2. The impact of future evidence on justified neutrality

We often take our future epistemic position into account when we decide what to think right now. By tomorrow, I might have forgotten some of the information I just acquired about different cellphone plans, so it seems like I should decide right now which one is better. By contrast, the fact that tomorrow Consumer Reports will publish a guide to buying cellphone plans seems like a good reason to suspend judgment now about which plan to purchase. Yet, standard accounts of epistemic justification or rationality don't make room for these factors.Footnote ¹ As McGrath (Reference McGrath2021a, Reference McGrath2021b) points out, it is often thought that only epistemic factors are relevant to which doxastic attitude regarding some claim p is propositionally (ex ante) justified for an agent. Epistemic factors include, for example, how much the agent's evidence supports p, or whether they know or appreciate how much their evidence supports p. Epistemic factors don't include whether the agent will have better or worse evidence in the future, whether their cognitive acuity will be better or worse in the future, how likely it is that they could come to know that p if they inquired further, or how much they value finding out about whether p is true.Footnote ²

I will not discuss the merits of abandoning traditionalism here – for the purposes of this paper, I will side with McGrath against allowing non-epistemic factors to bear on epistemic justification. Instead, I will examine his “divide and conquer” strategy for accommodating our intuitions while preserving traditionalism and argue that we need to divide and conquer more than he assumes. To see how his proposal works, let's focus on the following case:

This case is intended to elicit the intuition that the fact that Fletcher knows she'll get better evidence tomorrow makes it justified for her to suspend judgment now about who did it. But as we just saw, a traditional account of justification can't deliver this verdict. McGrath proposes to account for cases like this by distinguishing between two ways of being neutral: the first way is to suspend judgment, and the second way is to be agnostic (McGrath Reference McGrath2021a).Footnote ³

Suspending judgment, on McGrath's view, is not a doxastic attitude. Rather, “to suspend judgment on a question is to put off belief-forming judgment, that is, to omit it because one aims to have it later (and not before), or when and only when certain conditions obtain (which one does not yet believe obtain).” The way in which one suspends judgment (i.e., the product) is usually by postponing the process of deliberating until later. Since suspending judgment in this sense is not a doxastic state, we can admit that non-epistemic factors can be relevant to whether or not suspending is appropriate for a given agent at a particular time.

By contrast, when an agent is agnostic, this means that they have a neutral doxastic attitude. McGrath proposes “‘being agnostic whether p’ to mean having an intermediate state of confidence concerning whether p, i.e., having states of confidence for p and ~p such that one's state of confidence for p is neither strong enough for belief that p nor low enough for disbelief.” Because being agnostic is a genuine doxastic attitude, the traditionalist view McGrath endorses limits the factors that can be relevant to whether it is justified to standard epistemic factors, such as the agent's evidence and their appreciation of its significance.

With this distinction in mind, we can explain what's going on in DNA evidence #1. Before receiving the test result, Fletcher is both agnostic about who the murderer is, and she is also suspending judgment about it. She is justified in being in both of these states, but for different reasons. Her current agnostic attitude is propositionally (and doxastically) justified by purely epistemic factors: her current evidence doesn't support high confidence that any particular suspect is the murderer. Her suspension of judgment is justified by non-epistemic factors: she is justified in waiting to make up her mind because she knows that it will be more fruitful to deliberate about who the murderer is once she learns about the results of the DNA test. Hence, by distinguishing two ways in which Fletcher is neutral regarding who the murderer is, we can both accommodate the intuitive relevance of the forthcoming test results and retain our commitment to a traditionalist view of what's relevant to the epistemic justification of our doxastic attitudes.

3. A problem case

The distinction between being agnostic and suspending judgment gives us an elegant way of accounting for our intuitions about DNA evidence #1 without adopting a revisionist view of justification. But it runs into trouble in a slight variation of the case:

Our intuition about this case seems to be the same as in the first version: Fletcher seems justified both in being agnostic now and in suspending judgment until she receives the test result.Footnote ⁴ But there is an important difference between #1 and #2: Fletcher's initial evidence (without the DNA test) is inconclusive regarding who the murderer is in #1. By contrast, it supports high confidence in the gardener being the murderer in #2, even though recognizing this would require some additional deliberation about the evidence.

