1. NORMATIVE DEFEAT AND EVIDENCE YOU SHOULD HAVE HAD

Although it is by no means an uncontroversial matter, many epistemologists hold that evidence which you don't have, but which you ought to have, can defeat your justification for believing that p.Footnote ¹ Kornblith (Reference Kornblith1983) famously offered the case of the physicist who, owing to a personality defect in which he regularly tunes out his colleagues’ comments, fails to attend to what they are saying when they tell him of important counterevidence to his pet hypothesis. Kornblith argued that even if the evidence which the physicist had in his possession would have justified his belief in the pet hypothesis, that belief is not justified, since there is evidence he should have had – namely, the evidence of his colleagues’ say-so – which, if he had acquired that evidence, would have rendered his belief unjustified.

Kornblith offered the case as challenging evidentialism – the view that one's belief is justified when it fits the evidence one has in one's possession. Conee and Feldman (the leading proponents of evidentialism) responded by simply biting the bullet: they embraced the verdict that the physicist's belief in his pet hypothesis is epistemically justified after all, arguing that it was a moral or professional failure, rather than an epistemic one, that he failed to have the counterevidence on offer.Footnote ² But it is worth noting that not all evidentialists have followed suit; some want to grant Kornblith's ‘not justified’ verdict while trying to preserve the spirit of evidentialism itself. Thus Christopher Cloos has recently proposed to modify evidentialism so as to take into account not only all of the evidence one does have but also all of the evidence one should have, where the latter is determined by the evidence that one would have had if one had been epistemically responsible (Cloos Reference Cloos2015).Footnote ³ Whether or not this succeeds at saving evidentialism from Kornblith's objection, the point remains that Cloos feels the pull of the idea that evidence one should have had can render one's belief unjustified.

Epistemologists such as Kornblith and Cloos are not the only ones to embrace such a view. Recently, I have argued that epistemic assessment ought to take into account all of the epistemic expectations that are properly had of a subject in a given situation, so that if it is properly expected of S that she have certain sorts of evidence bearing on the matter whether p, then that evidence is relevant to an assessment of S's epistemic standing on the matter.Footnote ⁴ What is properly expected of one is a matter about which more theorizing needs to be done; but at a minimum we can say that legitimate social practices give rise to expectations that are proper. (See Goldberg Reference GoldbergForthcoming for details.) This is a view that appears to be endorsed by others as well.Footnote ⁵

In short, there is an emerging group – which is not to say a majority! – of epistemologists who regard evidence one should have had as potentially defeating the justification enjoyed by one's belief.Footnote ⁶ In this paper I will not be defending this hypothesis. Rather, I will be assuming that it is correct, with the aim of extending our reflections on various sorts of case in which the issue of evidence one should have had becomes live. For those who reject this hypothesis, I suggest that this paper should be read as aiming to establish a conditional thesis: if one accepts the possibility of such defeaters, the view to be proposed and developed below is the best account of these defeaters. To develop this account, I will be exploring cases illustrating two issues that have been underexplored in the relevant literature. One issue is how evidence one should have had relates to the evidence one does have, and in particular to the higher-order evidence one has regarding the likely existence and epistemic bearing of the evidence one should have.Footnote ⁷ For example, what happens when there is a discrepancy or clash between the evidence one should have had and the higher-order evidence one does have? I am also interested in cases in which the evidence one should have had would only have further supported one's belief that p. To the best of my knowledge, neither of these sorts of case have been discussed at any length in the literature. The issue of interest to me is how we should think about these cases, assuming that we come at them already holding that evidence one should have can defeat one's justification for belief. What are the principles that determine one's degree of justification in these sorts of cases?

It is important to be clear about the notion of justification I have in mind here.Footnote ⁸ Although I will be speaking of degrees of justification as well as justification simpliciter, the sort of justification I intend is doxastic justification, understood to be the property which renders a true unGettiered belief knowledge.Footnote ⁹ The talk of degrees is meant to capture our ordinary talk of being more or less justified, of something's being better or worse evidence for p, and so forth. I assume that it makes sense to have a graded notion of justification in a framework that regards justification as a necessary condition on knowledge: the satisfaction of this condition presumably will be a matter of having a high enough degree of justification. (I cannot pretend to have a theory that makes sense of all of this; but in this respect I am not alone, as anyone who thinks that justification is a necessary condition on knowledge, and who thinks that one can have more or less of it, is in the same boat.) My question is this: in contexts involving evidence one should have had, what are the principles that ought to guide how we think about how evidence bears on justification in this sense?

In addressing this question, I will explore particular cases that vary along one or another dimension of the relevant factors, and I will aim to identify the considerations that might incline us one way or another as we assess justification in these cases. In doing so, I will be approaching the cases from the perspective of one who antecedently accepts that evidence one should have had can defeat one's justification for believing that p. The idea will be to try to honor the motivation for such a view, while simultaneously arriving at an account that can do justice to the host of considerations in the more complicated cases. Although I am explicitly building the principles from the cases in a bottom-up fashion, I will conclude with some remarks about the principles themselves, aiming to identify a theoretical motivation for them that is independent of their ability to get the cases right.

2. A TAXONOMY OF CASES

Before moving on to the cases themselves, it will help to begin with some terms. Regarding subject S who believes that p on the basis of evidence E, let us suppose that there is further relevant evidence that S should have had – call this evidence E_SHH. (This is to say that E_SHH is not part of E.) Then we can characterize the epistemic bearing of E_SHH (with respect to S's belief that p) as either helpful, harmful, or neutral, according to which of the following conditions holds:

Further, we can speak of the special case in which E_SHH is (not merely harmful but) Harmful (with a capital ‘H’) if, in addition to HARM being true, the following condition holds as well:

Finally, let us distinguish between that part of S's evidence E which is first-order evidence bearing on whether p (call this ‘E_FO’), and that part of E (call this ‘E_HO’) which is higher-order evidence bearing on (i) how likely it is that there is further evidence S herself ought to have (which S does not currently have), and (ii) the likely epistemic bearing of that evidence on S's belief that p. Then we can describe the epistemic bearing of E_HO on S's belief that p as either happy, unhappy, neither, or null according to which of the following is true:

Given all of this, any case in which there is further relevant evidence S ought to have had (with respect to S's belief that p) can be categorized into one of twelve different cases, according to variations in the two dimensions above: the epistemic bearing of the evidence S ought to have had; and the epistemic bearing of the higher-order evidence S did have regarding the evidence S ought to have had. (In speaking of twelve cases rather than fifteen, I am opting to consider the more generic HARM rather than the case involving HARM together with ~J. This ensures that my taxonomy is exhaustive.) The aim is to offer a theory that enables us to understand the epistemic significance both of the evidence one ought to have had, as well as of the higher-order evidence one does have regarding the evidence one ought to have had.

