2. Background

The relevant question in the debate between permissivists and impermissivists is whether every body of evidence determines at most one rational doxastic attitude, where potential candidates for doxastic attitudes include credences and belief states.Footnote ² While my proposal will be flexible enough to understand doxastic attitudes as either credences or belief states, I will stick to belief states for ease of explication. Note, however, that there is ambiguity in whether a body of evidence determines at most one rational doxastic attitude for some particular agent or for all agents. To account for this, we can distinguish between intrapersonal and interpersonal versions of permissivism as follows:Footnote ³

The default use of “uniqueness” and “permissivism” in the literature is in the interpersonal sense, where the issue is whether different agents with the same evidence can rationally hold different doxastic attitudes. So, for the sake of continuity, “uniqueness” (or “impermissivism”) and “permissivism” will be used to refer to interpersonal uniqueness/permissivism unless otherwise noted.

Notice that the uniqueness/permissivism debate presupposes that evidence plays a key role in determining whether one’s belief is epistemically rational/justified.Footnote ⁴ The central question is whether more than one doxastic attitude is rationally permitted given a body of evidence. To illustrate with a more concrete example, consider a classic case of disagreement – the jury case:

It is understood here that Maria and Sanjay have the same evidence, where what counts as evidence – as Kopec and Titelbaum (Reference Kopec and Titelbaum2016) point out – is to some extent up for debate (191).

Since the uniqueness/permissivism literature typically presupposes internalism, it will help to define internalism and externalism. The debate between internalists and externalists deals with the question of what determines justification.Footnote ⁵ On internalist accounts, as I will understand them, an agent’s justification for her belief supervenes solely on her non-factive mental states. On externalist accounts, justification supervenes on more than an agent’s (non-factive) mental states (Pappas Reference Pappas2014). My focus will be on one particular kind of externalism, process reliabilism. In its most basic form, process reliabilism says that a belief is justified if and only if it is the product of a reliable belief-forming process. Process reliabilism is an externalist account of justification because although mental states often serve as inputs to belief-forming processes, it is – as Pappas points out – the reliability of these processes that determines whether the resulting belief is justified (Reference Pappas2014). And reliability depends not only on features of one’s mental states and processes but also on features of the world.

3. Why should we want to be rational?

A general challenge posed in the uniqueness/permissivism literature is the value problem, which deals with the question of why we would want to be rational given that we want to be accurate. If rationality entailed accuracy, of course we would want to be rational. If, however, being rational does not guarantee that one will be more accurate, why does rationality matter? The value problem is motivated partly by the idea that we should care about rationality because being rational will help get us to the truth. The value problem is taken by people like White (Reference White2005) and Horowitz (Reference Horowitz2014) to be an issue for permissivism in particular. Part of the issue Horowitz and White are capturing is that if rationality permits more than one response to a body of evidence, it is unclear how being rational helps us to be more accurate. Initially, then, it seems that the burden of proof is on the permissivist to demonstrate how their view accounts for the value of rationality.

Internalist responses to the value problem do not take into account what the world is like. Internalist accounts hold that an individual’s belief is rational on the basis of features internal to them. An individual has a true belief, however, when that belief matches the world. On internalist accounts, then, what makes a belief rational (i.e., one’s mental states) is not directly connected to what makes that belief true (i.e., the world). So, even though internalists often believe (or at least hope) that being rational will lead to the truth, they need to explain why we should care about rationality insofar as our primary aim is more accurate doxastic attitudes.

One such internalist response to this question comes from Schoenfield, who is a permissivist. On Schoenfield’s account, the credences one judges to be rational are those they judge to be compatible with the cognitive properties they endorse (Reference Schoenfield2018: 296). Cognitive properties are properties of the system through which an agent responds to evidence by forming doxastic attitudes, and they (a) can be captured solely in descriptive terms and (b) supervene on an agent’s non-factive mental states (288). Examples of cognitive properties include believing whatever is logically entailed by your evidence and believing that p when it perceptually appears to you that p (288). Schoenfield says that to endorse a set of cognitive properties is to regard those properties as accuracy-conducive. If Schoenfield’s account works, it seems to explain why one would want to be rational when their primary goal is accuracy – namely, one should adopt the credences they judge to be rational because they judge these credences most likely to be accurate (Reference Schoenfield2018: 296).

As an impermissivist, Horowitz gives a different internalist response to the value problem (Reference Horowitz2019: 238). Horowitz (Reference Horowitz2014) argues that rationality requires regarding the credences recommended by one’s epistemic rule as maximizing expected accuracy (44, 47). Suppose Wilma is a rational agent who knows she is rational and knows that her way of responding to a body of evidence is rational. Since Wilma also knows that impermissivism is true, she knows that her way of responding to the evidence is the only rational way – so any rational agent will respond the same way. Moreover, any rational agent must regard the credences recommended by this uniquely rational epistemic rule as maximizing expected accuracy given a body of evidence (Reference Horowitz2014: 46). So, on Horowitz’s account, rationality is valuable because being rational maximizes expected accuracy.

Both Schoenfield and Horowitz only establish that if one is rational, then by their own lights, they will expect their credence in a proposition p to be most accurate. So they only provide a connection between rationality and expected accuracy. When we are considering the value of rationality, however, we care about more than just expected accuracy. We want it to be objectively likely that what rationality tells us to believe is true. To do this, I argue, we need an externalist view like process reliabilism. I will also argue that process reliabilism is permissive – and recall that the value problem is thought to be an issue for permissivism in particular. So, one upshot of my argument is that there is a permissivist response to the value problem that is more satisfying than permissivist responses that presuppose internalism.

