Ought to believe, simpliciter

Abstract

According to many philosophers there are only pro tanto oughts to believe relative to a standard of assessment: there are epistemic oughts to believe, moral oughts to believe, prudential oughts to believe etc. But there are no oughts to believe simpliciter. Many of the same philosophers who hold this view, also hold that ought to believe is to be understood deontologically – such that if S violates such an ought without excuse, S is blameworthy for doing so. I here argue that on a deontological understanding of ought to believe there must be ought to believe simpliciter and that it is the violation of this ought that determines whether we are to blame for our beliefs.

1. The thesis

Some philosophers think that there are epistemic duties (obligations, or requirements) that render us blameworthy when we violate them without excuse. For instance, some philosophers think that epistemic justification should be understood deontologically. These philosophers think that epistemic justification should be understood in relation to whether or not some epistemic duty has been violated; for instance, they analyse justification as follows: S is justified in believing that p iff/only if S is blameless in believing that p (Chisholm 1988, notably). ‘Epistemic Ought’ for these philosophers is to be understood deontologically too, in the sense that to say that ‘S epistemically ought to believe that p’ is to say that ‘S has an epistemic duty/obligation/requirement with respect to the belief that p’ such that if S violates the requirement (without excuse) S would be blameworthy. Let's call this thesis Epistemic Ought Deontologism.

Epistemic Ought Deontologism: S ought_epistemic to believe that p iff S has an epistemic duty/obligation/requirement x with respect to the belief that p, such that were S to violate x (without excuse) S would be blameworthy.

Note that this is to understand “deontologism” in a “strong” sense, where it involves blame and resentment – the idea is that epistemic duties have to be understood to be analogous to paradigmatic ethical duties. Thus, under this construal, if epistemic duties turn out to be very different from moral duties then we have reason to abandon deontologism.¹

But many (if not all, or most) philosophers who hold Epistemic Ought Deontologism also hold the following view:

No Doxastic Ought Simpliciter: There is no such thing as doxastic ought simpliciter, all things considered or just plain ought to believe.

According to proponents of No Doxastic Ought Simpliciter, there are only pro tanto oughts to believe relative to a standard of assessment: there are epistemic oughts to believe, moral oughts to believe, prudential oughts to believe etc. But there are no oughts to believe simpliciter. No Doxastic Ought Simpliciter seems to be the received view; two prominent supporters are, for example, Kelly (2003), and Feldman (2000).² As they put it, respectively:

In cases in which what it is epistemically rational to believe clearly diverges from what it is practically advantageous to believe, there is simply no genuine question about what one should believe: Although we can ask what one should believe from the epistemic perspective and we can ask what one should believe from the practical perspective, there is no third question: what one should believe all things considered. In any case in which practical and epistemic considerations pull in opposite directions, there is simply nothing to be said about what one should believe all things considered. (Kelly 2003: 619)

For each “ought” there is an associated value. We ought, in the relevant sense, to do the thing that maximizes that value, or perhaps something that does well enough in achieving that kind of value. For example we morally ought to do what maximizes, or produces enough, moral value. We prudentially ought to do what maximizes, or produces enough, prudential value. If there is such a thing as “just plain ought” then there is a value associated with it. The thing we just plain ought to do is the thing that comes out highest, or high enough, according to that measure. It's far from clear what that value would be – it isn't to be identified with any of the more determinate kinds of value and there seems to be no uniquely correct (or range of correct) ways to combine moral, practical, epistemic, and other values. We've disambiguated “ought” and we can't put the various senses back together again. There is no meaningful question about whether epistemic oughts “trump” or are trumped by other oughts. (Feldman 2000: 663–4)

Both Feldman and Kelly have at least sometimes held both Epistemic Ought Deontologism and No Doxastic Ought Simpliciter. They are not obviously contradictory and lots of other philosophers defend this combination of views explicitly or implicitly – interesting recent examples include Stapleford (2015) and Brown (2020), whose paper has generated a new literature defending the idea of there being uniquely epistemic blame that attaches to such duties (e.g. Boult Forthcoming; Piovarchy Forthcoming). However, in this paper I will argue that in fact Epistemic Ought Deontologism and No Doxastic Ought Simpliciter are incompatible.

