Find EnlightenmentPrevious Chapter



I. Justification Standards.

In the last few decades, epistemologists have focused much attention on the notion of justified belief, often, but not only, to inform their accounts of knowledge. Discussions in more recent years, however, have made it plain that `justification' has some unwieldy ambiguities. One of these ambiguities, much-discussed, is that between justification as a deontological notion (one about how we are permitted or ought to believe; this has been associated with internalism ), and as a nondeontological notion (one about whether our beliefs are in a proper, e.g., causal, relation to the facts; associated with externalism ). But if any consensus has been built on this issue, it is that justification is indeed a deontological notion,[1] and in any case that is an assumption I shall be making in the following.

William Alston makes a worthy attempt at clarificatory reform in his article "Epistemic Desiderata."[2] Alston's basic point is that different epistemologists have had different – if related – concepts at which their theories are aiming. Thus, rather than quibbling over which concept to attach to `justification', for example, it would be far better to let that word go and focus instead on what he calls `epistemic desiderata'. In effect, Alston recommends that we give accounts of epistemically desirable states, states that are best expressed in entire sentences. Rather than formulating conditions for justification, Alston would have us focus on more precisely-described, but clearly epistemically valuable states, such as `S has adequate grounds (reason, evidence) for the belief' and `S's belief that p was formed in a reliable way'.

This is an extremely valuable suggestion but not one that I will be able to act upon – though my overall discussion gives it very roundabout support. This dissertation will focus on justification, since very many theories of positive epistemic status in the twentieth century were couched in terms of justification. This might turn out to have been a mistake (and by the end of Chapter 4 I hope to have made it clear why this has to some degree in fact been a mistake), but for the present I will follow accepted practice and usage.

`Standard' is used here as a catch-all term that can cover various manner of expression that epistemologists are prone to: `principle', `rule', `criterion', `norm', or even `desideratum'. The particular form of justication standard of which I will speak – for purposes of clarity and uniformity – is the following:

(JS) If S's belief that p meets (nonepistemic) conditions c (where c might include subjective or nonsubjective states of S and S's environment) at time t, then S is justified in believing that p at t.

So when I use `justification standard', `standard of justification', and `j-standard', I should be taken to mean something of the form (JS). Whether one wishes to say that this is a partial reduction, analysis, or definition of `justified belief', or a statement about supervenience conditions, is irrelevant to my project; what I have to say will apply equally to such epistemic generalizations, no matter how they are regarded.

Obviously, far from all claims that epistemologists have been concerned to argue about can be directly expressed in the form (JS). But the points I have to make about claims of this form can be applied, more or less straightforwardly, to at least many claims of other forms. Note that the problems I shall raise concern standards that do not merely translate talk of justification into some other epistemic terms (such as having evidence); the problems I shall raise concern standards that are attempts to explain justification in nonepistemic terms.

II. The Problem of Meta-Justification.

I will discuss at length The Problem of Meta-Justification (or PMJ ), which may be described, very roughly, as the problem of determining how we can justify standards of justification without ultimately begging the question. To help elaborate this rough account, I invite the reader to consider two closely related questions:

(1) For some standard of justification, what is it in virtue of which belief in it is, or could be, justified?

(2) For some standard of justification, how can we justify it?

Let us briefly set aside the differences between these two questions and discuss why we should care about any such question at all. The same curiosity that properly motivates investigation into first-order epistemological questions – e.g., "In what circumstances is a visual belief that p justified?" can and should also motivate investigation into second-order epistemological questions. If we are well-motivated in asking what the conditions of justification are, and (acting on our motivation) we produce a j-standard, then we would appear to be equally well-motivated in asking how we are justified in believing the j-standard.

Now consider the difference between the two questions. Question (1) asks for an account of that in virtue of which belief in a standard is justified; (2) asks that we justify the standard. Question (1) asks that justifiers be adduced, or at least that it be shown that they exist; (2) asks for an act of justification. That the questions are different is clear, even if it should turn out that they require the same sort of answer.

Indeed, one might well think that the questions require the same sort of answer: namely, something that can be converted into an argument that has a j-standard as its conclusion. I wish to defend this view. It assumes, in the case of (1), that

(1a) Whatever it is in virtue of which belief in a standard of justification is (or could be) justified, can be properly interpreted and evaluated as an argument.

and in the case of (2), that

(2a) The act of justifying a standard of justification is, or may be interpreted and evaluated as, giving an argument.

In what follows, I will examine the merits of (1a) and (2a).

To begin with (1a). If one does adduce the justifiers of a belief in a standard, then regardless of whether they are presented as premises of an argument in which the standard is the conclusion, nonetheless the justifiers-cum-standard – the standard's meta-justification – may be treated as an argument. Here is why. If the same relation of support that must hold between the premises and conclusion of a good argument does not hold between the adduced justifiers and the standard, then regardless of how the latter is presented, the justification will be rejected. The standards of successful support are the same as the standards of good argumentation. So other philosophers may interpret and evaluate the justifiers (regardless of how they are billed) as premises of a meta-justificatory argument.

Admittedly, evaluating a proposed meta-justification in this way might require considerable and difficult interpretation. If, for example, someone insists that it is coherence with a doxastic system that justifies a coherentist j-standard, it is not immediately obvious how this coherence-cum-standard is to be formulated as an argument. And unfortunately when philosophers do take up the Problem of Meta-Justification their attempts are often rather sketchy. But in order to evaluate their claims that their standards are successfully justified, such an argument – or some sort of story that can be converted into an argument – must be spelled out in sufficient detail. The coherentist should not expect us to accept on his say-so, or with only sketchy generalities, that his standard is successfully supported by the coherence of his (or some) doxastic system.

It is now common for externalists to insist – what is admittedly not obvious at first glance – that in some circumstances it is possible to be justified in a belief even when one cannot say what the justification is (what the justifiers are). Perhaps first-person appearance beliefs (e.g., "It seems to me I am seeing something orange and round") have such a justification. In such cases it is possible, perhaps, to adduce the justifiers for the belief, but obviously one should not attribute an explicit argument to the believer. So externalists might on these grounds disagree that a meta-justification ought be interpreted, and evaluated, as an argument. Our externalist might say that it is possible to be meta-justified in believing a standard, without being able to say what the justification is.[3]

As powerful as this basic externalist insight is for solving other problems in epistemology, it is of little help here. Suppose I advance (for example) a reliabilist j-standard, and when I am asked, "What justifies you in believing that?" I reply, "The fact that my belief in this standard is the result of a reliable process." Suppose, however plausible that might be, that in the next breath I insist that I need not be aware of these justifiers, and that hence they are not part of an argument. We scarcely know how to reply to such a move. Are we to evaluate whether the alleged meta-justification succeeds in supporting the standard (because it is presented as doing so), or aren't we (because it is denied that it constitutes an argument)? At any rate, until the meta-justification is actually spelled out in such a way as can be construed as an argument, we shall be unable to determine whether it actually does support the standard.[4] So, for all I have said so far, the reliabilist could be correct that his standard is meta-justified; but until he specifies what the meta-justification is, he hasn't done anything interesting philosophically.

Perhaps someone might try to bypass these concerns by arming himself with an existence proof: he can show that a meta-justification for a standard exists, even if he cannot say what it is. His proof is not an argument for the standard (and is not, thus, itself a meta-justification); it is, rather, an argument that a meta-justification exists.