This creates a problem for McGrath's traditionalist account, because it can't explain why Fletcher's initial agnostic attitude seems justified in #2. This is true regardless of whether we assume a standard evidentialist or a reliabilist view of justification. Each type of view implicitly or explicitly relies on a total evidence requirement, which means that an agnostic attitude/middling credence toward some claim p can't be propositionally or doxastically justified for an agent if their total evidence supports belief/high credence (and vice versa). This feature is important to ensure that these views can capture cases of defeat. Since Fletcher's total evidence incriminates the gardener, it can't also justify an agnostic attitude. On standard evidentialist views, a person's total evidence determines what they have propositional justification for (see, e.g., Feldman and Conee Reference Feldman and Conee2001). For the person to have a doxastically justified attitude, that attitude has to be supported by their total evidence and also be properly based on that evidence. This is not the case in DNA evidence #2. Since Fletcher's evidence supports a high credence that the gardener is the murderer, a middling or low credence that the gardener did it can't be propositionally or doxastically justified for her.

Reliabilists tend to see doxastic justification as fundamental, which is spelled out in terms of a process that reliably produces a valuable target state. Applied to credences, for an agent to have a doxastically justified credence, the credence must be formed via a process that reliably leads the agent to match the objective (or evidential) probability that is warranted by the agent's evidence (Comesaña Reference Comesaña and McCain2018; Dunn Reference Dunn2015; Pettigrew Reference Pettigrew2021; Tang Reference Tang2016). A doxastic state is propositionally justified for the agent when there is a reliable process available to the agent by which they could come to have it. Since Fletcher's evidence is such that she could come to see by deliberating thoroughly that it supports high confidence in the gardener being the murderer, a reliabilist view also can't explain why her agnostic state (i.e., her middling confidence) about who the murderer is seems justified. This problem is not particular to DNA evidence #2; it generalizes. Whenever an agent is agnostic (or has a non-extreme credence) during a deliberative process, although their total evidence supports a different attitude, standard views of justification can't explain why the agnostic attitude is justified.Footnote ⁵

But perhaps we can appeal to an “appreciation” condition, i.e., say that Fletcher's agnostic attitude is justified because she currently doesn't appreciate that her evidence incriminates the gardener. After all, an agent's appreciation of what their evidence supports counts as a permissible epistemic factor that can influence justification, at least according to McGrath. It certainly seems plausible to say that Fletcher's lack of appreciation of what her evidence supports explains why she couldn't currently have doxastically justified high confidence that the gardener did it. However, standard accounts of justification usually don't count an agent's current lack of appreciation of the significance of their evidence as a factor that justifies having a particular attitude. That is for good reason – we generally don't want to allow an agent's obtuseness to be a positive contributor to the justification of their attitudes. Suppose Detective Tupper, who is prone to jumping to conclusions, comes to believe that p is the case based on some of his evidence e. He doesn't appreciate that he has further evidence d that acts as a defeater, such that e + d at best supports middling confidence in p. It seems highly counterintuitive to say that Tupper's belief in p is justified because he doesn't appreciate the significance of d, rather, we want to say the opposite, and that's what all standard accounts of defeat predict.

Instead, the agent's appreciation of what their evidence supports is usually taken to play a more principled role in traditional theories of justification. What matters for justification is whether the agent is generally capable of appreciating an evidential support relation, not whether they currently do (see, e.g., Feldman and Conee Reference Feldman and Conee2001). The idea is to relativize propositional justification to an agent's epistemic capacities: for e to support p for some agent, the agent must be in principle able to see the relation between e and p. This is plausible – if it's far too complicated for an agent to see that e supports p, we might not want to say that an agent who possesses e has propositional justification for p. This type of view avoids making an agent's temporary obtuseness a positive contributing factor to the justification of their attitudes. However, this means that we can't say that Fletcher's agnostic attitude is propositionally or doxastically justified in DNA evidence #2, because we have stipulated that by deliberating a bit longer, she could come to see that her evidence points to the gardener being the murderer.

Thus, there seems to be a fundamental tension between our intuitions: in cases like Detective Tupper's, we don't want to say that the agent's belief is justified because they don't appreciate that it is defeated. But in cases like Detective Fletcher's, it seems natural to say that their agnostic attitude is justified if they don't yet appreciate what their evidence supports. We can't accommodate both of these intuitions on a traditional view of justification, even if we divide the neutral stance into the states of being agnostic and suspending judgment.