Below is my “master argument” for this claim.

The argument from identifying what is uniquely epistemic about epistemic requirements

1. 1. Deontological epistemic requirements/oughts are epistemic because:

   1. (a) they exclusively involve a unique object in the evaluation (belief).

   2. (b) they are made relative to an epistemic norm or standard.

   3. (c) they involve/warrant a special kind of epistemic blame or resentment, different from (say) moral resentment [the evaluation involved is of a different sort].

2. 2. f (1,a), there is doxastic ought simpliciter.

3. 3. If (1,b), there is doxastic ought simpliciter.

4. 4. If (1,c), there is doxastic ought simpliciter.

5. 5. There is doxastic ought simpliciter.

The rest of this paper is a defence of this argument. In section 2, I defend 2; In section 3, I defend 3; In section 4, I defend 4; Finally, in section 5 I give an account as to why it matters that the thesis under consideration is true. In summary, my answer is that it shows us that Epistemic Ought Deontologism is unstable, and I will show how this means that the ethics of belief (conceived of as concerning the question of what we ought to believe) should be thought to have independent standing, outrunning the concerns of epistemology. Further, that Feldman's concerns above, less than refute the idea of there being all things considered oughts to believe, illustrate certain constraints on our meta-epistemology.

Before I do that, let me just say something about how I will read premiss 1 above: I will treat 1(a), 1(b) and 1(c) as an exclusive (disjunction) for ease of exposition. However, the arguments for 2, 3, 4 will show the problems with an account that treats any of these conjunctively. That is because, I hope to show, any account that invokes 1(a) alone or in combination with another explanation will automatically entail that there is doxastic ought simpliciter, and that the explanation in 1(c) is identical to that of either 1(a) or 1(b) – so could not add anything as part of a conjunctive explanation.

2. Deontological epistemic oughts are epistemic because they exclusively involve a unique object in the evaluation (belief)

Perhaps we can use the word ‘epistemic’ and ‘doxastic’ interchangeably (indeed many people do): so that when we talk about epistemic ought, we are exclusively talking about what we ought to believe. The question [ought I to believe that p?] is no different from the question [ought I to epistemically believe that p?], and there is no sense in evaluating any non-doxastic state from an epistemic point of view. If this is true, then we can identify whether an ought is epistemic simply by considering the kind of object of evaluation – if the ought pertains to a belief, it is an epistemic ought; if the ought does not pertain to a belief then it is not an epistemic ought. Now, we should be careful to notice here that the antecedent in premiss 2 is not the claim that epistemic oughts govern only beliefs. The claim that epistemic oughts pertain only to beliefs is compatible with the claim that there are non-epistemic oughts pertaining to beliefs. As such, we could not identify the epistemic ought solely by invoking the object of evaluation. Just knowing that we are considering whether we ought to believe would not tell us whether we are considering epistemic or other kinds of ought. So, the antecedent in premiss 2 is the stronger claim that only beliefs are subject to epistemic ought, and there can be no non-epistemic oughts governing belief.

Sometimes this view – that there are no non-epistemic oughts for belief – is called ‘evidentialism’ (see, for example, Shah 2006; Skorupski 2010; Way 2016).³ Note that this is a different kind of ‘evidentialism’ from the ‘evidentialism’ according to which S is justified in believing that p iff S believes that p on the basis of sufficient evidence (see, inter alia, Feldman and Conee 1985).

If this former kind of evidentialism is true, and there are no non-epistemic oughts for belief, it is at least plausible that the only ‘ought’ pertaining to belief is epistemic ought.