Notice, however, that this does not contradict (1a), which says that a standard's justification can properly be expressed and evaluated as an argument. Indeed, the existence proof imagined here does not answer the original question (1): what is it in virtue of which belief in the standard is, or could be, justified? Simply to be told that the meta-justification exists does not satisfy our curiosity about what it is. Indeed it would only increase that very curiosity by showing that there is a meta-justification to be formulated. After all, we are probably not seriously in doubt about the meta-justifiability of some basic standards; what piques our curiosity is precisely how their meta-justification works.

Next consider (2a), according to which the act of justifying a standard is, or may be interpreted and evaluated as, giving an argument. This is not always true, at least according to the way that `justify' has been used by some philosophers. In his classic essay "De Principiis Non Disputandum...?",[5] Herbert Feigl distinguishes between two senses of `justification': validation and vindication . The Problem of Meta-Justification can be neatly solved, if we are willing to justify standards in the sense of vindicating them; validating standards might require arguments for them, but vindication does not.[6] Let us see what merit there is to this move.

As far as I can make out, for Feigl, to validate is to give a good argument, while to vindicate is to offer a kind of "`pragmatic' or `instrumental' justification"[7] according to which it is shown that the adoption of the principle (up for vindication ) is the best means to attaining some desired end. Whether a j-standard is successfully vindicated, then, depends on what end it is taken to serve, and on how well it serves that end.

A dilemma may be used to show that an argument (or something that can properly be regarded as an argument) will be required in any justification of the sort that satisfies a typical epistemologist's curiosity about the PMJ. Assume for both horns of the dilemma that someone advances a j-standard and then attempts to justify it by vindicating it.

Suppose, on the one hand, that the end that the standard is supposed to serve is a truth-linked quality; in other words, the end in question is truth, probable truth, or some epistemic quality that is supposed to secure one of these. If this is the end to which the vindication is directed, there is no substantial difference between vindicating and arguing. For the principle to be vindicated in this sense is simply to show that adopting the principle is an excellent means to gaining a true, or a probably true, belief. The vindication may then be evaluated in just the way that an argument is evaluated.

Here it might be objected that, notwithstanding the fact that the vindication may be evaluated as an argument, nonetheless the vindication is not itself, and is not intended as, any manner of argumentative support. But it does not matter how this sort of vindication is intended. It succeeds only if it can in fact serve as argumentative support for the standard. Correspondingly, the extent to which it fails to offer argumentative support for the standard is precisely the extent to which it fails as this sort of vindication.

Suppose, on the other hand, that the end that the standard is supposed to serve is not a truth-linked quality – for example, the end might be to secure as much pleasure for the believer as possible. In that case, the vindication of the standard, no matter how successful, will not satisfy a typical philosopher's natural desire to justify the standard. Evidently, when we ask how we might justify a standard of justification, we want to be told something other than that the standard will give us pleasure. That might be nice to know but it is irrelevant to what we were asking about. In short, we are seeking an epistemic justification, which excludes some kinds of vindication.

Hence, if someone advances a vindication of a belief in a j-standard, then either the vindication may be treated as an argument, or else it is not the sort of thing that we asked for, that will satisfy our curiosity.

The considerations of the last few pages are intended to support the claim that, however exactly the question is formulated, when we get curious about the justification of standards, it is only good arguments for those standards (or what may serve as such arguments) that will satisfy us.

Some more prosaic considerations can help convince us of the same thing. Epistemologists habitually advance standards of justification, and other epistemologists habitually call them to task for it, not only stating specific objections to those standards, but also asking for and evaluating positive arguments for the standards. And so, however all the talk about justification, justifying, validation, and vindication might be, we do as a matter of course require arguments for the standards we advance. To say this is not to argue that such behavior is rational – but it does, at least, lay a heavy burden of proof squarely on whomever wishes to deny that some particular j-standard does not require argument in order to be justifiably held or propounded. We certainly do not, for example, let reliabilists off the hook simply because, according to their theory, they might be justified in accepting their theory without knowing that that are so justified.

Philosophers are not, of course, the only people who accept j-standards. At the very least, scientists, lawyers, and other intellectuals accept various standards as well. And indeed such people could be justified in believing their standards without being able to produce anything like a rigorous argument for those standards.[8] Why not philosophers as well, then? It is the unique, special task of philosophy to face such problems as the PMJ. For a philosopher to renounce such a problem without giving it any serious consideration is to propose a radically different conception of what it means to be a philosopher.[9] So naturally, in our capacities as philosophers at least, we do not want to know simply whether we are justified in accepting our standards of justification but whether we can support those standards with good arguments, or what the meta-justification is.

So henceforth I will treat meta-justifications as arguments for j-standards.

One more point of discussion about the PMJ is apropos here. We might well imagine someone offering the following argument:

The PMJ is the problem that, when we try to go very deep justifying our j-standards, there is a circularity problem involved in the attempt: any belief in a j-standard that is justified will be justified in accordance with a j-standard. Now, ordinarily, in order to be justified in holding a belief (including a belief in a j-standard), one doesn't have to justify all the supporting beliefs all the way down; one can just take a lot of beliefs (including beliefs in standards) for granted. So on the ordinary sense of `justification', there is no problem about meta-justification. On a stricter (indeed, impossibly strict ) sense, which requires that one justify all the justifiers for any belief in order for the belief to be justified, it is obvious that one can't fulfill the requirement, since one will obviously have to take some j-standard for granted. That just means we can reject the stricter sense of justification, leaving us with the ordinary sense. In that case, we can indeed just take some standards for granted. So there's no real problem here; why go on talking about it, then?

This is, in fact, rather similar to my own approach to the PMJ. But I maintain that there is a real problem, or at least, that there is a lot of real work to be done in explaining why it is not such a problem after all.

Anyone who offers the above sort of argument has a lot of questions to answer. In what sense would a meta-justification involve us in circularity? If it is not simply premise circularity, then is there really anything wrong with it? (If there's nothing wrong with the circularity in question, perhaps we can provide a meta-justification in the stricter sense of `justification'.) Perhaps according to an ordinary sense of `justification', we can take a lot of beliefs for granted and still be justified in holding those beliefs – but what sorts of beliefs can be taken for granted? Those are just a few of the more obvious questions that come to mind; as we will see, there are many others that come up in an in-depth exploration of the PMJ.

So if the conclusion we will arrive at in Chapter 4 – that we will simply have to take some standards for granted already appears obvious, one should bear in mind that simply saying this does not constitute an adequate discussion of the problem, nor does it address the various bold attempts philosophers have made to solve it head-on. It might turn out that the PMJ, like many philosophical problems, is indeed another pseudo-problem; but in order to be fully justified in making that claim, an in-depth exploration of the problem is needed.

III. Track Record Arguments.

We advance j-standards and their meta-justifications for a good reason: it is precisely these standards that purport to say when we are justified and hence on the path to truth. Truth is what we are ultimately after, and the standards purport to tell us when we can say with impunity that we have it (or at least, under which conditions it exists, regardless of whether we can say this or not). This is why so-called criteria of truth can be regarded as falling under the purview of my discussion of j-standards.

