Epistemic Attitudes1 Scott Sturgeon

Certain propositional attitudes are capable of being reasonable or well taken in light of the evidence. These are the epistemic attitudes. It is not a trivial hypothesis that we manifest any of them in everyday life. It is a trivial hypothesis that by doing so we open aspects of that life to epistemic evaluation. Obviously, common-sense insists that we manifest epistemic attitudes with great frequency. But it also conceptualizes them in at least two different ways. On the one hand, common-sense recognizes a coarse-grained space of epistemic attitudes consisting in belief, disbelief and suspended judgment. On the other hand, it recognizes a fine-grained space of them consisting in countless levels of confidence. This raises a number of interesting questions. Two will be our main focus:

(A)

How do elements within one of these attitudinal spaces relate to one another?

(B)

How do elements across such spaces do so?

Question (A) concerns how the metaphysics of belief, disbelief and suspended judgement relate to one another, and how the metaphysics of levels of confidence do too. These intralevel topics apply within a given space of epistemic attitudes. Question (B) concerns how the metaphysics of coarse- and fine-grained epistemic attitudes relate to one another. This interlevel topic applies across attitudinal spaces. To a rough first approximation, there are reductionists and anti-reductionists about each of these issues. Intra-level reductionists about coarse-grained attitudes, for instance, say typically that one of belief, disbelief or suspended judgement is more basic than its coarse-grained cousins. And they use their take on the putatively more basic attitude to build an approach to the nature and norms of coarse-grained attitudes said to be less basic. Similarly, intra-level reductionists about fine-grained attitudes say typically that high or low levels of confidence are more basic than their opposite number. And they use their take on the putatively more basic attitudes to build an approach to the nature and norms of attitudes said to be less basic. Intra-level anti-reductionists reject this picture. They maintain that coarse- and finegrained attitudes are on an explanatory par with their level-mates, and, for this reason, that attitudes within each level are subject to proprietary epistemic norms. A familiar dialectic then plays out. Reductionists plump for a line blessed with ontic and explanatory unity; but they pay for it with substantive hostage to fortune. Anti-reductionists avoid such hostage to fortune; but they do so by admitting a wider expanse of self-standing mental and evaluative fact.

1

Acknowledgement.

1

The dialectic carries over, mutatis mutandis, to the inter-level issue at (B). Some maintain that coarse-grained attitudes reduce to their fine-grained cousins, while others maintain that fine-grained attitudes reduce to their course-grained cousins. These are interlevel reductionists about epistemic attitudes. A given level of attitude is said to be more basic than a distinct level, with the putatively more basic level being used to build an approach to the nature and norms for attitudes said to be less basic. Once again anti-reductionists reject the picture. They maintain that coarse- and finegrained attitudes are on an explanatory par with one another, and, for this reason, that attitudes at each level are subject to proprietary epistemic norms. On their view: belief, disbelief and suspended judgement do not reduce to levels of confidence, and levels of confidence do not reduce to belief, disbelief or suspended judgement. Once again reductionists put forward a picture blessed with ontic and explanatory unity, paid for it with non-trivial hostage to fortune. Once again anti-reductionists avoid such hostage to fortune by admitting a wider expanse of self-standing mental and evaluative fact. This paper argues for an anti-reductive stance on the intra-level issues but a reductive one on the inter-level issue. It will be shown that members of each class of epistemic attitude are on an explanatory par with their classmates, but, that one of those classes is more basic than the other. Moreover, the inter-level reductive view defended will itself manifest virtues had by each type of inter-level reduction: those which ground coarse-grained attitudes in their fine-grained cousins, and those which do the reverse. Attractive aspects of each strategy will be slotted into a unified whole. This will be done by locating explanatory resources more basic than any kind of epistemic attitude. Those resources will be used to build a fully reductive approach to epistemic attitudes. The paper unfolds in eight parts. §1 argues against a belief-first approach to coarsegrained attitudes. §2 develops a conception of such attitudes from an intuitive and functionalist perspective. §3 argues that point-valued subjective probability does not reduce to the belief-theoretic. §4 explains the most natural attempt to reduce coarse-grained attitudes to such probability, and why the attempt fails. §5 lays out a new approach to the nature of confidence by appeal to an overlooked resource, one which is more fundamental than any epistemic attitude. §6 uses that approach to create a new kind of confidence-first take on coarse-grained attitudes, and sketches its good-making properties. §7 shows how the new approach to confidence prompts a natural diagnosis of why the most popular formal kinematics in the area systematically breaks down. §8 develops a new formalism for confidence—prompted by the philosophical conception developed earlier--and uses both to clarify, for the first time, propositional and doxastic justification within a confidencetheoretic setting.

1.

Only One Coarse Attitude?

There is a curious asymmetry in the epistemology of coarse- and fine-grained attitudes. A common thought in the former is that there is only one explanatorily basic attitude: belief. Suspended judgment and disbelief are routinely reduced to belief.2 Nothing like that happens in the epistemology of confidence.

2

For instance, in the AGM or belief-revision literature: see Gärdenfors (1988) or Hansson (1999). Noted dissenters with respect to disbelief are Field (2003) and (2008), and Rumfitt (2000).

2

The relevant perspective on coarse-grained attitudes flows from a pair of reductive schemata: (a)

{SJ()} is nothing but {¬B() & ¬DB()},

(b)

{DB()} is nothing but {B(¬)}.

and

The idea here is that suspended judgement is nothing but the absence of belief and disbelief, and disbelief is itself nothing but belief in negation. The joint validity of these schemata entails that there is only one basic coarse-grained attitude: belief. No one plumps for an analogue view in the theory of confidence. Within a probabilistic setting this would amount to the idea that for any unit real n less than .5:

(c) {Cr() = n} is nothing but {Cr(¬) = (1-n)}.

