficient for the justification of my belief. For instance, another condition is that there should be no (non-overridden) defeaters, but the obtaining of that condition is not typically thought of as one of my *reasons* for p. Another condition is the basing condition, but again, the obtaining of this condition is not one of my reasons for p. It seems to me that the foundationalist would cite my having a reason for p as only an INUS condition on justification. Similarly, he would say my having a sensory experience is only an INUS condition on the justification for a perceptual belief. Did I correctly understand that the opposition between your view and foundationalism turned on the foundationalist’s claiming that experience was sufficient for justification of a perceptual belief?

  10. Richard,
    The view you described does not appear to specify what the structure of reasons is. On the surface, it doesn’t mention reasons at all. So the question would be whether you think people sometimes are justified in believing p because of the reasons they have for it; and then we ask about the structure of that series of reasons.

  11. Mike, exactly the right issue to raise, I think. As I understand foundationalism, the foundationalist claims that reasons for belief impart at least some degree of prima facie justification, and the coherentist denies this, claiming that only a system can do this. So notice that INUS conditions are insufficient for any degree of prima facie justification, as I define them.

    The result of this ploy on behalf of coherentism, I’m tempted to think, is that the foundationalism/coherentism controversy ceases to be as interesting as one would have thought (given the amount of inquiry devoted to it). There is a way of thinking of my proposal as generating something like a terminological variant of typical foundationalism.

  12. Michael, perhaps we could say the following: The fact that {B + p} is more coherent than {B – p} is the reason for S (anyone with belief set B) to believe p.

    (That sounds slightly awkward though. I think I would prefer to take justification – as defined above – as primitive, and then say you have reason to believe what you would be justified in believing.)

    The regress problem only arises if you insist that the reason for a belief derives from the reasons for our other beliefs. But that’s just what I’m denying. As a result, I think I can avoid your four “alternatives.” (Perhaps what I’ve described is different from the “holistic coherentism” you had in mind. But at least it exhibits the same interesting feature of justification, right?)

  13. Oh, unless you’d call these “foundational” reasons. But my account makes no distinction between basic and non-basic beliefs. And it seems odd to say that every reason is a foundational reason. (I don’t mean the account is odd — I happen to like it! Rather, it doesn’t seem that “foundationalism” is the right name for it.)

  14. This is a brief note about the idea that Richard has proposed. I won’t address Jon and Mike’s discussion of reasons and regresses (except to say that Mike seems to me to be doing just fine here). There are serious problems with proposals such as this:

    p is justified relative to some system of beliefs {B} iff {B + p} is more coherent than {B – p}.

    A lot depends upon how “more coherent” gets filled out. But I’ll use my intuitive understanding.

    1) Suppose I have some quite foolish (and unjustified) belief. Adding some equally unupported things that imply or are implied by it would seem be justified, since they would add some coherence. That looks like a bad consequence.

    2) Suppose I believe some things that do not fit together well. I could then increase coherence by dropping either one. So neither is justified, according to the theory. But that seems wrong – one of them might be well justified.

    3) Calculating how coherent (B-p) is will be really puzzling. Suppose I also believe [p.q). Then, (B-p) is going to include p.q, but not p. That looks pretty bad
    Will [B-p] be more coherent than B, even if p is not a reasonable belief?

    I discuss some of this stuff in my textbook.


  15. Richard,
    You gave two suggestions about one’s reasons for believing p:

    The fact that {B + p} is more coherent than {B – p} is the reason for S (anyone with belief set B) to believe p.

    I think I would prefer to take justification – as defined above – as primitive, and then say you have reason to believe what you would be justified in believing.

    Both of these seem wrong – your reasons for believing a proposition cannot be specified in such a general way. For instance, my reason for believing [it will rain tomorrow] might be [the weather report predicted rain]. It is false that my reason for believing [it will rain tomorrow] is [(B + {it will rain tomorrow}) is more coherent than (B – {it will rain tomorrow}]. Your second proposal seems not to identify anything in particular as one’s reason for p. But again, we do sometimes believe things for specific reasons – and they include reasons like “the weather report said it will rain”.
    You also wrote:

    The regress problem only arises if you insist that the reason for a belief derives from the reasons for our other beliefs.