That same desire for assurance of the truth of our beliefs also motivates epistemologists' attempts at meta-justification. It would indeed be puzzling for a philosopher to go to the trouble of distinguishing justified from unjustified beliefs, and then leave that work itself unargued-for – hence leaving his belief in his work itself possibly unjustified.

But attempts at, or descriptions of, meta-justificatory argument have led to the discovery of some famous circularities – the problem of the criterion, Descartes' circle, and the problem of the justification of induction, to name just three. These all display a single kind of circularity: epistemic circularity. To come to grips with such circularity, then, we must examine the sort of meta-justificatory argument – that gives rise to it.

I do not propose (the impossible) to examine each argument for every standard that has been proposed, or even to begin by examining a few. Instead, I will present the form of what is perhaps the most obvious inductive type of argument for standards: the so-called track record argument. In presenting this argument my goal is ultimately to explain what epistemic circularity is. (And thus my goal is not to reconstruct anyone's actual argument for a j-standard.)

Since they are generalizations, the most obvious – not to say the only – way to argue for justification standards is by inductive generalization. The strategy is to list a number of specific instances where the desired conditions hold, and where the subject's belief is justified; then generalize. The result is a track record argument . The argument's form is thus:

(1a) S's belief that p1 meets conditions c at time t1.

(1b) S is justified in believing that p1 at t1.

(2a) S's belief that p2 meets conditions c at time t2.

(2b) S is justified in believing that p2 at t2.

(na) S's belief that pn meets conditions c at time tn.

(nb) S is justified in believing that pn at tn.

(C) Generally, if S's belief that p meets conditions c at t, then S is justified in believing that p at t.

The "conditions c" here can refer to any of a variety of epistemically relevant conditions: cognitive states of S as well as facts about S's physical environment, reliability, proper functioning, habits, history, or other circumstances.[10]

The obvious question to ask is: "How was it determined that S's belief was justified for each (nb)?" In other words, there will be a demand for a systematic way to confirm, or at least to account for the possible justification of, each particular justification claim in the premises.[11] A very simpleminded reply[12] to this would be to specify, for each (na)/(nb) pair, a conditional "bridge principle" to the effect that

(BP) If (na), then (nb). (If S's belief that pn meets conditions c at tie tn, then S is justified in believing that pn at tn.)

After all, it is the fact that a belief that p meets conditions c that, according to the theory advanced in the conclusion, accounts for why the belief is justified.

But (BP) would just be an instantiation of (C) (as can be seen by inspection). One might hope for further support for each such instantiation of (BP), but this seems unlikely at best.[13] So one might as well consider any particular instance of (BP) as an arbitrary case; in other words, the purveyor of the track record argument appears prepared to advance the claim, for any of S's beliefs that meet conditions c at time t, that it is justified precisely because it meets those conditions. But then (C) would follow immediately by universal generalization, which would appear to show that the original argument was question-begging or circular.[14] In what precise sense it was circular remains to be explored.

Epistemologists faced with the problem of arguing for each (nb) in a track record argument – are apt to try a more oblique route. Some might, if pressed, suggest a pair of "bridge principles." The first would state that (na) implies that the belief is probably true (given the conditions of belief); this would specify how the conditions specified as epistemologically important are indeed truth-linked.[15] The second in turn would state that the probable truth (given the same conditions) of the belief implies (nb) – that the belief is justified. I take it that this is implicit in many epistemologists' attempts to argue for their epistemic standards; having allegedly shown the qualities they regard as epistemically important to be truth-linked, they infer that those qualities are what make beliefs justified (or warranted, etc.).

For example, a reliabilist might wish to argue for the following (still formulaic) version of (BP):

(BPR) If S's belief that pn is the result of a reliable belief-forming process (such as sense-perception, memory, etc.) at tn, then S is justified in believing that pn at tn.

My suggestion is that the reliabilist would argue that

(3) If S's belief that pn is the result of a reliable belief-forming process at tn, then the probability of the truth of S's belief that pn, given that it is the result of such a process, is >0.5.

Furthermore, as a reliabilist he holds:

(4) If the probability of the truth of S's belief that pn, given that it is the result of a reliable belief-forming process, is >0.5, then S is justified in believing that pn at tn.

From which (BPR) follows.

But this again allows our reliabilist to forego the earlier track record argument altogether: arguing for any arbitrary pn from (3) and (4) to (BPR) would constitute an equally good argument for a reliabilist version of (C). Now suppose our reliabilist – can defend (4) on some linguistic grounds (it expresses part of what `justified' means, say); he will still have to defend (3). He will have to show that a process of belief-formation that he regards as reliable does in fact elicit more true than false beliefs. And here again the most obvious sort of argument is a track record argument, which differs in a few details from the track record argument outlined above.

For purposes of clarity I will present the form of this second sort of track record argument:

(1a) S's belief that p1 meets conditions c at time t1.

(1b) p1 is true at t1.

(2a) S's belief that p2 meets conditions c at t2.

(2b) p2 is true at t2.

(na) S's belief that pn meets conditions c at tn.

(nb) pn is true at tn.

(C) Hence, if S's belief that p meets conditions c at t, then p is probably true at t. Or: the probability of p's being true, given that S's belief that p meets conditions c at t, is >0.5.

This argument is, perhaps, a little less obviously problematic than the first track record argument. But a similar question again arises: "How was it determined, for each (nb), that pn was true?" The epistemologist advancing the argument naturally believes he has a test for a claim's truth, namely, whether S's belief in that claim is justified; after all, justification is truth-linked on his view. And he believes he knows when a belief is justified, namely, when it meets conditions c. So the way to determine for each (nb), whether pn was true, is to see whether the belief that pn met conditions c; if the conditions are met then the belief is deemed true. But of course this is none other than the bridge principle – that besmudged our earlier track record argument with the charge of circularity.

A more concrete example should help clarify the problem. Suppose our reliabilist wants to argue that, when a belief is the result of something visually appearing to be the case, in excellent conditions and with no defeaters, then the belief is true. He amasses a number of cases such as the following. (1a) I believe there is a green teacup on my desk; I seem to see a green teacup on my desk; there is plenty of light, the air is clear, I am fully awake, etc., etc.; and I am aware of having no reason whatever to disbelieve that there is a green teacup on my desk. And, moreover, (1b) there is a green teacup on my desk.

When asked how I confirmed (1b), that there is a green teacup on my desk, the obvious (not to mention factual) answer is that I looked on my desk and saw the teacup. But then it was precisely a belief such as (1a) that confirmed (1b). Now, if our reliabilist maintains that (1a) actually does provide the needed support for (1b), then he also assumes that (1a) implies (1b). But this latter is simply a bridge principle – that instantiates (C) – what he was trying to argue for. And if it holds in the arbitrary case of the teacup, (C) follows immediately by universal instantiation. The reliabilist was assuming, or presupposing, his conclusion. So he was, in a certain sense, arguing in a circle.

None of what I have said so far should be taken to imply that all arguments for j-standards are so obviously subject to epistemic circularity. For all I have said so far there might be many that do not encounter the problem at all. I have spoken about one particular way of arguing for j-standards – but presumably, the most obvious way – in order to have a plausible context in which to discuss the topic I shall take up in the next section of this chapter, viz., epistemic circularity itself.