The idea would be that low credence is nothing but high credence in negation. The validity of this schema entails that low credence reduces to high credence in negation. That is obviously wrong. Low credence is not simply high credence in negation, but a range of proprietary kinds of attitudinal stance. Being one-quarter certain that it will snow, for instance, is not being three-quarters certain that it will not snow. The latter take on things may be rationally forced when the former is taken—we set that aside for the moment—but the former stance is something distinct from the latter, psychologically speaking. And so it goes with other states of low credence: they are not a state of high credence deep down. The implausibility of schema (c) prompts doubt about its cousins (a) and (b). After a bit of reflection, in fact, it does not seem plausible that suspended judgment is nothing but the absence of belief and disbelief; and nor does it seem plausible that disbelief is nothing but belief in negation. Just recall the schema for suspended judgement:

(a)

{SJ()} is nothing but {¬B() & ¬DB()}.

The tail end of (a) is insufficient for suspended judgement. When you fail to consider whether Caesar crossed the Rubicon, for instance, you fail to believe or disbelieve that he did so. But that doesn’t mean that you have suspended judgement in Caesar crossing the Rubicon. You have no view of the matter (by stipulation). Suspended judgement does not reduce to the absence of anything; a fortiori it does not reduce to the absence of belief and disbelief. More plausibly, suspended judgment concerns the presence of some kind of commitment, something itself more than an absence or lack. Suspended judgment concerns a 3

particular kind of epistemic stance, a proprietary sort of committed neutrality. It will take work to flesh out what this comes to, a central sub-goal of this paper. Recall next the schema for disbelief:

(b)

{DB()} is nothing but {B(¬)}.

The idea here is that disbelief is merely belief in negation. If that’s right, however, disbelief is itself something derivative which springs from more basic attitudinal fact. That too seems false. The tail end of (b) concerns the endorsement or acceptance of ¬. Disbelief in  does not concern the endorsement or acceptance of anything. Rather, it concerns the pushing-away of  in thought. Such a stance does not seem to involve acceptance at all, but rather a proprietary type of attitudinal condemnation. This too will take some explaining, another central sub-goal of this work. But our initial impression is clear: belief, disbelief and suspended judgment are each their own kind of thing. None of them initially seems to reduce to the others. We need a conception of coarse-grained attitudes which underwrites this initial impression. Developing one is our next task.

2.

Coarse-grained attitudes.

We need a conception of belief, disbelief and suspended judgement on which each has a psychological life of its own. To unearth one I shall lay out a pair of thought experiments. Then I'll sketch a take on coarse-grained attitudes suggested by them, and reflect on the result from a functionalist perspective. The initial thought experiment is

Three Doors You are told to exit a room by one of three doors: the Left door, the Middle door or the Right door. Hence there are four things that you could do: exit by L, exit by M, exit by R, fail to act. This means we can define a condition

C =df. (¬L & ¬R)

and note that there are two ways to get into C: exit by M, fail to act. Intuitively, a similar structure is in play with coarse-grained attitudes. Suppose you are given some data relevant to . You are instructed to examine them and to adopt the take on  which best reflects their relevance concerning whether or not . There are three 4

coarse-grained options before you: belief, disbelief and suspended judgement. Hence there are four psychological possibilities to hand: believe , disbelieve , suspend judgment in , fail to adopt a view. So there are two ways to fail to believe  while also failing to disbelieve . You might suspend judgement in . You might end-up with no commitment. Our second thought experiment is

The Judge You are a pre-trial judge. Your job is to assess evidence gathered by police to decide whether a trial is warranted. Three verdicts are available to you: evidence warrants trial, evidence warrants no trial, evidence does neither of these things. Mr. Big has been charged with stealing and you have been contacted by one of his powerful allies. You are urged to reach a verdict immediately to dampen public discussion of the case. But you have not yet received an evidential dossier from the police. Hence there are four options to hand: commit to evidence warranting trial, commit to evidence warranting no trial, commit to evidence doing neither, refrain from commitment. You should opt for the last of these choices, of course, refraining from commitment about whether evidence warrants a trial. After all, you have no clue about the evidence in the case. The same structure plays out in the epistemology of coarse-grained attitudes. There are four attitudinal stances available. Intuitively put:

settled endorsement = belief settled denial = disbelief settled neutrality = suspended judgment lack of a settled stance = no epistemic attitude.

There is a clear psychological difference between these four scenarios. The first involves a settled intellectual embrace, the second a settled intellectual rejection, the third a settled neutrality, and the fourth an absence of any kind of settled take. Within the epistemology of coarse-grained attitudes, then, we have four attitudinal possibilities: belief, disbelief, suspended judgment and unsettledness. None of these states inter-reduce. Disbelief does not boil down to any combination of belief, suspended judgement or lack of commitment, for instance; and nor do any of the other coarse-grained attitudes boil down to a combination of their coarse brethren. Moreover, a lack of attitudinal commitment obviously does not boil down to a combination of such present commitments. Thus it is that none of the four states just mentioned inter-reduce. We can make ready sense of all this from a functionalist perspective. Here is one way to do so. First, we align belief with the strong and stable disposition to act as if a particular content is true. This means we align belief in , for instance, with the strong and stable disposition to take bets on , to say things like “Yes, of course!” when asked if  is true, to 5

rely in reasoning on a pro- stance when appropriate--both practically and theoretically— and so on. This amounts to the alignment of belief in  with the constellation of its “signature functions”, with the collection of its functionally constitutive elements.3 Second, we align disbelief with the strong and stable disposition to act as if its content is false. This means we align disbelief in  with the strong and stable disposition to reject bets on , to say things like “No, don’t be stupid!” when asked if  is true, to rely in reasoning on an anti- stance as appropriate–both practically and theoretically—and so on. This amounts to the alignment of disbelief in  with the constellation of its signature functions, with the collection of its functionally constitutive elements.4 Third, we align suspended judgment with the strong and stable disposition to refrain from activities the strong and stable presence of which make for belief or disbelief. This means we align suspended judgement in  with the strong and stable disposition to refrain from accepting or rejecting bets on , to say things like “Oh, that’s unclear” when queried about , to refrain from relying on a pro- or anti- stance in reasoning–both practically and theoretically—and so on. This amounts to the alignment of suspended judgment in  with the constellation of its signature functions, with the collection of its functionally constitutive elements. And finally, we align being attitudinally unsettled with the absence of dispositions the strong or stable presence of which make for belief, disbelief or suspended judgement. This means being unsettled about  involves a lack of strong or stable dispositions to act as if  is true, a lack of such dispositions to act as if  is false, and a lack of such dispositions to refrain from those dispositions the strong or stable presence of which make for belief or disbelief. Hence attitudinal unsettledness about , intuitively, involves functional instability about whether  is true. Such instability is the signature function of attitudinal lack, and, as such, it is constitutive of such a lack, i.e. constitutive of attitudinal unsettleness. In this way we sketch the beginnings of a functional take on our coarse-grained attitudes. That sketch is perfectly consistent with the idea that such attitudes fail to interreduce. Hence we should affirm our initial impression: belief is a proprietary kind of intellectual embrace, disbelief a proprietary kind of intellectual pushing-away, and suspended judgment a proprietary kind of committed neutrality. When it comes to the relation between our coarse-grained attitudes, the most we can hope for is normative alignment under idealisation. For instance, we might embrace the following norm for ideally rational agents:

(a)*

SJ() iff ¬B() & ¬DB().