Can this warrant the further claim that there is no doxastic ought simpliciter, only epistemic ought? Unfortunately, I do not think that it can. There is a simple reason why not: if there are no other oughts pertaining to belief then all there is epistemically to consider relative to whether one ought to believe that p is all there is tout court to consider relative to whether one ought to believe that p! That is, what one ought epistemically to believe turns out to be identical to what one ought simpliciter to believe – since if there is nothing but epistemic considerations relevant to the question of what one ought to believe, settling whether one ought_epistemic to believe settles the question of what one ought all things considered to believe.

The argument here is a simple one, and that might lead one to wonder how anyone might be attracted to 1(a). My own conjecture is that this is due to the ambiguity of the term ‘Evidentialism’. Richard Feldman has argued, for example, for both No Doxastic Ought Simpliciter, and Evidentialism. So there is a precedent to combine these two views, and indeed to see them as intimately related. However, as he sees it, Evidentialism is compatible with the notion that there are non-epistemic oughts pertaining to belief. So I think the confusion may arise due to confusion as to what ‘Evidentialism’ is really committed to.

That is all that is needed, I think, to defend premiss 2. Given that the antecedent is 2 is the strong claim that only beliefs are subject to epistemic ought, and there can be no non-epistemic oughts governing belief, any conjunctive explanation for why epistemic requirements are epistemic that invokes 1(a) would be redundant. 1(a) would do all the work in the explanation, and as such, any putatively conjunctive explanation that invokes 1(a) would entail the existence of doxastic ought simpliciter. I now move on to my defence of 3.

3. Deontological epistemic oughts are epistemic [solely] because they involve epistemic norms/standards

The idea here is that we can identify epistemic oughts via invocation of the standard by which we are appraising a belief, typically involving an assessment with respect to the twin ends of believing truths or avoiding falsehoods, or the aim of knowledge (I will here take it for granted that there is some distinctively epistemic aim that would generate such a norm, but I need not here specify what it is exactly). I would like to levy three sub-arguments for premiss 3 – the claim that if we individuate deontological epistemic requirement via appeal to such a norm, there is doxastic ought simpliciter. Here is the first:

3.1. Sub-Argument (i)

Given premise 2, let us start this sub-argument by supposing that there are different norms/standards applicable as to whether to believe that p. For instance, suppose that there are moral norms pertaining to what we should believe, epistemic norms or standards pertaining to belief, prudential norms pertaining to belief etc.

Now, perhaps (at least sometimes) these norms or standards are commensurable. By ‘commensurable' I mean that we can weight them relative to one another, such that we can say that moral reasons to believe that p can outweigh epistemic reasons to suspend judgement on whether p. That is, there is a way, for example, to know what to believe when one ought_epistemic to believe that p, but one ought_moral to suspend judgment on whether p. If (at least some of) these norms or standards are commensurable, there is doxastic ought simpliciter – there is (sometimes) a way to weight competing oughts such that we can make an all things considered judgement regarding what we ought to believe. All things considered doxastic ought exists, therefore, if at least sometimes (or at least some of) the different norms governing belief are commensurable.

So the defender of No Doxastic Ought Simpliciter, it is clear, should claim that the different types of norms and standards governing are incommensurable.⁴ This looks like Feldman's view in the quote I started with, for example. I will now argue that this shows No Doxastic Ought Simpliciter to be incompatible with Epistemic Ought Deontologism.

Two really important points are in order here. First, we should not confuse the claim there is no doxastic ought simpliciter with the claim that there can be genuine doxastic dilemmas. (Where one is in a genuine dilemma, given Deontologism, when one is subject to competing oughts such that one cannot evade being blameworthy.) The claim that there are all things considered doxastic oughts is compatible with the claim that one can end up in a doxastic dilemma where: all things considered S ought_epistemic to believe that p and S ought_moral to suspend judgement on whether p, for example.⁵ Second, relatedly, while the defender of No Doxastic Ought Simpliciter will be happy to grant the existence of doxastic dilemmas within a domain – e.g. within the epistemic domain – she cannot think that there will be genuine dilemmas across domains. That is, it is incompatible with her view that there are dilemmas where, for instance, S ought_epistemic to believe that p and S ought_moral to suspend judgement on whether p. If there is no overarching standpoint from which these oughts can be weighted, and they are incommensurable, how could we come to the judgement that we are in a genuine dilemma? Recall we are here assuming that we are either blameworthy or not, and there are not substantively different kinds of blame (we will consider whether there can be such a thing as epistemic blame in section 5). No Doxastic Ought Simpliciter does say that there is no fact of the matter about what you ought – all things considered – to believe in such cases, but that is different from saying you are in a dilemma. Why? Because, as per the first point above, saying there is doxastic ought simpliciter is not to say there can be no genuine dilemmas. Saying there are genuine doxastic dilemmas and saying that there is no fact of the matter about what you all things considered ought to believe are two logically independent claims.