IV. What Epistemic Circularity Is.

The term `epistemic circularity' is a technical term, and hence its definition is open to some stipulation.[16] Nonetheless, the term serves a specific function, and many cases of its application are easy to recognize and difficult to dispute. We want a definition of `epistemic circularity' to cover those obvious cases. The definition should also identify what it is about an argument that impels us to apply the epithet. Given all that, there are grounds to dispute about different accounts of epistemic circularity and to argue for a better one, if available.

The task before us now is to decide what it is about the track record arguments explained in the foregoing section that is epistemically circular . To my knowledge it was James Van Cleve who coined this term[17] (though by no means is he the first to have identified the concept). There is no shortage of formulations of what epistemic circularity is in the literature; two prominent accounts are by Van Cleve and Alston. According to the Van Cleve, an epistemically circular argument is an argument such that

a necessary condition of using it to gain knowledge of (or justified belief in) its conclusion is that one already have knowledge of (or justified belief in) its conclusion.[18]

According to the Alston, epistemic circularity

involves a commitment to the conclusion as a presupposition of our supposing ourselves to be justified in holding the premises.[19]

Let me begin my own discussion of epistemic circularity – by repeating a few rather obvious things about it. First, however plausibly characterized, epistemic circularity is evidently different from what has been called premise circularity – in other words, that sort of circularity in which the conclusion, or some single claim that is the roughly same in meaning as the conclusion, is found among the premises. The conclusion of an epistemically circular argument need not be found in any form among the premises; that can be seen rather easily in the track record arguments discussed in the foregoing section.

Second, epistemic circularity is different from what has been called rule circularity – that is, that sort of circularity afflicting an argument where the conclusion is in the form of an inference rule, and that very rule licenses the inference from the argument's premises to its conclusion.[20] For example, rule circularity would afflict an argument in which modus ponens, stated as a theorem, were the conclusion, and the argument for this conclusion were itself in the form of modus ponens.

How then is epistemic circularity different from rule circularity? The mere fact that I use a certain inference rule in coming to a conclusion, as is necessary for rule circularity, does not in itself entail that I must have knowledge of that rule in order to get a justified belief by its use. I can use a rule without even knowing I am doing so. For example, I might conclude, from the knowledge that Joe had either soup or salad, and that he did not have salad, that he had soup; but to conclude this I evidently do not need even to be aware of disjunctive syllogism as a rule of inference. So, for Van Cleve at least, it is possible to have an argument that is rule circular but not epistemically circular.[21]

Here is a reply to the foregoing: to get a justified belief from an argument, one must not only believe each premise justifiably, but one must be justified in believing that the inference is correct.[22] But this reply cuts no ice, simply because one can be justified in believing that the particular inference is correct without knowing that the covering rule is correct.

Consider next how Alston – can draw the rule/epistemic circularity distinction. The mere fact that an argument is rule circular does not entail that we are committed to the conclusion as a presupposition of our supposing ourselves to be justified in holding any of the premises. Again, it is possible to get justified beliefs from arguments made when one is scarcely aware of just which rule one is following. So, on Alston's account too, it is possible to have an argument that is rule circular but not epistemically circular.

But according to my own account, yet to come, it might turn out that all rule circular arguments are epistemically circular; still, I doubt that this is a criticism of my account. Perhaps we should consider all rule circular arguments epistemically circular after all. Moreover, the fact remains that there are epistemically circular arguments that are not rule circular; examples were given in the previous section.

I have argued that epistemic circularity is different from premise and rule circularity. So what, exactly, is epistemic circularity itself? Van Cleve's and Alston's accounts are significantly different from each other; and there are some difficulties with both accounts. So, after examining their accounts, I will offer my own.

For Van Cleve, a necessary condition of using an epistemically circular argument to gain a justified belief in its conclusion is this: one must already have a justified belief in the conclusion in order to be justified (by the argument) in believing the conclusion. But on Alston's view, the necessary condition is this: the conclusion is presupposed true if one is justified (by the argument) in believing the premises.[23] Van Cleve focuses on the conclusion, and Alston on the premises; who is right, or does it even make a difference?

Van Cleve's account can in a sense accommodate Alston's. Suppose that for a certain argument a necessary condition of being justified in believing the premises is that one already believe the conclusion justifiedly. But now, since the premises are used to support the conclusion, from this it follows that a necessary condition of being justified in believing the conclusion on the basis of this argument is that one already be justified in believing the conclusion – as Van Cleve has it.

Similarly, suppose that for a certain argument, in order to be justified in believing that the inference from the premises to the conclusion is correct, one must already believe the conclusion justifiedly. But to be justified in believing the conclusion on the basis of this inference, one must be justified in believing this particular inference to be correct.[24] From these two claims it follows that a necessary condition of being justified in believing the conclusion on the basis of this argument is that one already be justified in believing the conclusion – again, as Van Cleve has it. If there were any such arguments, I think they would be epistemically circular; but on Alston's account they would not be.

Let me present some preliminary findings. Generally, I am discussing how one may be justified in believing a claim on the basis[25] of at least two sorts of beliefs about an argument, namely, (1) beliefs in each of the premises and (2) the belief that the premises are properly connected to (support) the conclusion. Both sorts of belief (at least) have to be justified, if one is to be justified in believing the conclusion on the basis of the argument.

For some arguments, one must either already be justified in believing, or at least presuppose (for Alston ), the conclusion, in order to be justified in believing the argument's premises or inference. For any such argument, it follows that one must already be justified in believing, or presuppose, the conclusion in order to be justified in believing the conclusion (on the basis of this argument). And, in any such case, it seems, the argument in question is epistemically circular. I will include these insights in my own account of epistemic circularity.

Now I come to a different point. For Van Cleve, epistemically circular arguments are such that one must already have knowledge of or a justified belief in the conclusion. Van Cleve's notion is stronger than Alston's, according to which one must only presuppose the conclusion. Again, who is right, or does it make a difference?

I think it does make a difference. Van Cleve introduces his notion of epistemic circularity with these words: "Under what circumstances is an argument viciously circular? I submit that it is so under one circumstance only: a necessary condition of using it to gain knowledge of (or justified belief in) its conclusion is that one already have knowledge of (or justified belief in) its conclusion."[26] Van Cleve is surely right to say that epistemic circularity is vicious,[27] on this account. Then he goes on to maintain that certain inductive arguments for induction not epistemically circular on this account.

But on first consideration of typical sorts of examples of epistemic circularity, it does not seem quite as obvious that the circularity is vicious as it is on Van Cleve's account. So if it is possible to define `epistemic circularity' in such a way that its viciousness does not in effect follow by definition, then we should do so.

Alston's approach is different. He says that epistemic circularity – involves only being committed to the conclusion as a presupposition, which might, perhaps, be a kind of belief. Presumably it need not be a justified belief. So it appears to be, as far as his definition goes, an open question as to whether epistemic circularity is vicious. This approach is better.

Another reason for preferring Alston's approach is that we want a term that actually does apply to a variety of different arguments supporting j-standards, or supporting claims about the reliability of sense-perception, induction, and other cognitive practices. But at the same time, if we do want to say that it applies to arguments regarding these various standards and cognitive practices, we will not want to have the term defined in such a way that epistemic circularity is vicious – by definition. Even if it should turn out that all epistemic circularity is vicious, we should have to argue for that point, if for no other reason than that some distinguished philosophers disagree with it.