In the event, we maintain that such agents suspend judgment exactly when they fail to believe or disbelieve. As it happens we have no need to pronounce on the bona fides of (a)*. We 3

Which functions are constitutive of belief? This is a good question on which we'll take no stand. Some kind of functionalism about belief will be assumed here, but nothing need be said about whether it is of the commonsense variety, the scientific variety, or something else yet again. Similar points apply to disbelief and suspended judgment. For elegant discussions of functionalism see Loar (1981), Lewis (1966) and (1994), and Schiffer (1987). 4 Ian Rumfitt uses an approach like this to disbelief in deductive reasoning. See Rumfitt (2000).

6

need only conclude that suspended judgment is a proprietary kind of committed neutrality, one which fails to reduce to other coarse-grained attitudes. Similarly, we might embrace the following norm for ideally rational agents:

(b)*

DB() iff B(¬).

In the event, we maintain that such agents disbelieve exactly when they believe in negation. But we also have no need to pronounce on the bona fides of (b)*. We need only conclude that disbelief is a proprietary kind of attitudinal stance, one which does not reduce to other coarse-grained attitudes. We are left, then, with a pair of interim conclusions: coarse-grained attitudes stand on their own with respect to one another, failing to inter-reduce; and coarse-grained attitudes are subject to proprietary epistemic norms. This means there is symmetry between coarse- and fine-grained attitudes: when it comes to how elements within a given attitudinal space relate to one another, anti-reductionism wins the day. Coarse- and fine-grained attitudes are on an explanatory par with their level-mates, and, for this reason, attitudes within each level are subject to proprietary epistemic norms.

3.

Reducing Credence to Belief.

In addition to belief, disbelief and suspended judgement, of course, we have fine-grained attitudes to deal with. How do states of confidence relate to their coarse-grained cousins? Can the former be reduced to the latter? Can the latter be reduced to the former? In this section we look at the first approach. In the next section we look at the second. Attempts to reduce credence to belief take the following form:

{Cr() = n} is nothing but {...B... ...n}.

The root idea is that n-level credence in  is some kind of belief-theoretic situation involving  and n. There are two natural ways to flesh out the idea. One involves locating bits of content believed to match credal strength. The other involves locating bits of the use of belief to match that strength. Consider each in turn:

[1]

The first strategy makes appeal to this schema:

(*)

{Cr() = n} is nothing but {belief in [prob() = n]}.

7

The idea here is that n-level credence in  is nothing but belief in the probability of  being n. I call this view “credence-as-belief”.5 It faces an obvious initial worry. We do not have explicit beliefs about probability for every claim to which we lend credence. Just consider the dead-obvious claim vocalized with the following sentence: "Either snow is white, or it is not the case that snow is white, or I'm confused". I invest maximal credence in the claim I make with this sentence. But I do not have an explicit belief about the probability of that claim. This is but an instance of more general phenomena. There are not enough explicit beliefs about probability to cover claims to which we lend credence. This means that credence cannot be reduced to explicit belief in probability. Defenders of credence-as-belief should maintain that credence reduces to normallytacit belief in probability. They should construe credal-making belief as routinely below the surface of our mental life, appearing on that surface, if at all, only in special informational settings, perhaps only when other probabilistic considerations are made explicit. Let us grant the point for argument’s sake and work through a particular case. Let us suppose that an agent is 70% sure that it will rain (R), and that the signature function of that credence is being 70% disposed to say that it will rain. The former supposition helps make for an illustrative case, and the latter helps vivify the functional signature targeted for reduction by credenceas-belief in such a case. The view then entails

(i)

{70% credence in R} is nothing but {belief in [prob(R) = 70%]}.

A defender of credence-as-belief cannot rest with this claim. If she is to place a serious proposal before us, she must tell us more about the credal-making belief in (i). Only after she does that—and in some detail, I should insist—will we properly understand what 70% credence in R is said to be on her view. So how should a defender of credence-as-belief make sense of the credal-making belief in (i)? Well, let us suppose—again for illustrative purposes—that she does so within a computational approach to the mind. More specifically, let us suppose that our defender of credence-as-belief says that thought takes place within an inner lingua mentis; and she says that belief is itself a computational relation taken to interpreted sentences within that mental language. The result is a view on which belief that it will rain is a computational state involving a mental sentence which means that it will rain, and, similarly, belief that the probability of rain is 70% is a computational state involving a mental sentence which means that the probability of rain is 70%. Our computational defender of credence-as-belief will say in general that belief in  is a computational state involving a mental sentence which means .

5

One might insist that proponents of credence-as-belief specify the kind of probability at work in their view. The idea would be that the credentials of credence-as-belief turn on the plausibility of credence reducing to belief in some theoretically-articulated kind of probability: nomic, metaphysical, evidential or whatnot. But that would be a mistake. After all, proponents of credence-as-belief can say that credence is nothing but belief in an everyday notion of probability, a notion no more detailed or theoretically articulated than any other ordinary notion. They can say that their notion of probability is pinned down by its conceptual and/or functional role in our life. And they need only insist, after making this claim, that belief in claims built from their notion of probability registers functionally as credal states do.

8

Let her computational relation be B, and  be a sentence in the language of thought. The picture is then one on which

(ii)

{belief in [prob(R) = 70%]} is nothing but {B() when  means [prob(R) = 70%]}.