In advance of trying to arrive at such a theory, it will be helpful to have brief descriptions of each of the twelve cases. For ease of reference, it will be best to describe them in terms of the epistemic properties of the evidence S should have had, E_SHH, and the higher-order evidence S does have, E_HO, as follows: when E_SHH is helpful, I will designate the situation as ‘E_SHH+’, when E_SHH is harmful I will designate it as ‘E_SHH-’, when E_SHH is neutral I will designate it as ‘E_SHH0’; and similarly for E_HO, so that when E_HO is happy I will designate this as ‘E_HO+’, when unhappy I will designate it as ‘E_HO−’, when neither I will designate it as ‘E_HO0’, and when there is no such evidence I will designate it as ‘E_HOϕ’. (This will make it easy to refer to the various cases in ways that will be easy to comprehend.) The following then constitute the twelve cases to be considered.

These are the cases, then.

In what follows I begin with the epistemic significance of S's higher-order evidence, and then move on to the epistemic significance of the evidence S should have had. After reflecting on these, I try to distill principles which cover both of them, and their interaction.

3. THE EPISTEMIC SIGNIFICANCE OF HIGHER-ORDER EVIDENCE

What is the epistemic significance of S's higher-order evidence?

Let's start with cases where I hope there will be widespread agreement. Consider then the three (E_HO−) cases, in which S has higher-order evidence which is unhappy: (E_HO−)/(E_SHH+), (E_HO−)/(E_SHH−), and (E_HO−)/(E_SHH0). In each of these cases there is a strong intuitive pull to the verdict that, at a minimum, the justification of S's belief that p is diminished in proportion to the unhappiness of her higher-order evidence. For the following seems plausible: if S has higher-order evidence to think both that there is further evidence she should have, and that this evidence will bear against her belief that p, then to just that extent she ought to question her entitlement to believe that p.Footnote ¹¹ This verdict is not merely plausible but inescapable if the higher-order evidence she has suggests that the further evidence she ought to have is not merely harmful but Harmful. For in that case she has evidence to think that there is evidence she should have which, if she had it, would render her belief unjustified. Since this higher-order evidence is part of her evidence, it renders her belief unjustified. Now consider the case in which her higher-order evidence suggests that the further evidence she ought to have is harmful (with a lower-case ‘h’). Here, her higher-order evidence gives her a reason to think that her belief that p is less justified than it would have been had she not had any reason to think that there was such evidence. Admittedly, it is unclear that this renders her belief unjustified full-stop. For she might have reasons to think that, while the evidence she should have had would render her belief slightly less justified, it would not do major epistemological damage. (Suppose S finds out that there has been new research which renders the scientific community only slightly less warranted in its confidence in p than it had been beforehand, where the community's degree of warranted confidence in p is still high.) In that case we might think that the scenario does not yet warrant a ‘not justified’ verdict regarding S's belief that p. Nor is it entirely clear what to say in cases in which the only thing S has reason to believe is that there is harmful evidence she should have had: if S has no reasons to determine whether E_SHH is merely harmful or Harmful, it is not clear whether, short of considering the actual epistemic bearing of E_SHH on S's belief that p, S's belief that p is unjustified. But even without reaching a final verdict on this case (which I will do below), we can say that at a minimum her justification is diminished in proportion to the unhappiness of the higher-order evidence.

In short, at least when it is unhappy, higher-order evidence is, by itself, epistemically significant, in that it decreases the justification of one's belief that p in a manner that is independent of the epistemic bearing of the evidence one should have had. The proof of this is simply that the considerations just presented (in defense of this conclusion) enabled us to reach this result without having to consider the (E_SHH+), (E_SHH−), or (E_SHH0) variations of the (E_HO−) cases. In particular, even if the evidence S should have had would only have been helpful, even so, if S had (what would then be misleading) higher-order evidence for thinking that the evidence in question would have been harmful, her justification for believing that p would have decreased accordingly. (And the same goes for the case in which E_SHH has a neutral bearing.) This is what I mean by saying that unhappy higher order evidence is epistemically significant in a manner that is independent of the epistemic bearing of the evidence one should have had: the former decreases justification in a way that is unaffected by the happiness or the neutrality of the evidence one should have had.

The dynamic is not entirely straightforward, however, as is seen in the cases where a verdict was hard to determine. Here I note that for all I said above in connection with higher-order evidence, it may be that (when it is harmful) evidence one should have had can further diminish S's justification for believing p, beyond the diminishment determined by S's (unhappy) higher-order evidence: as when, for example, E_HO suggested that E_SHH was harmful, but where in fact E_SHH was (not merely harmful but) Harmful. This remains to be seen. Still, even at this point we can capture the epistemic significance of higher-order evidence as follows:

Given (*), whatever epistemic bearing of the evidence S should have had – whether it is helpful, harmful, or neutral – S's total evidence, which here includes unhappy higher-order evidence, places a maximum on the degree of justification of S's belief that p: it can be no greater than it would be if all she had to go on were E (= E_FO + E_HO) itself.