So the adherent of No Doxastic Ought Simpliciter claims that, where moral ought and epistemic ought pull in different directions with respect to what we ought to believe, there is no fact of the matter as to what one ought all things considered to believe. But this is not to admit that there is a dilemma, since we cannot say whether we are blameworthy come what may. She might say that we are a kind of blameworthy come what may: epistemically blameworthy, when we violate the epistemic ought in favour of the moral, or morally blameworthy when we violate the moral ought in favour of the epistemic. But that is to say that the relevant evaluations are of a different sort, and I address this in section 5. For now we are assuming that there are not different kinds of blame or blameworthiness. As such, the adherent of NDOS has to deny that there are can be genuine doxastic dilemmas across domains.

With these two claims on board we are now in a position to see that NDOS is incompatible with Epistemic Ought Deontologism. Consider again the scenario where S ought_epistemic to believe p, but ought_moral to suspend judgment on whether p. As we have just discussed, the adherent of No Doxastic Ought Simpliciter holds that this is not a genuine dilemma (since there is no standpoint from which we can say that the subject is blameworthy no matter what), yet does not deny that the relevant oughts are genuine (though they are incommensurable). But the problem is that to say that there is no genuine dilemma here is to say that S can justifiably (cashed out in terms of whether blamelessly in EOD) evade the demands of either ought - one can pick either one - (without excuse). That is, if there is no dilemma, then S is not in a situation where S is blameworthy come what may: so she would not be blameworthy if she does what she ought_moral to do and suspends judgement on whether p, despite her having violated her epistemic requirement to believe that p. So then we can have, as per impossible for Epistemic Ought Deontologism, a genuine ought bearing on belief that can be blamelessly violated without excuse.

One could now retreat to the view that these doxastic oughts have authority in a different way from how moral ought usually does – but this would clearly be incompatible with strong deontologism as we defined it in §1 . And, further, to think that doxastic oughts have authority in a different way makes the following question hard to answer in a non-ad hoc manner: why does moral doxastic ought behave differently from ordinary (non-doxastic) moral ought?

To sum up then: if the way to identify epistemic requirements is via appeal to a special epistemic standard or norm of appraisal then that standard is either commensurable with other standards, in which case No Doxastic Ought Simpliciter is false, or it is not so commensurable. If it is not so commensurable, then cases when ought_epistemic clashes with another kind of ought (or other kinds of oughts) constitute genuine dilemmas or they do not. If they do, No Doxastic Ought Simpliciter is false. If they do not, No Doxastic Ought Simpliciter is false.

3.2. Sub-Argument (ii)

The following I think is the most obvious, natural sub-argument of the three that I will discuss. Unfortunately, I think it is also the weakest, since it relies on a controversial premiss. I discuss it mostly to show that it is, appearances to the contrary, compatible with the previous sub-argument. This is important to show, given what I have said about its being the most natural of the sub-arguments I am discussing.

Here is the sub-argument:

1. 6. If we individuate epistemic ought solely in terms of its relevant standard of appraisal then there is only one attitude of blame (but there can be different kinds of ground for holding it, incommensurable with each other).

2. 7. If 6, it is possible for S to be normatively required to hold conflicting attitudes at t.

3. 8. Principle NCA: S cannot be deontologically required to hold conflicting attitudes at t.