This joins with (i) to entail

(iii)

{70% credence in R} is nothing but {B() when  means [prob(R) = 70%]}.

I maintain that any proponent of (iii) faces a nasty dilemma. After all, any proponent of (iii) must tell us something about its explanatory resources. Failure to do so leaves us ill-equipped to understand what is said to reduce 70% credence in rain. In spelling out the explanatory resources in (iii), however, one must either appeal to the signature function of being 70% sure that it will rain, or fail to do so. In the first case a defender of (iii)'s spiel will vitiate the reductive ambitions of credence-as-belief. In the second case there will be little reason to suppose that (iii) is true, as no reason will have been given to think that standing in B to a mental sentence which means that the probability of R is 70% itself guarantees that one functions as those who are 70% sure that it will rain constitutively function. Consider each horn of this dilemma.

Horn 1 Suppose a computationalist defender of credence-as-belief spells out the explanatory resources in (iii) by explicit appeal to the signature function of 70% credence that it will rain. The simplest way for her to do so is to identify the relevant explanatory resources with the relevant signature function. The view would then be

(iv)

{B() when  means [prob(R) = 70%]} = {a 70% disposition to say that it will rain}.

There are a number of problems with this thought. For one thing--and most obviously—(iv)’s right-hand side does not specify a computational relation of any kind, much less one taken to a mental sentence which means that the probability of rain is 70%. Hence (iv) looks straightforwardly false from the start. More importantly for our purposes, however, is the fact that (iv) is incompatible with the reductive ambitions of credence-as-belief. To see this note how it joins with (iii) to entail the relevant computationalist's full-dress take on 70% credence in rain:

9

(vi)

{70% credence in R} is nothing but {a 70% disposition to say that it will rain}.

Relative to the illustrative assumptions we are making, of course, (vi) is perfectly true; but that is only because (vi) states what any functionalist would maintain relative to those assumptions, namely, that a 70% credence in rain is individuated by its signature function, by a 70% disposition to say that it will rain. Once the explanatory resources in (iv) are equated with the signature function of 70% credence in rain, the resulting view is not one on which credence reduces to belief in probability. It is rather one on which belief in probability reduces to credence. The result badly conflicts with the reductive ambitions of credence-asbelief. Suppose, then, that a computationalist defender of credence-as-belief forges a more subtle connection between the explanatory resources in (iv) and the signature function of 70% credence in rain. Suppose she does not simply equate the former with the latter, but, instead, uses the signature function of 70% credence in rain to explain how a mental sentence gets its meaning. Her line is then something like this:

Lending 70% credence to rain is nothing but believing that the probability of rain is 70%. Believing that the probability of rain is 70% is nothing but bearing computational relation B to a mental sentence  which means that the probability of rain is 70%. And a mental sentence  means that the probability of rain is 70% in virtue of those standing in B to it being thereby 70% disposed to say that it will rain.

This story does not equate the explanatory resources of (iv) with the signature function of 70% credence in rain. It makes a more subtle move. It deploys the signature function of 70% credence in rain in a story about the meaning-investing use of mental sentence . The story uses the signature function of 70% credence in rain to reveal what it is in virtue of which  means that the probability of rain is 70%. This too vitiates the reductive ambitions of credence-as-belief. After all, the very credence targeted for reduction by the computationalist is itself an explanatory primitive—an unexplained explainer—in her approach to a key resource meant to effect the reduction. This is no good. The picture put forward here maintains that manifesting the very function targeted for reduction is part of what it is for a mental sentence to mean what it does; but the picture also maintains that standing in B to a mental sentence with that meaning is what reduces the credence associated with the signature function. The result is manifestly at odds with the reductive ambitions of credence-as-belief.6

Horn 2 To avoid these sorts of problems a defender of credence-as-belief must specify the signature function of credal-making belief in a way which makes sense of that belief in its own terms, 6

Further: contrary to common-sense, the picture before us strongly suggests that credence is not subject to epistemic evaluation in the first place; for it locates credence as an explanatory primitive within psychosemantics, not as an attitudinal component of a psychology involving states subject to such evaluation.

10

with neither direct nor indirect appeal to the signature function of its reductive target. Yet a defender of credence-as-belief must also make clear that the manifestation of the signature function of credal-making belief brings with it the manifestation of the signature function of that target. Only then will the view be well grounded in theory. Doing the first of these things makes doing the second much harder. To see this, suppose that for any claim  the signature function of belief in  is the disposition to say that  is true. Then a defender of credence-as-belief can make sense of the signature function of credal-making belief quite independently of the signature functions of credence. She can note that 70% credence in rain is nothing but belief that the probability of rain is 70%; and she can go on to use the supposition at the start of this paragraph to make sense of the signature function of this credal-making belief. But the resulting position entails that the signature function of this credal-making belief is

(SFB) = the disposition to say that the probability of rain is 70%,

whereas the illustrative signature function of 70% credence in rain is

(SFC) = the 70% disposition to say that it will rain.

One can easily makes sense of (SFB) without appeal to (SFC), so one can easily make sense of the credal-making belief here without appeal to the signature function of its reductive target. The first task mentioned above is done without difficulty. But the second task is now out of reach. After all, possession of (SFB) does not guarantee possession of (SFC): being disposed to say that the probability of rain is 70% does not ensure a 70% disposition to say that it will rain. Relative to our illustrative suppositions, then, manifestation of the signature function of credal-making belief does not guarantee manifestation of the signature function of its reductive target. In turn that means the credal-making belief is no such thing: it does not make for credence. This is another failed defence of credence-as-belief.

I have no proof that every such defence will fail, much less one showing that every such defence will fail in a fashion similar to those we have seen. I do think, though, that working through the failures we have seen generates rational scepticism that credence-as-belief can be successfully defended. And that seems to be due to a simple fact, namely, that there are three things any full-dress defender of credence-as-belief must do:

(A)

She must specify the signature function of credal-making belief.

(B)

She must show that this function is neither identical to nor built directly from the signature function of its reductive target; and 11

(C)

She must show that possession of this function entails possession of the signature function of that target.