But what should be said about the cases in which S's higher-order evidence is either happy or neither? While it is hard in these cases to say whether S's belief that p is justified until we discuss the epistemic significance of the evidence S should have had – something I will do in §4, below – still, it seems that a generalization of (*) holds for these cases as well. This is for a simple, easily appreciable reason. (*) states a limit on the highest degree of justification there can be for S's belief that p: it can be no higher than S's total evidence, including her higher-order evidence, allows. This much seems true whatever the epistemic bearing of the evidence she should have had. To deny this is to hold that evidence one should have had can enhance one's justification for believing that p; and this seems crazy.Footnote ¹² Rejecting that crazy idea, we can embrace a generalization of (*) as capturing the epistemic significance of one's higher-order evidence, as follows:

(**) is a generalization of (*), as (*) is the special case of (**) in which E contains higher-order evidence which is unhappy. But the general point is this: one's degree of justification cannot be higher than what is determined by the evidence in one's possession – including, when relevant, one's higher-order evidence. This suggests that whatever the epistemic significance of S's higher-order evidence is, it has this significance independent of the epistemic bearing of the evidence S should have had.

But what is the epistemic significance of that evidence – the evidence S should have had? I turn to this question next.

4. THE EPISTEMIC SIGNIFICANCE OF EVIDENCE YOU SHOULD HAVE HAD

To this end, consider the cases in which the evidence S should have had was helpful, the (E_SHH+) cases. There are four such cases to consider: (E_HO+)/(E_SHH+), (E_HO−)/(E_SHH+), (E_HO0)/(E_SHH+), and (E_HOϕ)/(E_SHH+). Since I have already considered (E_HO−)/(E_SHH+),Footnote ¹³ I will limit myself to discussing the other three cases.

We can begin with (E_HO+)/(E_SHH+). Here, while there is evidence S ought to have had, this evidence would only have further supported S's belief that p (the case is a HELPful one). What is more, S has higher-order evidence that justifies her belief to this very effect (the case is a HAPPY one). In this case the following verdict seems intuitively plausible: S's belief is justified despite her failure to have had all of the evidence she should have had. To be sure, S had reason to believe that there is further evidence she should have had. But why should S's belief that p be rendered unjustified merely because she didn't acquire that evidence (E_SHH)? After all, she already had good reasons to believe that the evidence in question would only be helpful, and she was right about this. It seems rather that, on the strength of both E_FO and E_HO, together with the de facto helpfulness of E_SHH, her belief that p is justified.

At the same time, it must be acknowledged that even in this case, there are some considerations that can be mustered on behalf of the argument for a ‘not justified’ verdict.Footnote ¹⁴ So for example it might be argued that the very facts (one) that there is evidence S ought to have had, (two) that S has reason to believe as much, and (three) that S still didn't acquire that evidence, taken together, should be regarded as constituting a defeater for whatever justification is otherwise enjoyed by S's belief that p. One who argued in this way would acknowledge that S was justified in her true belief that the relevant E_SHH would have been helpful. Still, the argument would run, S had higher-order evidence justifying the belief that there was relevant evidence she should have had, and so her failure to have it defeats the justification of her belief that p.

The issue in play here strikes me as one of the central questions facing any account of the epistemic significance of evidence one should have had: what is the epistemic significance of the fact that one is aware that there is such evidence? For those who think that S's belief is unjustified in the case of (E_HO+)/(E_SHH+) above, this awareness itself is ascribed a great epistemic significance, independent of the actual epistemic bearing of the evidence itself. Matters are otherwise for those who think that S's belief is justified in this case. For such theorists, neither the fact that there is such evidence, nor the fact that one has reasons to believe as much, defeats justification. For now all I will say is this: any adequate account of the relevant phenomena should address this matter directly. In particular, an adequate account must be able to explain, both why there is a strong temptation to endorse the ‘justified’ verdict in (E_HO+)/(E_SHH+), but also why even so there remains some pull to the ‘not justified’ verdict. I will return to this case below, after I have developed principles regarding the epistemic significance of evidence you should have had.

Even before getting there, however, we are in a position in which we can discern some patterns in the other two cases in which the evidence S ought to have had was helpful:

$$(\hbox{E}_{\rm HO}0)/(\hbox{E}_{\rm SHH}+)\, \hbox{and} \,(\hbox{E}_{\rm HO}{\rm \phi})/(\hbox{E}_{\rm SHH}+).$$

Consider (E_HO0)/(E_SHH+) first. This case resembles the (E_HO+)/(E_SHH+) case in crucial respects. To see this, consider that we might employ a similar line of reasoning in defense of a ‘justified’ verdict in this case, (E_HO0)/(E_SHH+), as was employed in (E_HO+)/(E_SHH+). (E_HO0)/(E_SHH+) is a case of the (E_HO0) sort. Consequently, either S had good reason to believe that there was no relevant E_SHH, or else her higher-order evidence either warranted no view about the likely epistemic bearing of that evidence on her belief that p, or it warranted the view that the bearing was (likely to be) neutral. Since S's first-order evidence (E_FO), by itself, would justify S's belief that p, we might then think that the fact that E_HO was neither happy nor unhappy, together with the fact that E_SHH was helpful, yield the verdict that S's belief that p is justified – and this, despite the fact that there was further evidence she ought to have had. Admittedly, the case for a ‘justified’ verdict in (E_HO0)/(E_SHH+) is perhaps not quite as strong as the case for the ‘justified’ verdict in (E_HO+)/(E_SHH+), but it is still strong. Next, consider what might be said in defense of a ‘not justified’ verdict in (E_HO0)/(E_SHH+). As in the (E_HO+)/(E_SHH+) case, so too here there is some temptation to think that the very fact that there is evidence S ought to have had should count as a defeater for S's belief that p – and this, despite the facts, first, that S has no reason to believe (and some reason to deny) that there was negative evidence she should have had,Footnote ¹⁵ and second, that in fact E_SHH would have been helpful to her belief that p. This is very much like what was said in the (E_HO+)/(E_SHH+) case, adjusted for the difference in higher-order evidence between the two cases. In short, (E_HO0)/(E_SHH+) exhibits modified versions of the patterns we saw in (E_HO+)/(E_SHH+).