4. A possible solution

If we want to account for the intuition that Fletcher's agnostic attitude in DNA evidence #2 is justified, then we need to somehow depart from epistemic orthodoxy.Footnote ⁶ One option is to adopt the distinction between transitional attitudes and terminal attitudes that I have recently put forth in a pair of papers (Staffel Reference Staffel2019, Reference Staffel2021). I argue that to account for our intuitive verdicts about the rationality of attitudes we form in the process of complex deliberation, we need to appeal to a distinctive type of attitude and standard of justification that applies to attitudes formed in the process of reasoning, rather than as conclusions of reasoning. I distinguish in the usual way between two components of inquiry: learning new information, and deliberating about what it supports. In simple cases, we can see the impact of newly learned evidence immediately without needing to deliberate. But in more complicated cases like Fletcher's, deliberations proceed in stages: the stage of evaluating how one's information bears on answering some question, and the stage of reaching a conclusion. To characterize the attitudes formed during these two stages, I distinguish between terminal attitudes and transitional attitudes. A terminal attitude is any doxastic attitude formed as the conclusion of one's reasoning about how one's evidence bears on some question q.Footnote ⁷ By contrast, a transitional attitude is an attitude that is formed toward a possible answer to q at the start of or during one's reasoning about how to answer q.

These two attitude types play different roles in our reasoning, which means that they differ in their descriptive properties. Due to their different roles, they are also subject to different norms of epistemic evaluation. Typically, terminal attitudes are stable unless new evidence is gathered or the agent reopens deliberation. They are the conclusions of deliberative processes, in which an agent determines how their information bears on answering a particular question. They serve in the familiar way as bases for assertion, action, and as premises in further reasoning. Transitional attitudes, by contrast, reflect the agent's evolving take on the question under consideration, and so they can fluctuate as the agent keeps deliberating. They are typically not used as bases for assertion, action, or as premises for reasoning in unrelated contexts, unless the agent is forced to rely on a transitional attitude due to interruptions, time constraints, or other obstacles to completing their deliberation.

With regard to what justifies these attitudes, I propose that standard theories of epistemic justification like evidentialism or reliabilism are well suited for evaluating whether terminal attitudes are justified. This was presumably their intended application all along; hence, this part of the view aligns with epistemological orthodoxy. Departing from orthodoxy, transitional attitudes are claimed to be justified according to a different standard, called pro tem rationality, which takes into consideration stages of permissible reasoning processes. A transitional attitude is rational or justified relative to how much it is supported at the stage of reasoning the agent has reached. This idea is captured in the following schematic account:

This proposal can be filled in and further explicated depending on one's particular epistemological commitments, but roughly, condition (I) is motivated by the idea that agents cannot have rational transitional credences if they use a cognitive process that is completely unsuitable for settling the question at hand, for example, thinking only about astrological data to find a murderer.Footnote ⁸ Condition (II) captures the idea that the agent's attitude should be sensitive to the stage of reasoning the agent has reached, i.e., how much of their evidence they have incorporated into their deliberation and how their deliberation has been conducted up to time t.

We can now account for our intuitions about Detective Tupper and Detective Fletcher as follows: Tupper cannot justifiedly conclude that p is true because he possesses defeating evidence, just as standardly assumed. But Fletcher's agnostic attitude in DNA evidence #2 (and in #1!) is transitional, so what matters is whether it is supported at this intermediate stage of her reasoning. Since she's using ordinary reasoning about her relevant evidence as her reasoning method, condition (I) of the definition is met. But she has not yet been able to put together that her evidence points to the gardener, perhaps because some subtle evidential connections have eluded her so far. Hence, at this stage of her deliberation, being agnostic (and having middling or low-ish confidence) that the gardener did it is the pro tem rational transitional attitude for her to have.

This verdict doesn't depend on her getting more evidence later. The proposed view is thus “traditionalist” in McGrath's sense, because pro tem rationality does not depend on the kinds of non-evidential factors he wants to exclude. Yet, the view is also non-traditional in that it is not exactly parsimonious – instead of postulating one standard of justification that applies to all doxastic attitudes, it doubles how many doxastic attitudes and standards of justification there are. I will now determine whether such extravagance can be avoided by exploring another strategy, which appeals to changes in the agent's evidence while they are deliberating.