To these considerations I wish to add an insight of my own, inspired by my examination of track record arguments: to be justified in believing one or more of the premises of an epistemically circular argument, it must actually be the case that the conclusion is true. This leads me to suspect that it is irrelevant whether one have, prior to making the argument, a belief in the conclusion (and perforce it is irrelevant whether such a belief were to constitute knowledge, as Van Cleve has it). For, when we examine an argument, in order for us to pronounce it epistemically circular, it is sufficient that any justified belief in the premises (or the inference step) requires that the conclusion to be true. This in fact is just what happened – twice in our discussion of track record arguments.

Philosophers use epistemically circular arguments in an attempt to show that beliefs in various conclusions are justified. But after examination, we know that in order for us to use an argument to gain a justified belief in the argument's conclusion, we must also have a justified belief in the premises (and inference). And we realize that, in order for us to have a justified belief in at least one of the premises (or the inference), the conclusion must be true. That realization by itself, it seems, is enough to impel us to charge the argument with epistemic circularity. There is no need to consider whether the conclusion was antecedently believed; that's irrelevant to the argument's circularity.

Suppose a student used an epistemically circular argument to argue that sense perception is reliable. This student had no belief about the conclusion before constructing the argument, we will say. Moreover, the student did not see that his conclusion had to be true, in order for his belief in some premise to be justified. Surely that failure of insight on the student's part would hardly persuade a teacher, who did have the insight, that the student's argument was not circular. So there is no reason to hold that the proponent of an epistemically circular argument need antecedently know or for that matter believe the conclusion of the argument.

On Alston's account, we are committed to the conclusion as a presupposition  of taking ourselves to be justified in believing one or more of the premises. But an even simpler account is available. There is no need to puzzle out what our commitments are or what `presupposition' means exactly. It suffices simply to note that the claim that there is a justified belief in one or more of the premises implies that the conclusion is true.

So I propose the following account of epistemic circularity, similar to but also importantly different from both Van Cleve's and Alston's:

Def. An argument A for conclusion c (understood by S) is epistemically circular for S iff (i) if S were justified in believing that c on the basis of the set of S's beliefs about A, then S would be justified in believing each of A's premises and that A's inference is correct (= justifying beliefs about A), and (ii) for at least one of the justifying beliefs about A, if S were to have a justified belief in it, then c would be true.

The purpose of clause (i) is to identify a set of beliefs as those which, allegedly, justify S's belief in the conclusion of an argument. Clause (ii) then states the essential characteristic of epistemically circular arguments: the truth of the conclusion is a necessary condition of the justification of one of the aforementioned justifying beliefs.

George Schumm has produced an interesting possible counterexample to this definition:

John has a belief.
Therefore, someone has a belief.

This argument satisfies the conditions of the definition – if someone has a justified belief in the premise, it follows that the conclusion is true – but it is far from obvious that it is epistemically circular. For anyone who is unwilling to say that Schumm's argument is epistemically circular, it constitutes a counterexample to my definition.[28]

In that case, however, Schumm's argument would also be properly regarded as a counterexample to Alston's definition of `epistemic circularity' as well. As we have seen, according to Alston's definition, an epistemically circular argument is one in which we are committed to the conclusion as a presupposition of our taking ourselves to be justified in holding the premises. In the case of Schumm's argument, we are definitely committed to the conclusion – that someone has a belief – as a presupposition of our taking ourselves to be justified in believing the premises. If we suppose ourselves to be justified in holding a belief, we presuppose – that someone has a belief (we do).

So two distinct definitions of `epistemic circularity' have the same interesting feature, namely, being open to challenge using Schumm's argument. This might be regarded as some evidence for the view that in fact Schumm's argument is not a counterexample at all. But plainly such a view would have to be explained. After all, one might wonder whether the following argument might also be regarded as epistemically circular:

The moon is made of green cheese.
Therefore, someone has a belief.

This is an argument such that, if someone has a justified belief in the premise, it follows that the conclusion is true; it doesn't matter what the premise is, in fact. But why think this argument is epistemically circular?

Let me try to make it plausible that these arguments are in fact epistemically circular. Suppose I wanted to argue that someone has a belief. I offer some premise(s) and conclude, "Therefore, someone has a belief":[29]

(A)                       p.
Therefore, someone has a belief.

I then endorse (A) as follows: "This shows that I have a justified belief that someone has a belief."

My endorsement of (A) is odd; after all, if I want to say my belief in the conclusion of (A) is justified on account of (A), then I am presupposing that someone (namely, me) has a belief, which presupposition happens to be the conclusion (A). Like track-record arguments that are patently epistemically circular, if (A) is circular, it is not obvious that it is viciously circular; still, some might think, it does seem question-begging in some interesting way, at least insofar as I want to put it to use to show my belief that I have a belief is justified. Still, why should (A) be called circular at all?

Let's say you demand to know how (A) justifies my belief in (A)'s conclusion; I proceed to explain that my belief in the premise(s), p, is justified. You might then object – very sensibly – that I am presupposing the conclusion in defending the claim that I get a justified belief in the conclusion from (A). It is typically just this feature of (A) – that, in defending the claim that I get a justified belief in the conclusion, I presuppose the conclusion – that leads us to say, of other arguments that have this feature, that they are epistemically circular.

For example, consider a track record argument for the reliability of sense perception, of the sort discussed in Section III above. Suppose we wish to explain how such an argument gives anyone a justified belief that sense perception is reliable. In doing so, we defend the argument's premises, but we also presuppose the conclusion. Thus, in explaining how a track record argument gives anyone a justified belief that sense perception is reliable, we presuppose the conclusion. (A) shares that very feature; so we can, with good sense, call it `epistemically circular' as well.

So, on further consideration, perhaps it shouldn't be surprising that both Alston's definition of `epistemic circularity' and mine would have it that (A) is epistemically circular. As, for example, Alston's definition has it, epistemic circularity "involves a commitment to the conclusion as a presupposition of our supposing ourselves to be justified in holding the premises."[30] We may apply this to (A): we are committed to the conclusion, that someone has a belief, as a presupposition of our supposing that we have a belief in the premise(s) at all. It is admittedly a bit strange that the latter is true of (A) regardless of what the premises are – but what difference should that make for our identifying an argument as epistemically circular?[31]

This discussion has the curious and interesting implication that all arguments for the claims "Someone has a belief" and "Someone has a justified belief" – and perhaps others – are epistemically circular. Consequently, if I want to claim that all epistemic circularity is unacceptable, then I must be willing to defend a further claim, namely, that no argument either for "Someone has a belief" or for "Someone has a justified belief" is acceptable. Not everyone will be happy with this result.

It is fortunate for the consistency (not to say the plausibility) of my own views that, in Chapter 4, I sketch a theory on which one might want and expect this result. According to this theory, some claims that cannot be given plausible non-epistemically circular arguments are precisely those that may be taken for granted in argumentative discourse – claims that I will call, in Chapter 2, "philosophical starting-points."

But the considerations of the last two paragraphs suggest an even more serious objection[32] to my definition (and Alston's). Consider any argument the conclusion of which expresses a necessary condition (on my account) or a presupposition (on Alston's account) of any claim to the effect that one of the premises of the argument is justified. Any such argument is epistemically circular on these accounts of epistemic circularity.

