Deontological vs. Non-Deontological Justification
In his seminal paper āConcepts of Epistemic Justification,ā William Alston distinguishes between a deontological and a non-deontological concept.1 According to the deontological concept, Jd, justification is a matter of duty fulfillment. A belief that p, Bp, is deontologically justified, Jd, iff ~O(~Bp), that is, iff one doesnāt have an epistemic obligation to refrain from believing p.2 Alternative definitions are on offer, made available by more or less plausible synonymy relations between various deontic terms. For example, a synonymous alternative definition of Jd is this: Bp is Jd iff Bp is permissible. The definition Alston settles on is this one: Bp is Jd iff Bp is epistemically blameless. While the synonymy of āpermissibleā and ānot obligatoryā is unproblematic, identifying permissibility with blamelessness is not. Iāll return to this point further below.
Alston thinks that Jd is not the same as genuine epistemic justification, Je. He takes it that only a non-deontological concept captures what we have in mind when we classify a belief as epistemically justified. But what would it be for a belief to be non-deontologically justified? According to Alston, a belief is non-deontologically justified, Jnd, if having that belief is a good thing from the epistemic point of view, where the goodness in question has nothing to do with the fulfillment of any epistemic duties.3 This formula doesnāt say much unless we are told what it is for a belief to be a āgoodā belief from the epistemic point of view. Alston proposes a teleological understanding of epistemic goodness. The basic idea is that the kind of goodness that renders beliefs epistemically justified is a species of endsāmeans rationality. Forming beliefs that are epistemically good is the means to achieve the cognitive goal of maximizing truth and minimizing falsehood in a large system of beliefs.4 On this approach, a belief is Je iff it is formed in a truth-conducive way: in a way that is conducive to achieving the cognitive goal just mentioned.
Alston rejects Jd on the basis of two arguments. According to the first, beliefs in particular and propositional attitudes in general are unsuitable for deontological evaluation because we donāt have voluntary control over them. According to the second, Jd fails to secure truth-conducive belief formation and thus falls short of being genuine epistemic justification. It doesnāt deliver the goods when it comes to the goal of maximizing truth in a large belief system.
I will argue that Alstonās arguments against identifying Je with Jd fail. First, however, I will focus on the concept of epistemic duty and explore exactly what our epistemic obligations are.
Unconstrained Evidentialism
It is undeniable that Je is intimately tied up with the notion of truth. Since, as a purely conceptual matter, evidence is evidence of truth, itās natural to link Je to the possession of evidence. Let āEpā stand for āyour total evidence supports pā. To say that your total evidence supports p is to say that your total evidence makes it likely that p is true. Likelihood of truthāthat is what evidence is all about. Starting out with this thought, there is a straightforward way of fixing what our epistemic duty is: it is to believe in accord with our evidence. According to what we might call Unconstrained Evidentialism, this duty breaks down into two sub-duties:
D1~O(~Bp) ā Ep
D2Ep ā O(Bp).
According to D1, Ep is necessary for Bp to be permissible. According to D2, Ep is sufficient for Bp to be obligatory. Each of these duties is surprisingly problematic. Though evidentialists are unlikely to reject D1 as false, they might disagree about the nature of the ought D1 is about. That is, they might have different views on the kind of normativity the antecedent of D1 is about. There are three options: the ought in question is (i) sui generis, constituting an independent and irreducible domain of normativity; (ii) it is the ought of prudential rationality; (iii) it is the ought of morality.5 While option (i) would secure an epistemically pure version of evidentialism, it is not clear to me that (ii) and (iii) pose a threat to evidentialism. Consider (ii). If epistemic normativity is a species of prudential normativity, there will then be two different types of prudential oughts: epistemic and non-epistemic prudential oughts. Here is an analogy: dogs are not a sui generis genus of animal. Rather, they are mammals. But within the genus mammal, dogs constitute their own distinctive species. Likewise with epistemic normativity. Even if epistemic normativity is ultimately prudential, epistemic and non-epistemic oughts will nevertheless be distinctive species, just like dogs and cats are distinctive species of mammals. Analogous reasoning applies to option (iii). Here, I will not argue for or against any of these three options.
The Infinity Problem
The worry about D2 is that it is excessively demanding. If you have evidence for p, you thereby have evidence for all obvious deductive consequences of p. For example, p entails an infinite number of disjunctions: p v q, p v r, etc. Hence, if you have evidence for p, then you have evidence for an infinite number of disjunctions. D2 says you ought to believe p if you have evidence for p. So, D2 generates an epistemic duty to believe an infinite number of disjunctions.6 Thatās problematic. Since we cannot believe an infinite number of disjunctions, D2 generates an epistemic duty we cannot meet. As a result, D2 makes epistemic sin inescapable. No matter how epistemically virtuous you are, you cannot help failing massively in the attempt to live up to epistemic duty. A normative theory with such consequences, it seems to me, is highly suspect.