5. A parsimonious alternative?

What makes it difficult to say that Fletcher's agnostic attitude is justified in DNA evidence #2 is the assumption that Fletcher's evidence stays constant while she is reasoning. We might thus try out an alternative view on which Fletcher's evidence changes as she reasons, with additional evidence being generated by her deliberations. Call this the evidence generation view. It doesn't distinguish between transitional and terminal attitudes and it assumes that all doxastic attitudes are evaluated by the same, traditional standards of justification. On the evidence generation view, agents can have different justified attitudes at the beginning, middle, and end of deliberation because their evidence changes. Their total evidence keeps increasing, because as they reason, they generate new evidence, which changes what their total evidence supports.

Hence, in DNA evidence #2, the view says that Fletcher's evidence doesn't support that the gardener did it at the beginning and middle stages of her deliberation. She's lacking crucial pieces of deliberation-based evidence because she has not yet discovered the relevant evidential connections. Once she acquires more evidence that incriminates the gardener, either via her own reasoning or from the DNA test, she becomes (more) justified in believing/having a high credence that the gardener is the murderer.Footnote ⁹ Although the evidence generation view seems to be able to handle the problematic DNA evidence #2 case, it also encounters some serious problems. First, it's not obvious that the view can explain why Detective Tupper's belief that p is unjustified. Since he lacks deliberative evidence that would alert him to the relevance of the defeater he possesses, we might be forced to say that his current evidence justifies his belief. Secondly, we'll see that the evidence generation view falls into a vicious regress.

In Lewis Carroll's famous paper “What Achilles said to the tortoise,” the tortoise challenges Achilles to force her to make a logical inference (Carroll Reference Carroll1895). As Besson (Reference Besson and Star2018) explains, it is controversial what their dialogue teaches us, but one insight from it appears to be that not every feature of a reasoning process that makes it justified can be represented as a premise, since doing so generates a regress. Recently, this problem has been debated in the literature on how to formulate theories of explanationist evidentialism, and it is also relevant in our context of evaluating the prospects of the evidence generation view (Appley and Stoutenburg Reference Appley and Stoutenburg2017; Grosz Reference Grosz2020; Lutz Reference Lutz2020).

For the evidence generation view, a problematic regress appears once we combine three claims:

1. (i) What an agent is justified in believing depends on their total evidence.

2. (ii) Being justified in believing something requires insight into the relevant evidential support relations.Footnote ¹⁰

3. (iii) Those insights, once the agent has them, become part of their evidence.

The regress then appears in the following way: let E be a body of evidence that logically or probabilistically supports some claim h.

1. 1. For S to be justified in believing some claim h, h must be supported by S's total evidence.

2. 2. For S to be justified in believing h based on their total evidence E, S must have the insight that E supports h.

3. 3. S's insight that E supports h, once it obtains, becomes part of S's evidence.

4. 4. If (3), then this new evidence together with E yields a new body of total evidence, E*.

5. 5. For S to be justified in believing h, S must have the insight that E* supports h.

6. 6. S's insight that E* supports h, once it obtains, becomes part of S's evidence.

7. 7. If (6), then this new evidence together with E* yields a new body of total evidence, E**.

8. 8. Rinse and repeat… (inspired by Lutz Reference Lutz2020)

The question is then which of the three claims that generate the regress should be rejected. We'd like to know whether there is a way to escape the regress that lets us hold on to a conservative and parsimonious solution to the problem of how transitional attitudes are justified.

The first option is to deny that what an agent is justified in believing depends on their total evidence. This blocks the regress, because the agent doesn't necessarily have to keep reassessing their evidence as it grows when their insights about evidential connections are added. Unfortunately, this move won't save the evidence generation view. Denying a total evidence requirement means rejecting a basic assumption in most theories of justification. But even if we could make this palatable, the resulting view is unattractive for our purposes. If the agent need not take into consideration evidence that pertains to support relations they have recognized, then the Evidence Generation View no longer has a way of accounting for DNA Evidence #2. According to the evidence generation view, the reason why the justified attitudes change in the process of complex deliberations is that the agent's evidence changes as they gain more insight into evidential relations through reasoning. If this insight-based evidence is not part of the total evidence the agent needs to consider in order to be justified, the presence of this evidence can't explain why what they are justified in thinking changes.