    I don’t understand the point here. I think you may be saying that the regress problem assumes that “is a reason for” is transitive, so that if q is a reason for p, and r is a reason for q, then r is a reason for p. The regress problem does not assume this, because even if “is a reason for” itself is non-transitive, we can by stipulation construct a relation that is transitive – the ancestral of “is a reason for” – and ask about the set of things standing in this relation to p.

  16. Speaking of not quite understanding the point,

    “my reason for believing [it will rain tomorrow] might be [the weather report predicted rain]. It is false that my reason for believing [it will rain tomorrow] is [(B + {it will rain tomorrow}) is more coherent than (B – {it will rain tomorrow}]”

    B= the weather report predicted rain
    p = it will rain tomorrow

    It sure looks like {B + p} is more coherent than {B – p}.
    p does seems to cohere better with my belief B than does
    ~p = it won’t rain tomorrow. And it seems like one way in which beliefs cohere is that one belief follows logically from the another. So why isn’t it true that the reason I believe p is that it coheres with my other beliefs (i.e., in this case, follows logically from my other beliefs)? Or, rather, why is it so obviously false that this coherence is a reason to believe p?

  17. I mean for ‘B’ to denote one’s entire belief set, but assuming this includes the specific belief that the weather report predicted rain, then we still have the required result that {B + p} is more coherent than {B – p}. And while I grant Michael’s point that we usually understand reasons to be more specific than this, I’m not sure that this objection amounts to saying anything more than that most philosophers aren’t holistic coherentists!

    As for the “regress problem”: on the standard picture, I take it, one’s belief that p is justified by some other belief q, but only if q itself is justified. Hence the threat of regress. But it seems that (something like) my account can avoid this, because my account of justification (or reasons) does not appeal to other justified beliefs. It just appeals to coherence with other beliefs, without regard for whether those others are justified or not. (This motivates Rich Feldman’s first objection; but I instead see it as a virtue of the account.)

    Given that the reason for believing p is the fact that {B + p} is more coherent than {B – p}, it makes no sense to ask what the reason for this latter fact is. It’s just a fact, it doesn’t need to be justified in turn. So, we avoid regress. (Don’t we? I admit I don’t know what you mean by an ‘ancestral relation’.)

    Rich Feldman’s objections are well taken. I haven’t really thought this through very well. But perhaps the account could be improved by contrasting epistemic alternatives. Given a choice between inferring consequences from a present unjustified belief, or giving that belief up, it might be the latter that would lead to the most coherent belief set. I won’t fill out the details here, but I think there’s potential to avoid Feldman’s objections by something along these lines.

    Anyway, I take it Michael’s original question was not whether holistic coherentism is a good theory, but whether it’s even logically possible to avoid all four of foundations, regress, circularity, and skepticism. I’m not sure which of the four he thinks my suggestion exhibits.

  18. I think it might be helpful to distinguish at least two different “levels” of relations that go into explaining jusitification on a coherentist model: 1) relations among beliefs within the system by virtue of which the system has whatever degree of coherence it has, and 2) relations that each belief has to the system as a whole by virtue of which each belief has whatever degree of justification it has.

    To keep it simple, suppose we call the first category of relations “inferential” relations. These would be more or less “local” connections between the member beliefs by virtue of which each is inferable from others. Presumably the degree of coherence of the system is proportional, or in some other way determined by, to the number and/or strength of these connections (or something like that).

    But I take it that the holistic coherentist also recognizes a relation or relations between the member beliefs and the system. These are not inferential relationhsips (or at least not the same sort of inferential relationships as hold among the beliefs within the system), because one doesn’t strictly speaking “infer” each belief from the system as a whole, and the justification that attaches to each belief isn’t a function solely of its local inferential connections to others in the system.

    It seems to me that the four-fold list of possibilities that Mike identifies at the end of his initial post may presuppose that there is only one level of relation (rather than the two that a holistic coheretist might endorse) and thus does not exhaust the logical space of possibilities.