For example – arguably, and quite plausibly such propositions as `Something exists' and `There is consciousness' are presuppositions of the justification of belief in any premise (of any argument whatsoever), and their truth is a necessary condition of such justification. Consequently, on both Alston's view of epistemic circularity and mine, any argument with such a conclusion is epistemically circular. If I provide any argument whatsoever for the view that there is consciousness, one may forthwith charge my argument with epistemic circularity. Even more dramatically, if one also holds the view that epistemic circularity is unacceptable, he cannot accept any argument at all for such a view.

This is a very interesting result, and it is possible that we should regard the result as, simply, a philosophical discovery. If I were to bite the bullet at this point, however, and assert that such arguments are indeed epistemically circular, it would make my claim in Chapter 2, that epistemic circularity renders an argument unacceptable, far more controversial than I hope it to be. It might even be the case that, on the commonsense meta-epistemology urged in Chapter 4, we should welcome the view that certain obvious truths, such as those expressed by `Someone has a belief', `There is consciousness', and `Something exists', cannot be given an acceptable argument.[33] But as it is no part of this dissertation's purpose to defend such sweeping claims, I should avoid advancing a view that commits me to them.

Consequently, I propose simply to restrict the extension of `epistemic circularity' to those arguments whose conclusions either are j-standards or make assertions about the relibility of cognitive processes. This might have the effect of excluding some arguments that we might, on reflection, want to regard as epistemically circular. But none of the examples considered in this dissertation would be excluded, nor would many (or perhaps any) of the arguments discussed in the modest literature about this subject.

So here is a revised definition that avoids the above, alleged counterexamples:

Def. An argument A for conclusion c (understood by S), where c is either a justification standard or an assertion that some doxastic practice[34] is reliable, is epistemically circular for S iff (i) if S were justified in believing that c on the basis of the set of S's beliefs about A, then S would be justified in believing each of A's premises and that A's inference is correct (= justifying beliefs about A); (ii) for at least one of the justifying beliefs about A, if S were to have a justified belief in it, then c would be true.

I will address one last issue in this section. Central to this dissertation's argument is the claim that all epistemic circularity is vicious; there is no "benign" epistemic circularity. Without attempting to say what vicious circularity in general is, I can give an account of what it is for epistemic circularity to be vicious:

Def. The epistemic circularity of an argument is vicious iff the circularity renders the argument such that one cannot be justified (in either an externalist – or an internalist sense) in believing the conclusion on the basis of beliefs in the argument's premises and inference step.

This is what I will mean when I say that an argument's epistemic circularity is vicious. Moreover I think it is fairly obvious that it is also what other people mean when they discuss the viciousness of epistemic circularity. Epistemic circularity's particular vice is that it keeps us from getting justified beliefs. So, for example, when Alston says that the epistemic circularity of an argument does not prevent us from using it to gain a justified belief that sense perception is reliable, I will describe his view as that the argument's epistemic circularity is not vicious (and is, rather, virtuous – or benign ).

V. Licensing Standards.

The PMJ as I presented it in Section II consists of little more than a question: For any given standard of justification, how can one show that (belief in) it is itself justified? Formulated this way, there is not one problem but many; there is a different problem for each standard. But my thesis is that there is a problem about justifying standards of justification in general. I shall have to get some more background work on the table before I can present this general problem with adequate precision.

Central to the discussion to come is the view that standards can license belief in an argument – which is to say the standards license the beliefs in the premises of an argument and the belief that the inference from those premises to the conclusion is correct. So I will speak of licensing standards of premises, of inferences, and (by extension) of entire arguments.

I will say that meta-justificatory arguments have licensing standards. Examples of what I mean can be seen in the track record arguments discussed in Section III; the bridge principles, of the form "If (na), then (nb)," were presented as licensing, or explaining the justification of, each (nb). It will be helpful to clarify this talk of licensing standards further.

Suppose that I accept an argument, A, and on its basis have a justified belief that p. Then I must have been justified in believing the premises of A, and in making the inference from those premises to p. Now the core intuition behind this talk of a licensing standard is the notion that an argument's premises and inferential move are, insofar as they are justified for me at all, justified in accordance with some j-standard(s).

So, for example, if A is of the form modus ponens , then "there is" some j-standard that says, roughly,

If someone understands how the conclusion follows from the premises, and the argument is in the form modus ponens, then that person is justified in believing that the conclusion follows from the premises.

I want to say that something like this rule "licenses" my belief in argument A, by licensing my belief in A's inference. I shall not attempt to explain what understanding how conclusions follow from premises amounts to; that's obviously a matter for another day.

One difficulty comes in saying exactly what standard does license belief in the premises, or in the inference, of a given argument. When arguing, only very rarely do I advance any standards according to which I think I am justified in believing the premises and inference. Even if I did, I might be wrong about it. And even if we can agree that, in an argument of the form modus ponens something like the above-stated rule licenses belief that results from the inference, it is still a matter of considerable deliberation and potential disagreement exactly how the rule ought to be stated. It appears to follow that for any given argument, several competing standards might explain how belief in its conclusion is justified (or not); and so speaking of the licensing standard appears tenuous at best.

But this difficulty of arriving at an adequate account of licensing does not entail that one cannot explain the sense of `licensing standard' in general. I propose the following as a plausible account:

Where p is the conclusion of argument A and S's belief that p is based upon beliefs in A's premises and inference step, a justification standard, J, licenses S's belief that p iff J is the best explanation  for the fact either that (1) S justifiably believes one or more of the premises of A, or of the fact that (2) S justifiably believes the inference from the premises of A to its conclusion.

This account of licensing admittedly does not give any indication of what standards do "best explain" why S is justified in believing the premises or in making the inference. That is a topic for another, no doubt very lengthy discussion. Still, I would be remiss if I did not give a few words now about what I mean in saying that a licensing standard best explains why someone is justified in a belief.

Here I may advert to some platitudes about the methodology of epistemology (at least contemporary epistemology ). The central challenge of a theory of justification is to say what it is in virtue of which various of our beliefs are justified. In examining candidate accounts, the method most commonly used is to discuss whether an account "rules" correctly (correctness being decided by "intuitions") on a number of different (often highly contrived and difficult) cases. This is essentially abductive reasoning, or inference to the best explanation: the best account of justification will be the one that explains the highest proportion of cases, including tough cases.

I shall be making another assumption that will be discussed later, namely, that whenever one generates a justified belief through the use of an argument, some standards do license the argument. But at least one view – a version of particularism – would have it that some meta-justificatory arguments have no licensing standards at all. I will evaluate and reject this view in Chapter 3.

VI. How Justification Standards are Interrelated.

Just as the regress argument generates talk of "basic beliefs," the Meta-Regress Argument to be introduced anon will generate talk of "basic j-standards." It will be useful to explain the notion of basicality in advance. But basic j-standards are only some of the possible elements at which a meta-regress ends, if it does end. So it will be equally useful to define in advance several other relations that standards can have to each other: for example, standards might be, at least hypothetically, mutually supporting.