What, then, of (E_HOϕ)/(E_SHH+)? This case is like (E_HO0)/(E_SHH+) except that in (E_HOϕ)/(E_SHH+) the subject S has no higher-order evidence bearing on the existence (or likely epistemic bearing) of relevant evidence she should have had. This case raises the interesting question regarding how there can be evidence one should have had, when one has no reason to believe that there is such evidence. Some might question whether this is even possible. However, in Goldberg (Reference GoldbergForthcoming) I argue that such a scenario is possible. My idea there was this: the claim that there is relevant evidence which the subject ‘should have had’ can be generated by legitimate social expectations, and there are cases in which what is socially expected of one is legitimate despite one's lack of knowledge of the expectation itself. (I described such cases as ones in which, while there was evidence one should have had, one was blameless in failing to have it.) Here I will simply assume that my analysis was correct. If so, cases of the (E_HOϕ)/(E_SHH+) variety pattern like cases of the (E_HO0)/(E_SHH+) variety, with the exception that the case for ‘justified’ seems a bit stronger in (E_HOϕ)/(E_SHH+) – if only because S had no basis on which to believe that there was evidence she should have had – and the case for ‘unjustified’ seems a bit weaker in (E_HOϕ)/(E_SHH+) – again, because S had no basis on which to believe that there was further evidence she should have had.

Summarizing, we see that, in cases in which the evidence one should have had is helpful, so long as S's higher-order evidence is not unhappy, there is an argument to be made in defense of a ‘justified’ verdict, but also some considerations that seem to support an ‘unjustified’ verdict. A full account should be able to reach a verdict, and yet still explain what is going on in such cases – why there is some temptation towards an ‘unjustified’ verdict, even as the ‘justified’ verdict seems stronger. The account I offer below, in §6, will purport to do just this.

Next, let us move from helpful cases, to cases in which the evidence S should have had was harmful – the (E_SHH−) cases. Once again, there are four such cases to consider: (E_HO+)/(E_SHH−), (E_HO−)/(E_SHH−), (E_HO0)/(E_SHH−), and (E_HOϕ)/(E_SHH−). As we do so we will need to mind the distinction between mere harmfulness and Harmfulness – between mere HARM and (HARM + ~J).

We have already briefly considered (E_HO−)/(E_SHH−). To review, I argued that in such a case S's belief that p enjoys a degree of justification that can be no higher than it is on S's total evidence (which includes unhappy higher-order evidence). I added that if S's total evidence (including her higher-order evidence) suggests that E_SHH is not merely harmful but Harmful, then S's belief that p is unjustified. What we now need to consider is the epistemic significance of the de facto harmfulness of the evidence S should have had. Here I want to defend the idea that

We can motivate this thought as follows: if your evidence justifies you in believing that there is further evidence you should have (relevant to your belief that p), and justifies you in believing as well that this further evidence is (likely to be) harmful, then you are answerable to the harmfulness of that evidence – with the result that your belief that p can be no more justified than it would have been had you acquired that evidence and updated in a rational fashion. (It might be less justified, but it cannot be more justified.) Such an analysis allows that if E_SHH was merely harmful but not Harmful, S's belief that p might continue to be justified, albeit less justified than it would have been had the only relevant evidence been E_FO. But if E_SHH was not merely harmful but Harmful, then it seems that we have a very strong case for thinking that S's belief that p was unjustified. For in that case, S had good reason to believe that there was harmful evidence she should have had, where in point of fact that evidence was not merely harmful but Harmful. In such a case, even if S's higher-order evidence were misleading, in the sense that it justified S in believing only that E_SHH was harmful (but not Harmful), even so, the fact that S had reason to believe in the existence of harmful E_SHH, together with the de facto Harmfulness of E_SHH, renders her belief that p unjustified. This would seem to be a near-paradigmatic example of the phenomenon of normative defeat, where the evidence one should have had defeats the justification of one's belief.

In this regard an interesting variant to consider is (E_HO+)/(E_SHH−). This case is interesting since it involves a clash between the evidence one should have had, and the evidence one has regarding (the likely epistemic bearing of) that evidence. In (E_HO+)/(E_SHH−), there is further evidence one should have had, this evidence is harmful, but one has (misleading) higher-order evidence to think that whatever evidence one should have had is likely to be helpful. To make the case vivid, imagine S, the member of the research team, is responsible for tracing recent developments in a certain subfield; that S has excellent reasons to think that, while there are new developments in the subfield, they would only further support her belief that p (a usually reliable source told her as much); that as a result S fails to collect the relevant evidence, retaining her belief that p; but it turns out that, in point of fact, the evidence she should have had is (not merely harmful but) Harmful. Now it seems to me that anyone who thinks that evidence you should have had can undermine or defeat your justification for believing that p, should regard this as such a case – and this, despite the fact that S had misleading higher-order evidence. If this verdict is correct, it gives us an interesting result: the evidence you should have had trumps the higher-order evidence you did have regarding the (existence and bearing of the) evidence you should have had. In that case we have support for a generalization of (+), to the following effect:

(++) is a generalization of (+): (+) is the special case of (++) where S's total evidence includes higher-order evidence that is unhappy.

Why should we accept (++)? An initial thought (to be supplemented below, at the end of §6) is this. If there is evidence you should have, and you fail to get that evidence, you open yourself up to being assessed with respect to that evidence should it turn out that the evidence is harmful. And, at least in those cases in which that evidence is harmful,Footnote ¹⁶ then even when you had higher-order evidence which supported you in thinking (mistakenly) that the evidence in question would be helpful with respect to your belief that p, this higher-order evidence does not provide your belief that p with any epistemic protection. At best your higher-order evidence here provides you with some cover when others blame you for not having gotten the evidence you should have had.Footnote ¹⁷ (Below I will argue that there is a deeper motivation for (++). But for now this will have to do.)