4. 9. not-6. (from 6, 7, 8).⁶

Premiss 6 should not be at all controversial in this context, since it is fairly clearly entailed by 1(b): if there was a distinctive kind of epistemic blame, the standard of appraisal would not be the only thing that could individuate epistemic ought (whether the object of evaluation is a candidate for epistemic blame would also identify it). So let's move to premiss 7. Take the following contrast:

1. [α] at t₁ I blame S for believing that p and at t₂ I do not;

2. [β] at t₁ I blame and don't blame S for believing that p.

The situation we're invited to consider in premiss 6 is one where when I consider the epistemic reasons, I think S is blameworthy for believing that p (so blame S), and when I consider the moral reasons, I think S is blameless for believing that p (so do not blame) S. Is this more like [α] or [β] above? Clearly it is more like [α], since having considered both the epistemic reasons and the moral reasons I can end up both blaming and not blaming S at t₁, and be normatively required to do so by the two kinds of ground for blame.

To defend premiss 8, we need to ask the question: why hold Principle NCA? The best support is likely to be found in some further ought-implies-can principle – versions of the principle can be controversial, so I won't belabour the point here. As I mentioned, the important issue here is whether this sub-argument is compatible with the previous sub-argument.⁷ It is true that one of the premises of previous argument is that if there can be genuine normative doxastic dilemmas then there are doxastic oughts simpliciter. But the argument was not that there then must be doxastic oughts simpliciter because there are genuine normative doxastic dilemmas. Rather, the argument was that on all ways of answering the question of whether certain cases constitute genuine normative dilemmas one commits oneself to the claim that there are doxastic oughts simpliciter. If sub-argument (ii) is sound, then we are just restricted as to what counts as a viable answer to the question; it does not leave a scratch on whether all answers commit you to accepting that there are things you all things considered ought to believe.

3.3. Sub-argument (iii)

My third sub-argument is a kind of tu quoque argument, aimed at showing that – by parity of reasoning, and given what we already are naturally disposed to agree with – we should accept 3.

To start with, suppose we all agree that epistemic ought is characterised by a standard of assessment that involves considerations that are indications that p is true.

And suppose also that we also agree that predictive accuracy and explanatory power are both indications of truth: so both considerations are considerations relevant to epistemic ought.

If we suppose the previous two things, I think we are naturally disposed to agree with the following: If we thought that both considerations (predictive accuracy and explanatory power) are not relevant to the question [what should I believe all things epistemic considered?] then we wouldn't understand them as epistemic considerations relevant to what we ought to believe – we wouldn't agree that they are epistemic reasons (when by ‘epistemic reasons’ we mean considerations that are relevant to epistemic ought).⁸

Similarly, suppose we agree that some version of moral consequentialism is true: that we have moral reason to φ iff φ-ing promotes the good.

Suppose we also agree that being kind promotes the good, and that being just promotes the good: both considerations provide moral reasons. There are kindness-related moral reasons and justice-related moral reasons. So, if we didn't think that considerations relevant to whether the kindness-related and justice-related reasons were relevant to the question [what should I do all things moral considered?] we would fail to consider them to be moral reasons. We would disagree that they are both moral reasons after all.

Now, here is a general formula aimed at capturing this: if we agree on what x-type reasons are, and we agree that reasons relative to y-style valuation are x-type, then we think that they are relevant to whether one should φ, all x-type reasons considered.

If the formula applies within any domain, then it should apply in a ‘type-free’ domain.⁹ That is, the following should be true, via parity of reasoning:

If we thought that no considerations are relevant to the question [what should I believe all things [no type] considered?] then we would take no considerations to be relevant to what we ought to believe. We wouldn't agree that there are oughts to believe.

Put differently, to say that whether S epistemic ought to believe that p is not at all relevant to whether S all things considered ought to believe that p is to deny that epistemic ought is a genuine ought. Thus, given what we already think about what makes reasons of a certain kind be genuinely normative, we should, by parity of reasoning, admit that there are all things considered reasons to believe, else we should abandon the idea that there are any oughts to believe.