Doing (A) ensures that credence-as-belief is more than just talk. Any defender who fails to do (A) will not have been sufficiently clear about what credence is meant to be. Doing (B) ensures that a defender of credence-as-belief avoids using credence in her understanding of credal-making belief. Failure to do so—either by equating the signature function of credalmaking belief with the signature function of its reductive target, or by building the signature function of credal-making belief directly from the signature function of that target—vitiates the reductive ambitions of credence-as-belief. Either move undermines the very point of credence-as-belief. Yet a defender of the view must also do (C) to ensure that credence-asbelief is a genuinely reductive position, to ensure that it succeeds in making clear that credence is not only real but really something else.7 The challenge is to do (A), (B) and (C) together. Our discussion makes clear that surmounting this challenge is no easy thing. It is unobvious how to understand belief-inprobability so that our take on it is independent of the signature function of credence, yet our take on belief-in-probability underwrites the idea that so believing ensures that we function as credal agents function, that we manifest those functional properties constitutively possessed by those who lend credence. 8

[2] Let us consider, then, the other broad approach to defending credence-as-belief. Instead of looking for something within the content of belief to spell out credence, let us look for something within our use of belief to do so. This approach starts with a reductive schema:

{Cr() = n} is nothing but {Update-disposition[Att()] = n}.

The idea here is that n-level credence is nothing but an n-level disposition to do something or other with our attitudes upon updating. When n is high, it is fairly clear how the line should go: high credence is reduced to a strong disposition to retain belief upon updating. But what is the thought when n is not high? Suppose n is low. It should not then be said that low credence is nothing but low disposition to retain belief upon updating; for that would make low credence entail belief. Presumably, when n is low the thought should be something like this: low credence is nothing but low disposition to acquire belief upon updating. And, by analogy, when n is middling in strength the thought should be that such credence is samestrength disposition to acquire belief upon updating.9 Unfortunately, the approach is bound to fail no matter how it is spelled out; for there is an important and obvious difference between attitude taken to content and disposition to update. The present approach to credence-as-belief elides that distinction. To see this, recall an 7

To augment a phrase of Fodor's: see his (1987), p. 97. Conditions very like (A) thru (C) must be satisfied, mutatis mutandis, by any view which aims to reduce coarse-grained attitudes to fine-grained ones. Section 6 will detail how this can be done. 9 See Harman (1986) for an endorsement of this sort of view. 8

12

important exchange between Popper and Jeffrey.10 The former claimed that an agent’s epistemic state could not be well modelled with a single probability function. He noted that an agent might lend 50% credence when entirely lacking relevant data, contrasting the case with one in which someone lends 50% credence on the basis of copious amounts of data. Popper thought a single probability function was insufficient to mark differences between such agents. This led him to conclude that more than one probability function was needed to do so. Jeffrey responded by appeal to conditional credence. He noted that in Popper’s scenario one agent’s credences for the claim in question conditional on various bits of information look very different than the other agent’s credences for that claim conditional on said information. Jeffrey used this divergence in conditional credence to mark intuitive differences manifested by Popper’s agents. Moreover, Jeffrey endorsed the further idea that updating dispositions are themselves marked by conditional credence; so his response to Popper rested ultimately on the idea that two agents could invest the same credence in a claim yet manifest divergent credal-updating dispositions with respect to it. I do not agree with Jeffrey about his further idea. It does not seem to me that conditional credence is (or even marks) how an agent will (or should) update upon receipt of new information. But that difference between us is not presently relevant. After all, Jeffrey is surely right to think that agents can invest equal credence in a claim yet diverge in their dispositions to update credence. There is an important and obvious difference between lending credence to a claim and any disposition to shift credence in it. This is precisely the sort of difference needed to distinguish Popper's two agents. A similar point cuts against any attempt to reduce credence to the (rational) flux of coarse-grained states. For instance, two agents can adopt the same credence in  yet diverge in their disposition to update belief in , disbelief in , or an attitude of suspended judgment. Perhaps they cannot do so without also differing in their conditional beliefs: that is not presently relevant. What matters here is that adopting credence is one thing and being disposed to update is another. This is true when said disposition is indexed to a shift in finegrained states such as credence. It is also true when said disposition is indexed to a shift in coarse-grained states such as belief, disbelief or suspended judgment.

There are two main strategies for the reduction of credence to belief. One involves locating something to match credal strength in the content of belief. The other involves locating something to do so in the use of belief. Neither strategy is promising. We do not rationally believe contents with components to match credal strength. Nor does our use of belief contain the reductive seed of that strength. It does not look as if credence reduces to belief at all: not to belief in probability, not to rational or irrational shift in belief.

4.

10

Reducing Coarse-grained Attitudes to Credence.

Popper (1959: 414-415) and Jeffrey (1965: 195ff).

13

A more promising unification of the attitudes is one I call credal-based Lockeanism. This is the view that belief, disbelief and suspended judgement are credence in the rough.11 The picture is something like this:

2

Figure 1

Credal-based Lockeanism says that belief is nothing but sufficiently strong credence, disbelief is nothing but sufficiently weak credence, and suspended judgment is nothing but middling-strength credence.12 Faces in the credometer indicate my level of sympathy with a given aspect of the view. Before detailing various problems, however, it will be helpful to sketch the considerable good news surrounding credal-based Lockeanism.13

(i) Realism about the attitudes. Folk psychology is shot through with coarse- and finegrained attitudes. We routinely use each kind of attitude in a hugely successful predictive 11

See Foley (1993), Kaplan (1998)…. The approach will be plausible only if it is allowed that the level of credence needed for belief or disbelief is vague, and that the level of credence needed for belief or disbelief shifts with context. 13 We shall only touch on the relevant points. For further discussion see… 12

14

and explanatory practice. The scope and success of that practice is so great that the practice itself is all but invisible. But our use of coarse- and fine-grained attitudes in the prediction and explanation of one another is nothing less than a major success at the heart of social reality. It is a good thing, then, that credal-based Lockeanism entails realism about coarseand fine-grained attitudes. Eliminativism about either of them would render the success of our attitude-based practice an utter mystery.