  19. Jon,

    Let me try a different tack on the circularity objection (because I’m writing about this now). First, I suppose we’d agree that the following would be an example of a kind of circular reasoning: I claim that I am justified in believing A (solely) on the basis of B and C, I am justified in believing B (solely) on the basis of A and C, and I am justified in believing C (solely) on the basis of A and B. It also seems that we would have the same sort of undesirable circularity if I were to devise a similar example but involving a larger number of beliefs.

    Now, that example illustrates (in the small) a linear coherentist account. The question is whether the holistic coherentist account is importantly different. The holistic coherentist claims that a belief system is justified because of its coherence. What does this coherence consist in? Typically, it is supposed to consist in the presence of relations of entailment, probability-increasing, and explanation among beliefs. These are the very sort of relations that are typically displayed in inferences (minus a basing condition). So it appears that the holistic coherentist claims that a belief system is justified by virtue of instantiating the very sort of relations that the linear coherentist was displaying with his circular reasoning. Perhaps the holistic coherentist can deny that these relations make the rest of one’s beliefs *reasons for* any given belief since a basing condition is not required for the given belief to be justified, but this does not seem much of an improvement over the linear coherentist model.

    In my example in the first paragraph, suppose we agreed that my account of how I am justified in believing A, B, and C fails due to vicious circularity. How plausible would it then be for me to propose, instead, that the whole *system* consisting of beliefs A, B, and C is justified by virtue of B and C supporting A, A and B supporting C, and A and C supporting B?

  20. A holistic coherentist (or at least one type of holistic coherentist) might say that while the inferential connections within the system of beliefs form a closed curve, the justification of each belief is noncircular because (1) there is a noncircular reason for believing that the set of beliefs is coherent and (2) there is a noncircular reason (e.g., an a priori reason) for believing that a system of coherent beliefs is likely to be true.

    There are, of course, many good reasons for being suspicious that there could be noncircular reasons for either (1) or (2), but I take it in this discussion we are focusing merely on the structure of the view. The point is that, on at least one variety of the view, circularity of inferential relationships within the system doesn’t translate into circularity in the justification of beliefs, because the relations by which beliefs are inferred have a different structure than the relations by which they are justified.

  21. Mike, yes, you’re right that many of the same relations will appear in the account of coherence that holists will need. If the basing requirement is clarified separate from the coherence requirement, the coherentist will be able to clarify proper basing so that circularity is ruled out. So, the difference is that the coherentist will have to use basing to rule that out, whereas the foundationalist can use the epistemic support relation itself to rule it out.

  22. Mike is my (M.A.) thesis advisor, and I’m writing on epistemic circularity, so he’s probably decided to scoop me. But even so, I recalled a few footnotes on precisely this topic, so I thought I’d throw them out, in particular because they point to further discussion. I hope I’m not forcing Bergmann into this thread against his will. This is from Bergmann’s “What’s NOT Wrong with Foundationalism”, linked here:


    The first footnote reads “See BonJour 1985, 89-93 on the distinction between holistic and linear coherentism. As Sosa (1980), Plantinga (1993, chapter 4) and Klein (1999 and 2000) rightly point out, holistic coherentism is just a version of foundationalism.”

    In “Epistemic Circularity: Malignant and Benign” this claim is again endorsed, and he adds that holistic coherentists accept the Regress Argument. There seems to be a minor (probably superfluous) qualification of the claim that holistic coherentism is a version of foundationalism at fn. 9:

    “See Plantinga 1993a, ch. 4 and Klein 1999, 298 for a defense of the view that what I’m calling ‘holistic coherentism’ is a version of foundationalism insofar as it disapproves of circular reasoning and infinite chains of reasoning while endorsing F [non-inferential justification].”

  23. Scott, this idea that holistic coherentism is a version of foundationalism is, I think, a mistake. What a theory says about reasoning and proper basing can be held distinct from what it says about the conferring of justification. That’s the point of talking about INUS conditions above.

  24. Pingback: Certain Doubts » September Stats

  25. Pingback: Show-Me the Argument » Holistic Coherentism and Circular Reasoning

Leave a Reply

Your email address will not be published. Required fields are marked *