I want to say something similar in the two remaining cases of (E_SHH−), namely (E_HO0)/(E_SHH−) and (E_HOϕ)/(E_SHH−). My reasoning here is this. I have argued that the de facto HarmfulnessFootnote ¹⁸ of E_SHH is sufficient to render S's belief that p unjustified in (E_HO+)/(E_SHH−), where S's higher-order evidence justifies S's belief that E_SHH would be helpful. But (E_HO+)/(E_SHH−) is the (E_SHH−) case in which S's belief that p has the strongest claim on justifiedness (given the happiness of S's higher-order evidence). So if S's belief that p is not justified in that (E_SHH−) case – if the de facto Harmfulness of E_SHH trumps S's higher-order evidence in that case – then S's belief that p is not justified in any case in which E_SHH is Harmful. This reasoning seems patent in (E_HO0)/(E_SHH−). And the only wrinkle in (E_HOϕ)/(E_SHH−) is that in that case S has no higher-order evidence at all. But insofar as this is compatible with the existence of evidence S ought to have had, as I argued in Goldberg (Reference GoldbergForthcoming) and as I am assuming here, it would seem that this case patterns like (E_HO0)/(E_SHH−). The Harmfulness of E_SHH trumps, and so S's belief that p is unjustified. More generally, in each of these cases the degree of justification for S's belief that p diminishes in proportion to the harmfulness of the evidence S ought to have had.

I want to round out my discussion of the epistemic significance of evidence you should have had, by considering cases I have not yet considered: cases in which the epistemic bearing of the evidence S ought to have had is neutral. Now we have already considered the case in which S's higher-order evidence is unhappy: (E_HO−)/(E_SHH0). In that case, I argued, it is plausible to think that S's belief that p is downgraded in justification in proportion to how harmful she (with her higher-order evidence) has reason to think E_SHH is. If her total evidence E, including her higher-order evidence, suggests that E_SHH is merely harmful, her belief that p might remain justified (albeit less so than it was given only E_FO); but if her higher-order evidence suggests that E_SHH is Harmful, her belief that p is unjustified. Here, the de facto neutrality of E_SHH ensures that E_SHH has no further epistemic effect. It would seem, then, in cases in which the evidence she should have had has an epistemically neutral bearing, S's belief that p is justified to the degree determined by her total evidence E (= E_FO + E_HO). So in (E_HO+)/(E_SHH0), S's belief that p remains justified; in (E_HO0)/(E_SHH0), S's belief that p is justified if and only if S's first-order evidence (E_FO) renders it justified; and the same can be said in (E_HOϕ)/(E_SHH0). But even here we can discern the same temptation as we saw in previous cases: we might ascribe epistemic significance to the very fact that there is further evidence S should have had (and that S had reasons to believe as much). Until we have addressed this temptation (endorsing it or explaining it away), we cannot be confident in the proper verdicts for these cases. What remains unclear is whether it is the very fact that there is such evidence, one's reasons for believing that there is such evidence, or the epistemic bearing of that evidence, that carries the epistemic water (so to speak).

5. CEILING PRINCIPLES AND THE DOCTRINE OF EPISTEMIC STRICT LIABILITY

Before addressing this matter, it will be helpful first to state the generalizations that emerge from our reflections so far. I submit that the following two generalizations emerge. First, one's higher-order evidence regarding the evidence one ought to have had is itself epistemically significant. However, and second, one's higher-order evidence does not exhaust the epistemic significance of the evidence one should have had. On the contrary, when the evidence one should have had is harmful, one's justification for believing that p decreases accordingly, in proportion to the degree of harmfulness of the evidence one should have had – whatever one was justified in thinking regarding the evidence one should have had. In this way the epistemic significance of the higher-order evidence one has can be trumped by the harmfulness of the evidence one should have had.

With this in mind I submit that the combination of (**) and (++) captures the most general principle governing the interactions between higher-order evidence and the evidence you should have had. I call the result the Epistemic Ceiling Principle (ECP):

Now, if we assume that ECP is the only epistemic principle bearing on the epistemic significance of evidence you should have had – something I will defend in §6 – we can then go on to determine verdicts in the range of cases discussed above.

For example, return to the various cases in which there is a clash between one's higher-order evidence (E_HO), on the one hand, and the evidence one should have had (E_SHH), on the other: the former is misleading either as to the (likely) existence or epistemic bearing of the latter. ECP entails that in such cases, the degree of justification enjoyed by S's belief that p at t can be no higher than whichever of E or E_SHH Footnote ¹⁹ determines the lower degree of justification. So if ECP is the only relevant principle governing the epistemic significance of evidence you should have had, then whichever of E or E_SHH determines the lower degree of justification will determine the degree of justification enjoyed by S's belief that p.

With this in mind, let us consider the two types of clashing cases in turn, one in which E_HO represents the likely epistemic bearing of E_SHH as worse than E_SHH actually is, and the other in which E_HO represents the likely epistemic bearing of E_SHH as better than E_SHH actually is. In the former case, because E_HO represents the likely epistemic bearing of E_SHH as worse than E_SHH actually is, the result is that it is E (which contains E_HO), rather than E_SHH, which determines the lower degree of justification for S's belief that p. So, given the assumption that ECP is the only relevant principle governing the case, the verdict we reach – that the degree of justification enjoyed by S's belief that p is determined exclusively by E – accords with the traditional view that one's degree of justification is determined by one's total evidence. Next, consider the case in which E_HO represents the likely epistemic bearing of E_SHH as better than E_SHH actually is. In this case it is E_SHH (rather than E) which determines the lower degree of justification, in the sense that the degree of justification S's belief would have enjoyed had S acquired E_SHH (and updated rationally) is lower than the degree of justification on E alone. The verdict we reach in this case – that the degree of justification enjoyed by S's belief that p is determined by the degree it would have been had S acquired E_SHH (and updated rationally) – reflects the epistemic potency of the evidence you should have had. What can be seen by reflecting on these clashing cases, then, is that the fact that there is further evidence one should have had is never, by itself, of any epistemic significance. Rather, this significance is always carried, either by the evidence one has (including one's higher-order evidence), or by the epistemic bearing of the evidence one should have had. Consequently, any impression that we ought to ascribe epistemic significance to the fact that such evidence exists must be explained away – something I will aim to do below, in §6.