(ii) The Marching-in-Step phenomenon. Human action looks to spring from belief exactly when it springs from high confidence. And such action looks to spring from disbelief exactly when it springs from low confidence. These facts generate the impression that human action springs from suspended judgement exactly when it springs from middling-strength confidence. Hence coarse- and fine-grained causes of action look to march in step with one another. That is exactly what you would expect if coarse- and fine-grained attitudes stand to one another as graded determinable stands to grade of determinate. Yet credal-based Lockeanism maintains precisely that coarse-grained attitudes are graded determinables with fine-grained attitudes as grades of determinate. Hence the causal marching-in-step of coarseand fine-grained attitudes looks entirely natural from the view’s perspective.14

(iii) Linguisitic Practice. We say things like: “Maja and Sascha both believe that snow is white, but Maja believes it less firmly than Sascha” or “Fritz disbelieves that snow is white, but he does not do so fully”. Everyday description of coarse-grained attitudes prompts the idea that belief and disbelief are graded, that they come in strengths. Everyday linguistic practice seems to underwrite the idea that you and I can each believe the same thing, for instance, even though what we thereby have in common comes off more strongly in my hands than it does in yours. Given the belief-making threshold is below certainty, that is exactly what credal-based Lockeanism entails. The view squares well with quotidian description of epistemic attitudes.15

This is a non-trivial amount of good news. Credal-based Lockeanism underwrites robust realism about the attitudes, with commonsense. It clarifies how causal profiles of epistemic attitudes fit together, in line with folk psychology. It helps make sense of how we ordinarily speak of belief, disbelief and their degrees. These good-making features of the view jointly make credal-based Lockeanism an attractive position. They seem jointly to be sufficient to render small, for now anyway, the notional chance that there is no truth in the view. Having said that, credal-based Lockeanism faces stiff worries on at least two fronts. One concerns the psychological implications of the view. The other concerns its epistemic implications. Consider them in turn.

1. Psychology. Credal-based Lockeanism is built from three bi-conditionals. In short form they are

14 15

For classic discussion of the causal marching-in-step of determinables and determinates, see Yablo (1992). See Stanley (2005) for further discussion of this sort of linguistic practice.

15

(B)

Belief



Strong credence

(D)

Disbelief



Weak credence

(S)

Suspended Judgement



Middling credence.

Credence counts as strong here exactly when it does not fall below the Lockean beliefmaking threshold, weak exactly when it does not fall above the Lockean disbelief-making threshold, and middling in strength exactly when it falls between such thresholds. Credalbased Lockeanism is the conjunction of (B), (D) and (S). Unfortunately, the left-to-right direction of each bi-conditional is false. To see this, focus on an obvious fact about our mental life: we do not always invest credence when we invest confidence. After all, credence is point-valued subjective probability. To invest credence is to configure an exact spread of confidence across every niche of epistemic possibility. To invest credence is to adopt a hyper-precise epistemic attitude. We do manage it from time to time, in my view, but mostly we do not invest credence when investing confidence. Mostly our take on things is not so exact. For instance: suppose a given coin has a 95% bias to heads. I know this and consider whether the coin will land heads on the next toss (H). In the event, I will invest 95% credence in H. This does not seem to be metaphor or literally untrue idealisation. It seems plainly the case that in these circumstances I will be 95% sure that the coin will land heads on the next toss. I will adopt a hyper-precise attitude to H. I will invest credence. But consider the claim T: that it will be sunny in Tucson tomorrow. As it happens I am highly confident that it will be sunny in Tucson tomorrow. Indeed I would go further and say that I believe that it will be sunny in Tucson tomorrow, and, that I do so precisely in virtue of being highly confident that it will be sunny in Tucson tomorrow. But there is no credence that I invest in T, no hyper-precise confidence that I invest in the claim that it will be sunny in Tucson tomorrow. My take on the issue is rougher than any state of credence. I am highly confident that it will be sunny in Tucson tomorrow; but no real number r is so that the strength my confidence in T is r% of certainty's strength. I invest credence in T, though, only if there is such an r. Hence I invest no credence in T. My high confidence is no state of credence at all, yet it makes for belief in T. So it is possible to believe in the absence of credence. The left-to-right direction of (B) is false. A similar point holds for (D). Just think of the claim ¬T: that it will not be sunny in Tucson tomorrow. Being clear-headed and investing high confidence in T, I invest low confidence in ¬T. Indeed I would go further and say that I disbelieve that it will not be sunny in Tucson tomorrow precisely in virtue of investing low confidence in ¬T. But just as there is no credence that I lend to T, there is no credence that I lend to ¬T, no hyper-precise confidence invested in the claim that it will not be sunny in Tucson tomorrow. My take on the matter is rougher than any state of credence. I do invest low confidence in the claim that it will not be sunny in Tucson tomorrow. But no real number r is so that the strength my confidence in ¬T is r% of certainty's strength. I invest credence in ¬T only if there is such an r. So I invest no credence in ¬T. My low confidence in ¬T is no state of credence at all, yet

16

it makes for disbelief in ¬T. So it is possible to disbelieve in the absence of credence. The left-to-right direction of (D) is false. And a similar point holds for (S). Just think of the claim V: that there is a smelly vinegar factory in Slough. I have no idea if this claim is true, having no evidence on the matter to speak of. Yet I do lend some kind of confidence to V (now that I think of it, at least). And it is clear that the confidence I lend is neither strong nor weak. Indeed I would go further and say that I suspend judgment in V precisely in virtue of lending such middlingstrength confidence to the claim that there is a smelly vinegar factory in Slough. But there is no credence that I invest in the matter, no hyper-precise confidence that I lend to V. My take on things is rougher than any investment of credence. I do have middling-strength confidence in V. But no real number r is so that the strength of that confidence is r% of certainty's strength. I invest credence in V only if there is such an r. So I invest no credence in V. My middling-strength confidence in V is no state of credence at all, yet it makes for suspended judgment in V. So it is possible to suspend judgement in the absence of credence. The left-to-right direction of (S) is false. The left-to-right directions of (B), (D) and (S) entail, contrary to psychological fact, that belief/disbelief/suspended judgment require the investment of credence. Nothing of the sort is true. It is possible to believe, disbelieve or suspend judgment without investing credence. And it is psychologically common to do so. The point does nothing to suggest that belief, disbelief or suspended judgement can occur in the absence of confidence. It simply makes clear that credence is not needed for any of the coarse-grained attitudes. As a matter of psychological fact, any of them can be adopted without adopting credence. This conflicts with credal-based Lockeanism.