The epistemic duty to believe an infinite number of deductive consequences can be avoided by limiting the antecedent of D2. We could say that you have a duty to believe p if you have evidence for p and you consider p. This gives us:
D2* (Ep + Cp) ā O(Bp).
According to D2*, itās only your considered evidence that determines what you ought to believe. Letās call this view C-Constrained Evidentialism. Since you cannot consider an infinite number of propositions, this version of evidentialism does not generate a duty to believe an infinite number of deductive consequences. Alas, C-Constrained Evidentialism runs into another problem.
The Epistemic Felicity Problem
Suppose the following is true of me:
Unless something weird is going on, we can go from (1) to:
Since (1) obviously entails a trivial and utterly uninformative disjunction such as
(3)Iām hungry or Napoleon is hungry,
we have as well:
(4)E(Iām hungry or Napoleon is hungry).
Suppose I consider (3). D2* tells us that I ought to believe (3). Is this plausible?
An epistemic duty to believe a disjunction need not be objectionable. Suppose Sherlock Holmes has two strong pieces of evidence. The first suggests that the butler did it, the second that the gardener did it. Itās plausible to say Holmes ought to believe:
(5)The butler did it or the gardener did it.
It seems to me, however, that there is a significant difference between Holmes believing (5) and me believing (3). Here is a Gricean rule of felicitous conversation: āMake your contribution as informative as is required (for the current purposes of the exchange).ā7 Suppose Watson were to ask Holmes: āWhom do you suspect?ā If Holmes asserted (5), he would not be violating this Gricean rule. However, if someone asked me whether I was hungry and I replied by asserting (3), I clearly would. Could it be that, whereas asserting (3) would be conversationally infelicitous, adding (3) to my belief system would be epistemically felicitous? Letās see whether saying ānoā can be defended.
Here is a suggestion I will briefly explore: we should distinguish between evidential support and epistemic felicity. Ep supports Bp iff Ep raises the probability of p above 0.5. Bp is epistemically felicitous iff Bp fits your evidence. Evidential support is a necessary condition of evidential fit. But evidential support is not sufficient for evidential fit. Thus, we get the following possible situation: although your total evidence supports believing p, believing p would be epistemically infelicitous.
According to evidentialism, itās our duty to follow our evidence. Arguably, that doesnāt mean that, according to evidentialism, itās our duty to believe everything our total evidence supports. Rather, evidentialism, properly construed, tells us that we ought to proportion our beliefs to our evidence. If we do that, our beliefs fit our evidence.8 To proportion our beliefs to our evidence, we must, as a first approximation, do two things: not believe more, and not believe less, than our considered evidence supports.9 For example, when I am hungry and I believe (3) instead of (1), then I believe less than my evidence supports. Therefore, if I believe (3) but not (1), I donāt proportion my belief to my evidence. This makes my believing (3) epistemically infelicitous.
But what about believing (1) and (3)? Suppose I consider both (1) and (3). Let E1 be my awareness of my being hungry. Let E2 be my recognition that that (1) entails (3). E1 supports believing (1). E2 supports believing (3). E1 and E2 jointly support believing (1) and believing (3). If I acquire both beliefs, I believe (3) in addition to (1). Now I no longer believe less than my evidence supports. Nor do I believe more than my evidence supports.
According to evidentialism, do E1 and E2 make it the case that I ought to believe both (1) and (3)? I would prefer to say ānoā. The proposal Iām exploring defends saying ānoā on the ground that having both beliefs would not be epistemically felicitous. For this move to work, epistemic felicity would have to involve more than believing no more and no less than oneās evidence supports. Perhaps we could say that epistemic felicity is to believe no more and no less than oneās evidence informationally supports. Compare:
(1)Iām hungry.
(3)Iām hungry or Napoleon is hungry.
(7)Iām hungry and Napoleon is hungry.
Itās plausible to say that (3) is weaker than (1) and (7) is stronger than (1). (3) carries less, and (7) carries more, information than (1) does. This appeal to informational strength calls for elaboration. Here, I will deploy the concept of informational strength without offering anything in the way of an analysis of that concept. I merely want to suggest that the following claim might be defensible: believing p is epistemically felicitous iff believing p fits the informational content of my evidence.
So, given E1 and E2, what ought I to believe? According to the proposal under consideration, while I ought to believe (1), itās not the case that I ought to believe (3). Believing (3) in addition to (1) does not add informational content to my belief system. Therefore, believing (3) does not fit the informational content of my evidence. So, believing (3) would be an ...