Next, consider the cases in which the evidence one should have had would only have offered further support to one's belief that p. ECP entails that in such cases (as in all cases), the degree of justification enjoyed by S's belief that p at t can be no higher than whichever of E or E_SHH determines the lower degree of justification, from which it follows trivially that this degree of justification can be no higher than what is determined by E. So if ECP is the only relevant principle governing the epistemic significance of evidence you should have had, then we can see that evidence one should have had can never enhance one's justification above the degree determined by one's total evidence itself. This accords with the traditional view that evidence not in one's possession cannot enhance one's degree of justification (for which see footnote 11).

Still, what remains to be seen is whether we can explain (or explain away) the battery of factors that pull us in different directions in the various cases. I believe that ECP itself equips us to do this as well. But before I offer my explanation, I want to draw attention to one of ECP's implications. According to ECP, when evidence you should have had is harmful, the degree of justification of your belief diminishes accordingly (in the sense that the belief's degree of justification can be no higher than what it would be if one had acquired that harmful evidence and updated rationally). This is so whatever higher-order evidence you have. In this way we can say that, by the lights of ECP itself, the evidence you should have had is to be regarded from a standard of what I will call Epistemic Strict Liability:

I now want to suggest that this doctrine of epistemic strict liability can enable us to explain the variety of factors that pull us in one direction or another in the entire range of cases I have been discussing. In particular, it can enable us to understand the temptation to ascribe epistemic significance to the very fact that there is further evidence one should have had – even as it reveals this temptation to be wrongheaded. If this is correct, it gives us some reason to think that ECP itself is the only principle we need when considering the epistemic significance of evidence you should have had.

6. APPLICATIONS

Return to the set of cases discussed in §3 in which I maintained that we are pulled in different directions. These were cases in which one's higher-order evidence is not unhappy (it is either happy or neither), and the evidence one should have had is not harmful (it is either helpful or neutral). In this range of cases, I suggested, we feel a pull in both directions. On the one hand, a ‘justified’ verdict seemed correct since, while S was aware that there was further evidence she should have had, she had good reason to believe (or no reason not to believe), what was true, that that evidence would not bear against her justification for believing that p. On the other hand, there was something to be said on behalf of an ‘unjustified’ verdict as well. After all, there was evidence she should have had, and she had reasons to believe as much. In the latter respect, there was some temptation to treat the very fact that there is evidence she should have had (or perhaps her reasons for thinking that there was such evidence) as itself epistemically significant – independent of the actual epistemic bearing of that evidence on her belief. Call this The Temptation.

Now above I appealed to ECP, together with the assumption that ECP was the only relevant principle governing the epistemic significance of evidence one should have had, to reach verdicts in such cases. If those verdicts were correct, The Temptation should be resisted. But it is one thing to have a theory that entails verdicts inconsistent with The Temptation; it is another to defend that theory in the face of The Temptation itself. In what follows I want to do so by arguing that ECP can account for the various factors pulling us in different directions in the range of cases I have been discussing. To the extent that this account is correct, ECP appears to be the only relevant principle we need; and, what is more, The Temptation itself can be diagnosed as an understandable, albeit wrong-headed, reaction to the various factors in play in these cases. To make out this diagnosis, I will be appealing to the doctrine of epistemic strict liability.

It will be helpful to have a concrete case in front of us, if only to make vivid the pull of The Temptation. Consider the following illustration:

In this case, the evidence Friend should have had – Jones’ current avowals of her continuing commitment to come to the party – would have reconfirmed what Friend already believed on good evidence.Footnote ²⁰ But easily it could have happened that the evidence would have been otherwise (as would have been the case had Jones changed her mind, as she almost did). In light of this, we can bring out the source of The Temptation by pointing out that there is some intuitiveness behind the ‘unjustified’ verdict. To get a sense of what might be the source of the intuitiveness of an ‘unjustified’ verdict, we can note this: since both the evidence Friend has and the evidence Friend should have had (by getting Jones’ current avowal) are supportive of Friend's belief that Jones is coming to the party, insofar as an unjustified verdict itself is warranted, this warrant must be seen as reflecting the epistemic significance of (Friend's awareness of) the fact that there is evidence Friend should have had. (There would appear to be no other way to account for the ‘unjustified’ verdict itself.) In this way we see that if the proponent of The Temptation can defend an ‘unjustified’ verdict in PARTY, she can thereby defend The Temptation itself.

What might be said in defense of an ‘unjustified’ verdict in PARTY? Well, Jones almost changed her mind, and had she done so, the evidence Friend should have had would then have been Harmful – yet Friend would have been none the wiser. This seems to suggest that in basing her belief only on E, and ignoring E_SHH, Friend's belief was (in some yet-to-be-identified sense) epistemically risky. On the assumption that the presence of epistemic risk in this sense is incompatible with justification,Footnote ²¹ we would then be in a position to conclude that this riskiness undermines justification. Further, if this is the proper analysis of this case, we might hope to be able to fashion a notion of epistemic risk on which this risk is present whenever there is further evidence that the subject should have had, or at least whenever she has reason to suppose as much – no matter how objectively likely it was that the evidence would have turned out harmful. If successful, the result would be The Temptation: one would then endorse an ‘unjustified’ verdict whenever there is further evidence one should have had (or at least whenever one has reason to suspect as much).

But this case for an ‘unjustified’ verdict in PARTY does not withstand scrutiny: while there is a sense in which Friend's belief (and the testimony she offered) were epistemically risky in PARTY, it is not a sense that undermines justification. Two points are relevant here.