2. Epistemology. Along with psychological worries for the view there are epistemic worries as well; and just as the former spring from obvious aspects of our mental life, so the latter spring from obvious aspects of our epistemic life. In particular, they spring from the incompatibility of credal-based Lockeanism and a highly intuitive epistemic principle. I call this principle the “Norm of Character Match”. In a nutshell, it says that one should adopt an attitude to a content only if that attitude’s character matches the character of evidence used in its formation. The sense of match at work in the norm is best gleaned by example. We'll systematize things later in a way which makes the norm more transparent, tractable and theoretically fecund. For now we work with intuition on a case-by-case basis. For instance: when your weather-related evidence comes solely from a fully trusted source, and that source says that there is an 80% chance of rain, then, on that basis, you should be 80% sure that it will rain.16 If your source says only that it is likely to rain, however, or only that there is a strong chance of rain, or anything like that, then, on that basis, your take on the weather should be appreciably rougher. You should be confident that it will rain in such evidential circumstances, on the basis of your source’s testimony; but there is no exact credence that you should have, in such circumstances, that it will rain. The point is well known in the literature.17 Everyday evidence is often meagre and vague, too weak to rationalize credence. The view that belief-making confidence should 16

See Lewis (1980). Classic discussion of thick confidence can be found in Hacking (1975), Jeffrey (1983) and Levi (1974). More recent discussion can be found in Halpern (2003), Joyce (2005), Kaplan (1998) and Walley (1991). 17

17

always be credence conflicts with the Norm of Character Match. After all, credence is a hyper-precise spread of conviction which ranges, as it were, over every niche of epistemic space. Quotidian evidence routinely fails to rationalise such hyper-detailed conviction, even under idealisation. Such evidence routinely demands an attitude which is confidencetheoretic yet coarser than credence, something I call thick confidence. We'll say more about how to think of such confidence in the next few sections—both formally and informally—but here we note merely that everyday evidence is often too weak to warrant the adoption of credence, and, for this reason, that the Norm of Character Match ensures that such evidence often warrants the adoption of thick confidence rather than credence. To adopt the latter on the basis of such evidence is to outrun one’s evidential headlights, so to say, and to do so precisely with respect to the character of the attitude adopted on the basis of one’s evidence. Fortunately, neither credence nor thick confidence is a theoretical postulate. Common-sense recognizes being 50% sure that it will rain, for instance, a paradigm credence if ever there were one; but it also recognizes, as we have seen in our discussion of psychological objections to credal-based Lockeanism, that there are states of confidence untethered to credence (like being non-credally confident that it will rain). Both credence and thick confidence are manifest components of everyday psychology, manifest elements of our attitudinal repertoire. And this is very much an epistemic good thing, for both types of confidence are mandated by total evidence routinely in our possession. Credal-based Lockeanism is incompatible with this fact about our epistemic life.

Credal-based Lockeanism insists that course-grained attitudes spring from credence. This is false as a matter of psychological fact and wrong as a matter of epistemic norm. It is possible to believe, disbelieve and suspended judgement without lending credence. Often our evidence mandates doing exactly that. Nevertheless, we have seen that credal-based Lockeanism generates such good news that it is hard to suppose there to be no truth in the view. This makes it natural to augment credal-based Lockeanism to put forward a view on which all states of confidence make for coarse-grained attitudes. The result is a liberal version of Lockeanism: it is still said that confidence is required to make for belief, disbelief or suspended judgement; but now it is allowed that coarse-attitude-making confidence can be credal or non-credal in character. Such a Lockeanism sidesteps both the psychological and epistemic difficulties faced by its credal-based cousin. Our next task is to develop the view.

5. Confidence as Mischung. Our discussion of coarse-grained attitudes ended with the idea that belief, disbelief and suspended judgement fail to inter-reduce. Each of them is on an explanatory par with the others. I shall use a similar thought to develop a reductive theory of confidence—both credence and thick confidence—and then I shall use that theory to create a new kind of confidence-first approach to coarse-grained attitudes. The perspective developed will be similar to those which use belief, disbelief and suspended judgment in a theory of confidence. It will also be similar to approaches which use confidence in a theory of belief, disbelief and suspended judgment. The trick will be to

18

locate resources more basic than any attitude of epistemic appraisal, coarse- or fine-grained. Those resources will now be unearthed with an analogy. Suppose a waiter asks you to sample a piece of Limburger cheese. You’ve never tried Limburger before, but you know that it is not to everyone’s taste. You inspect the cheese carefully, prodding it with a finger, sniffing it with your nose. One of three things is then likely to happen:

(i) (ii) (iii)

You are drawn to the cheese; You are repelled by the cheese; You are indifferent to the cheese.

The first upshot involves gustatory attraction: a force which drives you to the cheese, prompting a taste, a sniff, a bite. The second upshot involves gustatory revulsion: a force which drives you away from the cheese, prompting avoidance of it outright. The third upshot involves gustatory indifference: neutrality which sits between you and the cheese, prompting a stable laissez-faire disposition with respect to it. Each of these reactions comes in degrees, to be sure, but set that aside for the moment. Focus instead on the fact that there is good pretheoretic reason to recognize three kinds of gustatory force: attraction, repulsion and neutrality.18 There is also good reason to insist that these forces can cohabit, that they can arise side-by-side in reaction to food. A chunk of Limburger can attract and repel at the same time, for instance; and when this happens instability can be the result. There can be gustatory battle between powerful elements of one’s reaction to food, gustatory tension, as it were, which makes for behavioural instability. When that happens no strong or stable disposition to act with respect to the food springs from your gustatory reaction to it. But there needn’t be such instability generated by one's reaction food. There may be no gustatory conflict at all: one’s reaction to Limburger may be solely one of attraction or revulsion or neutrality, for instance; and then an easy gustatory equilibrium is reached. Or— more importantly for the analogy being developed—there may be functionally harmonious mixtures of gustatory force, stable blends of attraction tempered by a whiff of revulsion, for instance, or revulsion muted by a whiff of neutrality. These are complex ways of achieving functional stability in relation to food, complex-but-stable reactions to food. We may think of gustatory equilibria, then, as stability points which arise from the interaction of gustatory attraction, revulsion and neutrality. To a rough first approximation, manifesting such a stability point is manifesting a strong and stable disposition to eat a given piece of food, to push it away, or to be neutral concerning interaction with it, with the overall disposition itself springing from the interaction of gustatory attraction, repulsion and neutrality. And each of these gustatory forces itself carries a signature function of its own.