So if we think that there is no all things considered doxastic ought, we cannot think that what makes epistemic evaluation epistemic is solely the standard by which we evaluate. We must think instead that there is a special kind of epistemic evaluation. I thus now turn to evaluating premise 4.

4. Epistemic oughts are epistemic because they involve a unique kind of evaluation/blame

A deontic epistemic evaluation with respect to a belief that p involves the assessment that a reactive attitude (typically blame and resentment) would be appropriate in response to a belief that p. So one might think that epistemic ought is distinctive, because the relevant deontic epistemic evaluation (evaluation of whether an ought has been violated) involves a distinctive kind of epistemic blame and resentment.

For someone who denies there is doxastic ought simpliciter, the difficulty is to give an account of epistemic blame that is distinct from (say) moral blame but yet belongs to the same genus – and do so without solely invoking a differential norm of assessment (as I showed in section 4, this leads one to accept that there is doxastic ought simpliciter).¹⁰ To be clear, this is not to demand that the explanation as to why epistemic requirements are epistemic be uniquely to do with involving a distinctive kind of blame (such that the explanations in 1(b) and 1(c) cannot be combined), but rather the demand that the explanation in 1(c) is not just the same explanation as 1(b).

The most obvious thing to appeal to (it seems to me) are functionalist accounts of blame. Here people may appeal to the typical functionalist idea of multiple realisability: lots of things might qualify as blame/resentment, so long as they perform the same function. So epistemic blame might have a completely different character to moral blame (involve a distinct sort of resentment perhaps), but yet nonetheless be a kind of blame in that it performs the right function.

What then is the right function? Here are a couple of candidate aims for blame.

BlameAim₁ [To repair relationships after wrongdoing and so keep communities together] (see MacNamara 2013).

BlameAim₂ [To increase the alignment of the blamer and the wrongdoer's moral understandings] (see Fricker 2016).

It is hard, however, to see how epistemic blame can perform the function described in MoralBlameAim₁. Do we really wrong when we fail to be an epistemological externalist, for example? Contrary to how certain externalists might make one feel, I do not think they are being wronged in any substantial sense of the word, even if epistemic externalism is true.¹¹ MoralBlameAim₂ seems to make the aim of blame be a kind of moral aim, and as such the aim of epistemic blame would just be a special case of moral blame. And so not the kind of sui generis thing our opponent would be after. Perhaps then we need to try to formulate a distinctive, but nonetheless similar aim for epistemic blame? Here is a first stab at formulating such an aim:

EpistemicBlameAim₁: [To increase the alignment of the blamer and the wrongdoer's epistemic understandings].

But what are epistemic understandings? Understanding already seems to be an epistemic notion, and as such we do not need to say ‘epistemic’ twice (just as it is odd to talk of legal laws). So, we might then have:

EpistemicBlameAim₂ [To increase the alignment of blamer and the wrongdoer's understandings].

But does not EpistemicBlameAim₂ look like an account of blame simpliciter? I think it does, since, on this account, the only thing that would make moral blame be moral blame would be that the considerations the blamer has taken into account are restricted to the moral domain. In other words, we are back to appealing solely to a standard of appraisal and not really invoking a distinctive kind of blame. So we might then think that epistemic blame has its own unique aim, independent from the aim of other kinds of blame:

EpistemicBlameAim₃ [To increase people's compliance with epistemic norms].

But if epistemic blame has its own independent aim, then it is hard to see why it is a species of blame. The fact that it fulfils some function or other surely is not enough to qualify it as a kind of blame. Perhaps though it has a similar function (or the same kind of function)? For example, its aim is similar to the function of moral blame in that it gets us to comply with norms we ought to follow –we ought to follow moral norms, and we ought to follow epistemic norms. But if that is true, then there is after all an overall aim of blame: to get us to comply with norms that we ought to follow. And if epistemic blame has the same overall aim as other kinds of blame, then it is hard to see what makes it epistemic without appeal solely to the standard of appraisal (and as we have seen this entails there is doxastic ought simpliciter; assuming epistemic ought deontologism). The central underlying worry here is that without some unifying account of what blame involves, epistemic ought deontologism looks vacuous.