Throughout this section I will speak of support that one standard may have for another. On this, three points of clarification are in order. First, I mean this not in a "success" sense, but a merely putative sense. So you may substitute "alleged support" wherever I have used "support." Second, when applied to j-standards, `support' is to be understood in a special sense, as according to the following recursive definition. Clause (i) is the base clause; (ii) is the recursive clause; (iii) is the closure clause:[35]

For any justification standards, J1 supports J2 in a doxastic system D iff

(i) J2 is the conclusion of an argument in D that includes J1 among its licensing standards, or

(ii) J1 supports Jn (in the sense explained in (i)) and Jn supports Jm, etc., which supports J2; and

(iii) no other j-standards support any other j-standards in D.

Third, strictly speaking, this supports relation exists between beliefs in standards; i.e., it is merely shorthand to say that one standard supports another. Hence supports relations between (beliefs in) standards will be said to exist in a given doxastic system. There are, granted, rather few j-standards in anyone's doxastic system, and so there are rather few of the above-defined supports relations. Thus we might, alternatively, speak of supports relations between standards apart from any doxastic system – in much the same way that one might speak of propositional justification rather than doxastic justification. But I am supposing we will want to formulate the Meta-Regress Argument as an argument about the (doxastic) justification of beliefs in j-standards, which is the natural way to understand the subject of the PMJ, after all. So it will be best to speak of supports relations as holding between beliefs in j-standards that exist (where else?) in a doxastic system.

For any given standard, as I will now explain, there are three possible supports relations that it can bear to other standards.

In the ordinary regress argument, the possibility is raised that some beliefs are justified but not justified by any other beliefs. Such beliefs are often said to be `self-evident' or even `self-justifying'. But these beliefs need not be, and typically are not, held to be beliefs that (whatever this might mean) bear the (ordinary) supports relation to themselves.[36] The latter sort of possibility is rarely considered seriously at all.

In the Meta-Regress Argument, however, a closely analogous possibility must be considered seriously. A surprising number of people have held that there is nothing wrong with an argument for a standard that is licensed by that same standard. In such a situation one may properly identify a "self-supporting" standard:

A justification standard J is self-supporting in a doxastic system D iff J supports J in D.

This means that a self-supporting j-standard will be one that licenses an argument with itself as the conclusion. The track record arguments from Section III contain a couple of examples. A plausible historical example might be Descartes' criterion of truth; as the famous "Cartesian circle" would have it, a long chain of argumentation is supposed to support the claim that whatever I clearly and distinctly perceive is true, but arguments for crucial elements of this chain (viz., that God exists and is not a deceiver) are licensed by this same criterion.[37]

Another possible relation between standards is mutual support :

A justification standard J1 is mutually supported , and J1 and justification standard J2 are mutually supporting in a doxastic system D iff J1 supports J2 in D and J2 supports J1 in D.

Suppose A1 is used to argue for standard J1, and belief in one of A1's premises is licensed by standard J2. Suppose further that in the doxastic system under consideration, J2 is the conclusion of another argument, A2, itself licensed by J1. I want to say that in such a situation J1 and J2 are mutually supporting and, individually, each is on that account mutually supported .

There might be, however, a whole group of standards with complex support relations in this way. It might be the case that J1 supports J2 only by licensing the first in a long chain of arguments, each licensed by a different standard, and the last argument in this chain has as its conclusion J2. And so long as J2 also licenses an argument for J1, they are mutually supporting and both, individually, mutually supported.

There is one other possibility: the notion that there is an infinite regress of standards supporting a given standard. One may speak of a standard that is "infinitely" supported, so long as there is an endless chain of standards supporting it:

A justification standard J1 is infinitely supported in a doxastic system D iff J1 is supported by J2, and J2 is supported by J3, and so on ad infinitum, in D.

Self-support, mutual support, and infinite support – are all properties that standards can have in virtue of their support relations with other standards. But there are other properties that standards can have in virtue of their lack of such support. Such a property is basicality. Again, the notion of basicality must not be confused, in this context, with the notion of self-support. As I will use the word,

A justification standard J is basic in a doxastic system D iff the belief that J is justified in D, but the belief that J is not[38] supported (in the ordinary sense) by any argument (or by any justifiers – that bear the ordinary supports relation to J) in D.

So a standard is basic only if a belief in the standard is both justified and not the conclusion of any argument (or, per the remarks in Section II above, by anything that may be converted into an argument). Basic standards exist only if it is possible to meet both conditions at once.

Finally, it is quite possible that a standard is simply unsupported in a doxastic system. The following definition names and describes that situation:

A justification standard J is a posit in a doxastic system D iff it is not the case that J is justified in D, and J is not supported by any argument in D.

It is a logical possibility, of course, that the same standard be self-supporting, mutually supported, and infinitely supported. But posits are necessarily not basic and vice-versa, and neither can be self-supporting, mutually supported, or infinitely supported.

With this background in place, we are ready to come to grips with the Meta-Regress Argument.

VII. The Meta-Regress Argument.

In the ordinary infinite regress argument, well known to epistemologists, a given belief is said to be justified. Then it is asked what justifies that belief, and then what justifies that, and so on until we have run to the end (or the beginning) of the justifying beliefs. This series of beliefs is the "regress."

The meta-regress I wish to introduce consists of a series, not of ordinary beliefs, but of (beliefs in) licensing standards. It is "meta-" in that it concerns beliefs about which beliefs are justified. It has an ancient predecessor in Sextus Empiricus' diallelus .[39]

The ordinary regress argument – can be used to support various contrary conclusions; it can, for example, be used to argue for either foundationalism – or coherentism. I intend to use the Meta-Regress Argument (MRA ) to support the view that some standards of justification cannot themselves be shown to be justified; i.e., one cannot justify belief in certain important standards. And so, in Chapters 2 and 3, I will be filling out Chapter 1's sketch of the MRA to show that the Problem of Meta-Justification – cannot be solved by any frontal attack. Having done that, though, I will argue in Chapter 4 that that this position is not necessarily a skeptical position. So on to the argument itself.

Suppose S accepts a certain standard of justification, J1. We raise the question central to posing the PMJ: what justifies S in accepting J1? Suppose S claims to have an argument and says that his beliefs in each of the premises and in the connection of the premises with the conclusion are what justify his belief that J1. In other words,

(1) S's beliefs about an argument (perhaps a track record argument ), A1, justifies S in accepting J1.

This is the answer one should expect, given what I wrote in Section II about what will constitute a solution to the PMJ.

Assume further that

(2) A1 has at least two[40] licensing standards; call one of them J2.

In that case, there are two basic possibilities:

(3a) J2 is identical to J1.

(3b) J2 is not identical to J1.

If (3a), then J1 is self-supporting. In this case we may say S is committed to self-support meta-foundationalism (or self-supportism for short), which is the view that some self-supporting standards of justification exist.

Suppose instead that (3b): J2 is not identical to J1. Then we ask what justifies S in believing J2. Here there are a number of possibilities. First, suppose:

(4a) S has an argument A2 for J2, which is licensed by J1.

Or (the analysis will be the same) suppose:

(4b) S has a series of arguments supporting J2, and at some point in this series, an argument is licensed by J1.

In either case, J1 is mutually supported (because J1 and J2 are mutually supporting ). S is committed to meta-coherentism , the view that some mutually supported j-standards exist.