18

I also note that there is a salient difference between gustatory attraction and repulsion on the one hand and neutrality on the other. The former seem to be forces which happen to one, whereas the latter seems more like a force one creates. I find myself being attracted or repelled by a piece of food; but I often emit, as it were, neutrality about it. There is no need here to work out the asymmetry. It is worth noting, though, in the gustatory case. Something very like it also seems present in the attitudinal case to follow.

19

The main contention of this paper is that all of this carries over, mutatis mutandis, to attitudes of epistemic appraisal. To see how, suppose you are asked to consider the claim C: that Mars is larger than Venus. You understand C fully but have never wondered whether it is true. You comb through your information for anything relevant to whether Mars is larger than Venus. One of three things is then likely to happen:

(iv) (v) (vi)

You are drawn to C; You are repelled by C; You are indifferent to C.

The first upshot involves intellectual attraction: a force driving you to C in thought, prompting embrace of the claim with your mind. The second upshot involves intellectual revulsion: a force driving you away from C in thought, prompting rejection of the claim with your mind. The third upshot involves intellectual neutrality: mental indifference which sits between you and C, prompting a stable cognitive laissez-faire stance between you two. These reactions also come in degrees, but set that aside as well. Focus instead on the fact that there is good pre-theoretic reason to recognize three kinds of attitudinal force: attraction, rejection and neutrality. Reflection on cases provides strong reason to insist that such forces can likewise cohabit. They too can spring up side-by-side in reaction to a given piece of information. There can be attitudinal battle between elements of one’s take on a claim, for instance; and such battle can also lead to a kind of instability, intellectual instability. This occurs when there is no strong or stable disposition to adopt a take on a given claim. For example, consider a penumbral case of vagueness. Suppose Fritz is hair-wise midway between a clear case of bald and a clear case of non-bald. You inspect his head thoroughly in good daylight and consider whether Fritz is bald. This is a borderline case par excellence, and so the case is thoroughly puzzling. As Crispin Wright notes, the “absolutely basic datum” here is

that in general borderline cases come across as [cases] where we are baffled to choose between conflicting verdicts about which polar verdict applies, [true or false], rather than as cases which we recognize as enjoying a status inconsistent with both.19

The psychological phenomenon induced by such a borderline case involves the interaction of attitudinal force. Once we possess every bit of constitutive evidence there is to be had in such a case—once we know exactly how Fritz is hair-wise, for instance—we are attracted and repelled by the claim that Fritz is bald, with these attitudinal force reactions springing from the very same evidence. This kind of mixture of attitudinal force is different than any

19

Wright (2001: 70)

20

which makes for a stable attraction to the claim that Fritz is some way other than bald or nonbald.20 The result is a special kind of attitudinal instability, the kind which results from conflicting attitudinal forces springing from apt sensitivity to a single corpus of evidence. You are intellectually drawn to the claim that Fritz is bald, on the basis of how he is hair-wise in the case, and you are also intellectually repelled by that claim on that basis. This unique sort of attitudinal-cum-epistemic situation is what generates the distinctive "intellectual vibe" of a borderline case. That vibe consists in attitudinal instability resulting from force-based conflict generated by the same evidence. On the other hand, distinct intellectual forces working on a single mind and a single claim do not always lead to attitudinal instability. Just consider the axiom of comprehension in set theory. It says that whenever a condition is understood well enough to apply clearly in a case, or be ruled out clearly in a case, there will exist a set of things satisfying that condition. Solid comprehension is said by the axiom to be sufficient for the existence of a set, namely, the set of things satisfying the condition so comprehended. It is easy to be attracted to the axiom, as it seems to apply well in everyday cases and it is difficult to see how it could be false to begin with. Yet the axiom of comprehension leads directly to paradox. This makes it easy to be repelled by the axiom too. In my own case I feel both sorts of intellectual force: attraction and revulsion, and, as a result, I settle into an epistemic attitude of disbelief-via-low-confidence in the axiom. This squares perfectly with the view of attitudinal forces being sketched. According to that view, such attitudes are nothing but stable configurations of attitudinal force. Every kind of confidence—both thick confidence and credence—along with every kind of coarse-grained attitude—belief, disbelief and suspended judgment—is nothing but a stable configuration of attitudinal force. In essence the view is that epistemic attitudes are equilibria which grow from mixtures of attitudinal force, attitudes-as-equilibria so to say. It is the view that epistemic attitudes are stability points produced by blends of cognitive force: each type of confidence is a stable mixture of attitudinal attraction, repulsion and neutrality; and each type of coarse-grained attitude is too. First I shall explain how the approach works in the case of credence, and then I’ll generalise the approach to cover thick confidence. After that we’ll see how to recover belief, disbelief and suspended judgement from the framework. This will result in a confidence-first approach to the attitudes which has a great deal in common with views on which coarsegrained attitudes are explanatorily basic. The resulting confidence-first approach to the attitudes should be welcomed by those who are normally unsympathetic to such a view. To begin, I propose that credence is the result of mixing attitudinal attraction, rejection and neutrality in a way that involves no neutral force at all. Put another way: each grade of credence involves a level of neutrality that is literally nothing, mixed with blends of other attitudinal force. I further propose that this is what sets credence apart from thick confidence: thick confidence involves non-zero attitudinal neutrality, potentially mixed with blends of other attitudinal force. As we’ll see when the story unfolds, such neutrality is anathema to credence.

20

Or equivalently: this kind of attitudinal instability is different than being stably attracted to the view that the claim that Fritz is bald enjoys some third semantic status, one incompatible with being true or false.

21

There are several ways to picture the line. Since the main suggestion about credence is that it springs from combinations of acceptance and rejection, take any credence whatsoever. Let its strength be represented by unit-real n. Then picture the credence as composed of acceptance and rejection forces:

|--a--->||--n--|||--n--|