I begin with a criticism of the key intuition driving the ‘unjustified’ verdict in PARTY. The argument on offer (in defense of an ‘unjustified’ verdict in PARTY) ascribes great importance to the following modal fact: easily the situation in PARTY could have been one in which the further evidence – the evidence Friend should have had – was Harmful (and the proposition in question, to the effect that Jones is coming to the party, was false). The analysis just offered regards this modal fact, together with the fact that Friend did not acquire the further evidence, as rendering Friend's belief ‘epistemically risky’ in a way that undermines justification. But the modal fact cannot carry the weight that the argument assigns to it.

For one thing, the modal fact would appear to favor an analysis on which Friend (remains justified in her belief but) fails to know that Jones is coming to the party. For if the modal fact establishes anything relevant to the epistemic riskiness of Friend's belief and her testimony, it is that Friend's belief/testimony is unsafe: easily could Friend have used the very same belief-forming method and yet formed a belief (testified to a proposition) that was false.Footnote ²² But it is standard to use considerations of safety, not to cast doubt as to the justification of the unsafe belief, but rather to challenge its status as knowledge. This is why we see safety considerations in connection with barn beliefs in Fake Barn County, or with Nogot's belief that Smith owns a Ford, etc. To be sure, this use of safety considerations (in Fake Barn County etc.) takes place on the assumption that the belief in question is justified – precisely what our imagined defender of the ‘unjustified’ verdict in PARTY means to deny. But given that Friend does have evidence for her belief/testimony that Jones is coming to the party – after all, Friend heard her last week say she was coming – it would seem that the burden is on the defender of the ‘unjustified’ verdict to say why this case isn't relevantly like the other cases, in which safety considerations establish a lack of knowledge rather than a lack of justification. In short, the appeal to the modal fact sits awkwardly in an attempt to analyze PARTY as a case in which Friend's belief lacks justification.

Nor is this the only problem facing those who would appeal to the modal fact to defend the ‘unjustified’ verdict in PARTY. To see this, suppose that Friend had gotten the further evidence in PARTY: she asked Jones, and Jones confirmed that she is coming. (Call this case, PARTY+.) In PARTY+ it seems patent that Friend's belief is justified. But notice that even in PARTY+ it remains true that easily Jones might have changed her mind (in which case the evidence in question would have been Harmful). That is, even in PARTY+ the modal fact in question remains in place. Since this is so, and since (by hypothesis) Friend's belief remains justified in PARTY+, we can draw the following conclusion: the ease with which evidence one should have had could have been harmful, by itself, does not affect justification. So – turning back to PARTY – if in PARTY Friend's belief is ‘epistemically risky’ in a sense that is incompatible with justification, this verdict must be supported by more than a mere appeal to the modal fact itself; that is, an ‘unjustified’ verdict in PARTY does not follow merely from an acknowledgement of the fact that easily the evidence Friend should have had could have been harmful. But if the modal fact itself does not affect your justification in PARTY, it is hard to see how the modal fact, together with Friend's failure to have gotten the evidence she should have had, affects Friend's justification in PARTY. (As we saw in the previous paragraph, at best it affects her knowledge.)

Still, there remains a strong intuition in PARTY that Friend did something less than ideal, and that as a result she subjected Mr. Partymaker to a risk – we might even call it an “epistemic” risk – to which Friend ought not to have subjected him. Can we explain what lies behind this intuition without calling into question Friend's justification for believing as she does? I think we can, and this is the second point I wish to make about the scenario in PARTY. In telling Mr. Partymaker that Jones was coming to the party, without first having checked with Jones (and so without having acquired the evidence Mr. Partymaker asked her to acquire), Friend subjected Mr. Partymaker to a risk, and in particular an epistemic risk, to which she ought not to have subjected him. The risk was epistemic in that Friend risked testifying to a proposition she was not justified in believing – as would have been the case if the evidence Friend should have had had turned out to have been harmful. Friend ought not to have risked this in testifying to Mr. Partymaker. The “ought” here is moral: assuming that Mr. Partymaker's request that Friend confirm with Jones was itself legitimate, Friend owed it to him morally to do as he requested, or else to indicate to him why she did not do so. (For example, she might have presented her old evidence in the very exchange in which she told him that Jones was coming.) She did neither. Now as it happened, Friend was right in what she told Mr. Partymaker, and so, had she collected that evidence, Friend would have been further supported in this; but still it was risky to continue believing as she did, and risky as well to assert as she did to Mr. Partymaker, given the (to him) salient possibility that Jones had changed her mind. And if Jones had changed her mind, then Friend's belief (and her testimony) would have been (both false and) unjustified. In failing to check up on this, Friend improperly subjected Mr. Partymaker to a risk of (falsity and) unjustifiedness. This is a moral failing of Friend's; and it is a moral failing with an epistemic content, since the failure involved her having subjected Mr. Partymaker to a heightened risk of (falsity and) unjustifiedness. Still, we must distinguish between the risk of unjustifiedness with unjustifiedness proper. Since in point of fact the epistemic risks did not materialize, Friend's moral failure is not at the same time one deserving of epistemic downgrade in the justification of her belief as well.

But how can we make sense of the idea that one can run the risk of unjustifiedness in belief without being unjustified in one's belief? To do so, we can advert to the notion of epistemic strict liability.

In the law, when effects are governed by a standard of strict liability, the violator is strictly liable for any harms that result to others, even if the violator had excellent reasons to think that no one would be harmed by the product. To illustrate, consider a company that makes a product which has the potential to harm people if it were to malfunction. Insofar as the possibility of harm was foreseeable, tort law may then impose a standard of strict liability on the company that manufactures it: if the product harms anyone, then the company is held liable – whatever the intentions or degree of care the company exhibited in manufacturing the product. Now it might happen that, despite its best efforts, the company failed to rule out a possible source of malfunction that might easily have arisen in a given case. Suppose though that, even so, the product doesn't malfunction, and so those using it are unharmed. Then the company isn't liable for any damages. Still, we might think that the company did not do all that it should have done to ensure the protection of those using its products. There was an unacceptable risk of harm, even though no harm transpired. This, I want to suggest, is the analogue of the PARTY case: there was no epistemic harm done, but in fact Friend did not do all that should have been done to protect Mr. Partymaker from the risk of epistemic harm. This is a moral failing of Friend's, but – lucky for Friend! – given how the case turned out the justification of Friend's belief doesn't suffer any corresponding epistemic downgrade.