So again we are faced with a dilemma:

Either there is an aim that epistemic blame shares with other kinds of blame such that they are members of the same genus or there is not. If there is, then there is doxastic ought simpliciter as we would have to identify the epistemic as in 1(b). If there is not, then there is no doxastic deontological ought.¹²

Brown (2020) has recently defended an account of epistemic blame. In her analysis one blames S when one has a particular belief/desire pair (as per Sher 2009). One morally blames S when one believes that S has acted morally badly, and desires that S had not so acted; one epistemically blames S when one believes that that S has believed badly, and desires that S had not so believed. Note that the account of epistemic blame does not involve one's believing that S has believed epistemically badly, since this would not give us a non-circular account whereby we can individuate the epistemic. But this means the account pinpoints the difference between moral and epistemic blame on the object of the assessment, and nothing else. This brings us back to premiss 2, and the conclusion that there is doxastic ought simpliciter.

This then completes my defence of premisses 2, 3 and 4. I have taken premiss 1 to be obviously true. It would be false if there was some other way than described in 1(a), 1(b) and 1(c) for our identifying why epistemic ought is epistemic ought, and it seems logically impossible (given that epistemic ought will involve an assessment of some sort) for there to be any other way. I have treated the explanations in 1(a), 1(b) and 1(c) disjunctively, however, and one could ask whether some combination of them can save the day. Unfortunately, given what I have argued, it looks like if any account invokes 1(a) as an explanation, even as part of a conjunction, then there is doxastic ought simpliciter. The trouble with the explanation in 1(c) is that it collapses either to the explanation in 1(a) or 1(b) so that it cannot offer us really any further explanation, either alone or in combination. We are then left with 1(b), which I argued entails that there is doxastic ought simpliciter. I now move to discuss why I think it matters that there is such a thing.

5. Why does it matter that there is doxastic ought simpliciter?

I think the answer to the above question has to do with how the issue bears on the issue of justified belief. What we have been discussing raises the following question: if there are all things considered doxastic oughts, and there are differing kinds of (pro tanto) doxastic oughts, what kind of ought is relevant to the justification of belief? Here is an extremely simple argument for the claim that it is our all things considered oughts that determine the justification of belief (formulated by Booth 2012):

1. 10. S is justified in believing that p iff S is blameless in believing that p. [deontologism]

2. 11. S is blameless in believing that p iff S has not violated a (non-excused) all things considered ought in believing that p.

3. 12. Therefore, S is justified in believing that p iff S has not violated a (non-excused) all things considered ought in believing that p (see Booth 2012: 510).

In other words, there aren't any epistemic oughts for which we are blameworthy. Since Epistemic Ought Deontologism understands oughts in terms of blameworthiness, the thesis is unstable. That would be quite a significant result, it seems to me, so the argument merits some consideration. The crucial premiss in the above argument is clearly 11, assuming Deontologism. Is it defensible? I think two points help make it so, at least prima facie:

First, if there are doxastic oughts simpliciter (all things considered) then epistemic oughts must be merely pro tanto [in determining that S ought_epistemic to believe that p, one has merely considered epistemically relevant factors, one has not considered everything].

Second, it is widely held that one can be excused for violating an ought (e.g. when one is non-culpably ignorant that it applies) and justified for violating an ought when a competing ought trumps the ought in question (e.g. I ought to teach my classes to my students, but the ought relative to whether I should strike is greater, so trumps the first).

To say, then, that S ought to believe that p all things considered, is to say that S can only be excused for violating that ought – S cannot justifiably violate it.