The second way of blocking the regress involves denying that being justified in believing something requires insight into the relevant evidential support relation. If this kind of insight is not needed for being justified in believing something, then the agent doesn't have to keep acquiring it whenever their evidence changes. Unfortunately, this option is no more attractive than the previous one. It opens the door for someone to have a justified belief, even though they have no idea that their evidence supports it. Most theories postulate something like a basing requirement to rule out this possibility. Recall that we're not making any very specific assumption about what “insight” amounts to, so this problem can't easily be avoided by changing one's explanation of what the nature of this insight is. Further, this way of blocking the regress would have the effect that the evidence generation view no longer has an explanation of why Fletcher's agnostic attitude is justified in DNA evidence #2. Her growing insight into evidential support relations was supposed to account for the fact that Fletcher can initially be justifiedly agnostic about whether the gardener did it, even though further deliberation would reveal that he is the murderer. But if insight into evidential support relations is not required for having justified doxastic attitudes, then changes in one's insight into these relations can't explain why what is justified for an agent changes as their insight grows.

This leaves the third way of blocking the regress: denying that insights into evidential support relations, once the agent has them, become part of their evidence. This way of avoiding the regress is adopted by Lutz (Reference Lutz2020) and Weatherson (Reference Weatherson2019) in formulating their respective views of justification. Here's how they explain their positions:

Denying that insights constitute evidence is also part of my view mentioned above. On my favored account, which conclusion an agent can rationally adopt doesn't depend on the agent's current insight into their evidence. If their total evidence supports a high credence in p, then that's propositionally rational for them. By contrast, their current insight into their evidence matters for what is rational for an agent to think at an intermediate stage of their reasoning process. But this insight is not part of their evidence.Footnote ¹¹

Hence, Lutz and Weatherson and I all deny in some way that gaining insight into evidential support relations always adds to one's evidence. This allows all of the views to escape the regress, because they don't entail that agents have to keep reassessing what their total evidence supports as their insight grows. Of course, this means they all reject the central claim of the evidence generation view. This brings us back to the question whether we really need two standards of justification for transitional and terminal attitudes to account for our intuitions in DNA evidence #2. Perhaps Weatherson's or Lutz's proposals about the role of our insight into the evidence could do the trick?

Lutz suggests that, as long as an agent has “suitable background beliefs involving explanatory connections,” the agent can be justified in making inferences without a regress being generated. However, while appealing to standing background beliefs can block the regress, it doesn't help us explain why Fletcher is initially justified in being agnostic in DNA evidence #2. This is because her already held background beliefs about what sorts of things can explain what other sorts of things presumably don't change while she deliberates, and without such a change, we can't say that different attitudes are justified for her at different stages of her reasoning.

What about Weatherson's view? He wants to say that there is a difference between obvious and non-obvious entailments.Footnote ¹² Agents can be immediately justified in making obvious inferences, but for non-obvious entailments, they sometimes have to gather additional evidence by reasoning their way to intermediate premises. This makes it seem like Weatherson might be able to explain what's going on in Fletcher's case without having to appeal to two standards of justification. Perhaps in cases like DNA evidence #2, an agent can't be justified in believing a conclusion unless they have generated a crucial intermediate premise, and if they have not added this intermediate premise to their evidence via reasoning, their evidence doesn't yet support the relevant conclusion.

Unfortunately, this explanation is not general enough to cover all the different ways in which we could spell out what happens in DNA evidence #2. For all we have said, the way in which Fletcher could see that her evidence supports that the gardener did it does not actually require a complicated derivation. Perhaps it only requires a very simple inference, and the difficulty lies in attending to the right pieces of evidence at the same time in order to see that the gardener must have done it. Fletcher just needs to attend to the crucial facts simultaneously, and then she will find it obvious right away that the gardener is the murderer. If we fill in DNA evidence #2 in this way, then this is not a case whether the agent must generate an intermediate premise in Weatherson's sense in order to have sufficient evidence for the conclusion that the gardener is the murderer. Hence, his view can't give us a general explanation of why agents are sometimes justified in being agnostic while they are still deliberating, even though their total evidence does not support agnosticism. This leaves my proposed two-standard view as the only remaining contender, at least for now.