I think that this sort of case offers a model for thinking about The Temptation more generally. As I see things, the temptation itself (to ascribe epistemic significance to the fact that there is evidence S should have had) can be diagnosed as deriving from a confusion. The confusion involves the conflation of (i) a moral claim with epistemic content for (ii) an epistemic claim. To wit: it is one thing to say, in the manner of (i), that, a speaker should not knowingly subject a hearer to a risk of epistemic unjustifiedness regarding whether p; but it is quite another to say, in the manner of (ii), that in such cases the speaker's own belief, as well as any “telling” based on this belief, is unjustified. The temptation to regard the very existence of evidence you should have had as itself epistemically significant confuses the former for the latter. The confusion is understandable, given that one might well have thought that to run a serious risk of unjustifiedness is itself an instance of unjustifiedness. But it is still a confusion: the risk of unjustifiedness need not be an instance of unjustifiedness; a moral claim with epistemic content, as in (i), is not (and does not imply) an epistemic claim, as in (ii).

However, at this point the defender of The Temptation might think to go on the offensive. Given that we already need to distinguish the moral dimension from the epistemological dimension in cases in which the evidence one should have had is neutral or helpful, as is seen in the discussion above, why not endorse this sort of analysis in cases in which that evidence was harmful as well? Such an analysis would have it that, while the evidence one should have had is always epistemically inert – it never has any epistemic significance on one's justification for believing that p – even so, a failure to have it can be a moral failing, something for which one is criticized. (This appears to be the view of Feldman Reference Feldman2000.)

In this paper I have been assuming that such a view is false.Footnote ²³ Still, we might want to know what motivates this assumption. In particular, what reasons are there for thinking that evidence you should have had can have more than a merely moral significance? I have two related reasons for thinking that such evidence can have an epistemic significance (in addition to its moral significance).

The first is simply that denying this – or holding that such evidence has only a moral significance – leads to false predictions. In particular, it predicts that the epistemic standing of S's belief that p is the same in the (E_HO+)/(E_SHH+) case as in the (E_HO+)/(E_SHH−) case. I regard this as a cost. Return to the case of PARTY. Do we really want to say that Friend's belief to the effect that Jones is coming to the party would have remained justified even if Jones herself had changed her mind after all – and this, despite the fact that Friend was aware of further evidence she ought to have had, where the evidence was such that if she had it, her belief would be unjustified? This seems highly implausible.Footnote ²⁴

But there is a second, more theoretical reason for resisting any analysis which would have us treat evidence one should have had as invariably epistemically inert. It has to do with the point of epistemic assessment in the first place. The sort of picture suggested by ECP can be broadly motivated by two claims. The first is that we exhibit a profound, systematic, and ineliminable reliance on others for what we know, or take ourselves to know, of the world. The second is that a good deal of this reliance is rationalized by the mutual expectations that we have, and that we are entitled to have,Footnote ²⁵ of one another in the various roles we play – sometimes personal, sometimes professional or institutional – as we go about sharing information with one another. The point of epistemic assessment, on this picture, is to capture a notion of belief-worthiness that (a) is attuned to the materials the subject has to go on, but also (b) is sensitive to what others are entitled to expect regarding the materials the subject has to go on. If you like, we might think of dimension (a) as capturing the evidence the subject has in her possession, and dimension (b) as capturing the evidence it would be proper to regard her as having. Legitimate expectations regarding the evidence another subject should have capture dimension (b): what we expect from the subject, given the relevant role(s) she is manifestly playing in the various social practices through which we share information.

The picture behind my account of evidence one should have had, then, is this. Because our epistemic reliance on others is so profound and systematic, and because this reliance is largely rationalized by the expectations we have (and which we are entitled to have) of other epistemic agents playing various roles in our knowledge community, other subjects are properly held accountable to these expectations when they are legitimate. In the first instance, this “holding accountable” is a species of practical/moral accountability: those who violate legitimate expectations are prima facie socially or morally responsible (and are socially or morally blameworthy unless excused). However, this social or moral failure opens up subjects to epistemic criticism as well. At a minimum, a subject who fails to have all of the evidence she should have had exposes her interlocutors to an additional risk of falsity and unreliability in the information she passes along. Whether this risk materializes depends on whether the evidence she should have had was harmful to the relevant belief. If it is, then the subject is not merely prima facie morally responsible, she suffers from an epistemic diminishment as well: her justification for the relevant belief diminishes in proportion to the degree of harmfulness of the evidence she should have had. This is so no matter the reasons she had to expect that this evidence would not have been harmful. This, in effect, is the doctrine of Epistemic Strict Liability; my suggestion here is that it can be motivated by appeal to the sorts of expectations we have of our fellow agents in a knowledge community.

7. CONCLUSION

In this paper I have tried to provide the outlines of an approach for thinking about evidence you should have had, in two sorts of cases that have not received much attention in the literature: cases in which there is a clash between one's higher-order evidence and the evidence one should have had; and cases in which the evidence one should have had would only have further supported one's belief. The approach I presented here derives from an “epistemic ceiling principle” composed of two subprinciples, to the effect that the degree of justification enjoyed by a subject's belief can be no greater than (i) what it is on her total evidence, and (ii) what it is when we add to her total evidence the evidence she should have had. On such a view, we can allow that there are cases in which, given the happiness of one's higher-order evidence and the helpful bearing of the further evidence one should have had, one's belief remains justified despite one's failure to have all of the evidence one should have had. At the same time, if we see this “ceiling principle” and the resulting doctrine of epistemic strict liability as deriving from the legitimate expectations we have of one another as fellows in a knowledge community, then we can also explain why there is some temptation even in such cases to downgrade the belief's justification. Still, if my account is correct, we should resist that temptation, as it conflates the (morally or socially problematic) increased risk of unjustifiedness with unjustifiedness itself.Footnote ²⁶