So, to say that S pro tanto ought to believe that p, is to (at least) say that we do not know whether S would be justified in violating that ought. As such, blame is inappropriate: we need to know not only whether there is a relevant excuse but also a relevant justification. Here is a nice articulation of this thought by Scott Stapleford:

When deciding what to do, we need to consider everything. What remains after considering everything is the weightiest obligation. The weightiest obligation is all that counts, since it outweighs every other consideration. And the weightiest obligation is generic (rather than, say, epistemic), since it's based on more than one kind of consideration. (Stapleford 2015: 1863)

Nonetheless, 11 seems problematic. Here is a correlatively simple argument as to why (where by ‘normative dilemmas’, I mean cases where S ought to believe that p and suspend judgement as to whether p):

1. 13. If 11, there cannot be normative dilemmas.

2. 14. There can be normative dilemmas.

3. 15. Not-11.

But 13 is false. There may be two equally weighted, but conflicting oughts all things considered. Since neither ought can be thus overridden, one cannot blamelessly violate it without excuse. Thus, it may be the case that S ought all things considered to believe that p and S ought all things considered to suspend judgement on p: aka a normative dilemma. As we mentioned in section 3, to think there are all things considered oughts is compatible with the claim that there are normative dilemmas. One could question how there could be two equally weighted, but conflicting oughts all things considered, by invoking principle NCA we previously discussed. But that amounts to denying premiss 14, so could not figure in the argument denoted by 13, 14, 15.¹³

So here is another attempt at resisting premiss 11 (which I owe to Scott Stapleford):

1. 16. If 11, there cannot be different grounds for blame.

2. 17. There can be different grounds for blame.

3. 18. Not-11.

We can easily generate counterexamples to the principle [premiss 11], if we only recognize that there are different grounds of blame. Precisely because there are different grounds – stemming from different sources of obligation – we can be simultaneously blameworthy for ϕ-ing on x-grounds and blameless on y-grounds even when the y-grounds outweigh the x-grounds. I am to blame on legal grounds for driving in the wrong lane to evade the ducks – a pedantic judge could impose a fine – though my moral obligation to these creatures outweighed legal, prudential and all other considerations. All things considered, my best bet was to swerve and save the ducks, but I am still legally culpable. (Stapleford 2015: 1873)

This looks like a much more promising argument. Unfortunately, it too, does not succeed, since if there is all things considered ought, you are justified in violating the legal requirement above, so blameless (that is just what it is on a deontic understanding of justification to violate a deontic requirement). To insist that one is still legally blameless is thus to deny that there is here a fact of the matter with respect to what one ought all things considered to do. This is because this last claim is equivalent to: one cannot be justified in violating a legal consideration without an excuse. That is, that other considerations cannot compete with considerations relevant to legal ought. But we just showed that – for ought to believe – we must reject the thesis we called No Doxastic Ought Simpliciter or else give up Deontologism.

6. Conclusion

In conclusion, then, Epistemic Ought Deontologism is unstable – only our all things considered doxastic duties are relevant to the question of what we ought to believe. However, we should be careful to note that this does not mean that we cannot ask (and answer) that question – it is just a question that has independent standing, and outruns concerns in epistemology.

I have not addressed Feldman's worry mentioned in section 1, that:

If there is such thing as “just plain ought” then there is a value associated with it. The thing we just plain ought to do is the thing that comes out highest, or high enough, according to that measure. It's far from clear what that value would be – it isn't to be identified with any of the more determinate kinds of value and there seems to be no uniquely correct (or range of correct) ways to combine moral, practical, epistemic, and other values. (Feldman 2000: 664)

I think, however, it is enough to parry the objection to deny the assumption (since it is merely asserted by Feldman) that if there is such a thing as just plain ought, there is a value associated with it. This seems to, broadly, assume a type of realism about the source of normativity, according to which all values are things out there in the world that determine what one ought to do and believe. And many ethicists (who work on precisely the issue of what all things considered oughts are) reject this account – most notably Kantian constructivists, I think nicely illustrated by Chang's (2017) account of hard choices where what determines what one ought to do when confronted with a choice between two courses of action that are ‘on a par’ is not any mind-independent normative fact. So, if I am right about the instability of Epistemic Ought Deontologism, then Feldman's concern might give us a further restriction on what might be called our ‘meta-epistemology’: we should not be realists so construed.¹⁴