Another option is:

(4c) S claims to accept (dispositionally) an infinite series of standards, or that "there is" an infinite series of them, supporting but not containing J1.

Then S believes that J1 is infinitely supported. And then S apparently accepts meta-regressism , the view that infinitely supported j-standards exist.

Each of these possibilities is unacceptable.

The trouble with self-supportism is that arguments containing self-supporting standards are, necessarily, epistemically circular. This can be argued for briefly as follows. Since J1 is self-supporting, J1 licenses belief either in one of the premises or that the inference holds; but if the premises and inference are justifiedly believed and J1 succeeded in licensing them, then J1 is true. But according to the definition defended earlier in this chapter, to say that A1 is epistemically circular is to claim: if S were to have a justified belief in one of the premises of A1, or that its inference holds, then (it follows that) J1 would be true. Hence, if J1 is self-supporting, then A1 is epistemically circular.

The crucial premise here is that, in order to be able to succeed in licensing an argument – to explain, successfully, that in virtue of which the argument gives someone a justified belief in its conclusion – a standard must be true. Admittedly, a false j-standard could be the best explanation available for all we know; but if it really is false, then from an objective point of view it fails to account for the justification of anything. Or else calling the standard "false" is meaningless: I think the only good meaning to give to the claim that a standard is false is that it does fail to account for the justification of that for which it purports to account.

As I will be concerned to argue in Chapter 2, epistemic circularity is vicious; hence self-supportism, committed as it is to justifying standards through the use of epistemically circular arguments, must be rejected.

In Chapter 3, I will be concerned to discuss the merits of meta-coherentism (in large part through examining the prospects of the method of reflective equilibrium as a solution to the PMJ ). It too, however, rather obviously falls prey to epistemic circularity. But defenders of ordinary coherentism – are concerned to argue that a "large enough" circle can mitigate the viciousness of circularity. I will examine (and reject) a similar claim in the context of meta-coherentism and epistemic circularity. So meta-coherentism must be rejected as a solution to the PMJ. In the same chapter, meta-regressism (none too plausible to begin with) will be dispatched.

So now suppose, after we reject (3a), (3b), (4a), (4b), and (4c), S returns to the original claims that were needed to generate these possibilities: (1) and (2). If all of these possibilities are incorrect, it must be because either (1) or (2) was incorrect.[41] But is it reasonable to reject either of these claims?

Suppose S accepts (1) but rejects (2), and so argues:

(5) There is not at least one licensing standard in virtue of which belief in the premises (of A1) or in the inference is justified.

In other words, there are some (successful) meta-justificatory arguments that lack any licensing standards; there are some standards that lack any support by any other standards. Can this claim be made out plausibly? It would seem so, since this is essentially the approach that so-called particularism (Chisholm's word[42]) takes to the PMJ. While particularism was formulated as a solution to Chisholm's so-called Problem of the Criterion (which differs somewhat from the ancient problem), it can be readily construed as a solution to the PMJ.

The particularist holds that we must, in order to defend any criterion of justification successfully, begin with particular instances of justified belief, to be determined without reference to criteria. It so happens, however, that getting very clear on the structure and presuppositions of the MRA, as I have tried to do in this chapter, also makes it clear that the particularist is committed to some very implausible claims. As I will explain in Chapter 3, I think it is especially implausible that there might be a belief that is justified, but not justified in accordance with any standard of justification at all. I will also argue in Chapter 3 that particularism, construed as a solution to the MRA problem (as it is plausible to do), is committed to this view. I will explore the issues involved in construing particularism in this way, and will ultimately reject it.

So let us suppose that S backs up to the beginning of the MRA and rejects the premise that undergirds all of the other options: (1). So it is nothing like an argument that justifies S in accepting J1. S claims, instead:

(6) S is justified in accepting J1 without argument and without making any reference to justifiers – that can be interpreted, or construed, as an argument.

So S is committed to the view that there are basic j-standards. We might well call S's view basic meta-foundationalism . Surely, S's view is at least as deserving of the name `foundationalism' as self-support meta-foundationalism, since (unlike the latter) it posits genuinely basic beliefs, i.e., beliefs that are justified but not by other beliefs.

Nonetheless I will prefer a briefer and more familiar name, again due to Chisholm: methodism . The methodist offers a solution to the Problem of the Criterion – that can also be construed as a solution to the PMJ, as follows. We begin our work in epistemology by (somehow) fixing upon a set of j-standards, and subsequently applying those standards to particular beliefs to argue that they are, or are not, justified. We can also use those standards to argue for derivative, more specialized standards.

In Chapter 3, I will introduce a fundamental objection to methodism – that I will briefly outline now. If some belief may be called `justified', then there is something (called its `justifier(s)') in virtue of which it is justified. This is a fact about the meaning of `justified'. Even if, as (6) has it, the justifiers – are not beliefs, there are some sort of justifiers, i.e., facts in virtue of which the belief is justified. Moreover, if the belief is justified by those justifiers, the supports relation must hold between these justifiers and the belief. And so S, in claiming that he is justified in accepting J1, is committed to the existence of some items the descriptions of which must bear the same (supports) relation to the belief that J1 that the premises of an argument would have to bear to J1. In that case, (6) is false. In short, in Chapter 3, I will elaborate some points from Section II above as a refutation of methodism.

It would appear that we are left with the view that nothing (ultimately) justifies S in accepting any standard of justification. Put differently, all j-standards are, if supported at all, ultimately supported only by mere posits.

It might be thought, then, that we are left with what may be called meta-skepticism . But an option remains, that, if correct, promises to keep us from having to swallow this bitter pill. Suppose that, according to some established sense of `rational', it is possible for belief to be rational without supporting reasons. I claim that there are some standards that S can believe rationally but not, in any ordinary sense, justifiably (or unjustifiably, for that matter), and for which he cannot offer any even slightly plausible argument (or describe supporting circumstances that can be restated in argument form).

Some standards, I will maintain, describe rational belief-formation. To accept such standards is perforce to accept that certain kinds of beliefs are rational. Given such standards, noncircular arguments for j-standards are in the offing. In Chapter 4, I will elaborate and defend this position, which is deeply indebted to and informed by the meta-epistemologies of Strawson, Wittgenstein, and most of all the great Scots philosopher Thomas Reid.

It should be enormously obvious to epistemologists – that the MRA covers ground that has already been covered both recently and in the very distant past in various ways. Some of the more obvious historical figures that come to mind are Sextus Empiricus, Hume, Reid, and Wittgenstein. Recently this ground has been trodden, most prominently, by Alston, but also by Chisholm, Sosa,[43] and many others. I believe the present approach to organizing the problem and related arguments has the advantage of being particularly clear and straightforward.

This approach also has a further advantage, specifically over Alston's approach in The Reliability of Sense Perception. Alston devotes about one hundred pages of that book to showing that various attempts to solve (something like) the PMJ – cannot escape epistemic circularity. This work, while doubtless useful and interesting, may be bypassed if the MRA is successful. For whereas Alston might have shown piecemeal that many particular theories each cannot escape epistemic circularity, the MRA aims in part to show why those theories in general, and any of their undiscussed competitors, could not escape it. Indeed, I point to the MRA as the best explanation of why the arguments of those one hundred pages of Alston's were suffused with a strong odor of inevitability.

Next Section