4. Process reliabilism and the uniqueness/permissivism debate

It might not be obvious that standard process reliabilism is permissive because it has largely been excluded from the debate between permissivists and impermissivists.Footnote ⁶ To see how a process reliabilist might understand the debate, consider Alvin Goldman’s account. Goldman (Reference Goldman and Pappas1979) says that the justification of a belief comes from the reliability of the process or processes that cause the belief – and a process is reliable if it tends to produce true beliefs rather than false ones (Reference Goldman and Pappas1979: 11).Footnote ⁷ Goldman notes several suggestions for determining the truth-conduciveness of a process, but for the purposes of this paper, I will assume that a process is truth-conducive just in case it has a general propensity to produce true beliefs in the world in question.^{Footnote 8,Footnote 9} There is also the question of how reliable a belief-forming process must be to confer justification (Goldman, Reference Goldman and Pappas1979: 11). Goldman says that some vagueness is expected here, but it seems clear that perfect reliability is not required. Many (but not all) belief-forming processes include other beliefs as inputs. For example, tokens of memory process types often involve beliefs about past experiences (Goldman, Reference Goldman and Pappas1979: 13). Goldman says that in cases where some of the inputs to a belief-forming process are other beliefs, the reliability of the process depends in part on the justification of these input beliefs. If the input beliefs are not justified, a process involving those beliefs does not count as reliable (Reference Goldman and Pappas1979: 13).

To sum up: Whether one’s belief is justified depends on the reliability of the process that produced the belief and, where relevant, whether the input beliefs are themselves justified.Footnote ¹⁰ One need not know that the process is reliable or be justified in believing that the process is reliable for the resulting belief to be justified.

Notice that the process reliabilist view I have highlighted has not yet mentioned evidence. In fact, when Goldman (Reference Goldman and Pappas1979) presents his initial process reliabilist view, he deliberately avoids appealing to evidence because he wants his theory to be free from normative epistemic terms (2). Since the debate between permissivists and impermissivists assumes that evidence is relevant when determining what it is rational to believe, we will need to translate the process reliabilist view to square it with the debate. Unfortunately, the question of what a process reliabilist means by “evidence” is not easy to answer. For example, Kornblith (Reference Kornblith2015) – who is a proponent of process reliabilism – argues that we should take seriously the possibility that evidence should not play as large a role in our epistemological theorizing as we often think (238).

One plausible strategy for the process reliabilist is to consider “evidence” as inputs to the belief-forming process(es).Footnote ¹¹ As Goldman and Beddor (Reference Goldman and Beddor2021) point out, even though process reliabilists do not often call inputs to belief-forming processes “evidence,” there seems to be no reason precluding them from doing so. For one thing, this proposal meets Goldman’s stipulation that if the term “evidence” is to be used, it must be fleshed out in non-epistemic terms (Goldman and Beddor Reference Goldman and Beddor2021). Since inputs are typically things like doxastic attitudes and perceptual or memory experiences, they seem to meet this requirement.

If we understand evidence this way, then for the process reliabilist, the uniqueness/permissivism debate concerns whether a set of inputs determines at most one rational doxastic attitude. On Goldman’s view, there are many cases in which some of the inputs to belief-forming processes are other beliefs (Reference Goldman and Pappas1979: 13). This means that for agents to have the same inputs, they must not only take in the same information, but they must also have the same initial beliefs that are relevant to the question at hand.^{Footnote 12,Footnote 13} I suggest that a process reliabilist can understand the permissivism/impermissivism debate in terms of the following statements:

At this stage, it is worth asking whether introducing a process reliabilist notion of justification to the uniqueness/permissivism debate is changing the subject. It seems like people involved in the debate care about what is rational, and it is not clear how a process reliabilist account of justification is supposed to capture the notion of rationality. But an internalist conception of justification as determined by reasons that are cognitively accessible to an agent presumably does capture a core feature of rationality. So, one might worry that even if process reliabilism provides a more satisfying response to the value problem than internalist alternatives (as I will claim), it does not address what it is to be rational.

In response to this worry, I accept that equating a process reliabilist account of justification with rationality is changing the subject. However, this change of subject is motivated partly by the value problem itself. Insofar as we are concerned with the value problem, we have reason to prefer an externalist notion that builds in a direct connection between what makes a belief rational and what makes that belief true. Additionally, before the uniqueness/permissivism literature came to fruition, most people – including process reliabilists – were assuming the truth of uniqueness. Thus, an argument (like the one I will put forward) showing that process reliabilism lends itself to permissivism encourages people (especially process reliabilists) to rethink this assumption. Finally, introducing an externalist perspective into an internalist-dominated debate may pave the way for the inclusion of other externalist voices.

5. A defense of process reliabilist permissivism

With this background of process reliabilism in mind, we are now in a position to ask whether process reliabilism is permissive or impermissive. I will argue that it is permissive regardless of whether all reliable processes are deterministic.

For a process reliabilist, there is a question of whether all reliable belief-forming processes are deterministic. In other words, does every reliable process always yield the same output given the same inputs? Weisberg (Reference Weisberg2020), for example, argues that evidence-processing can be stochastic. Specifically, he claims that there may be some randomness in how a body of evidence is processed such that different agents with the same evidence (or the same agent given her evidence) may believe either p or ∼p depending on how their processing of this evidence goes.Footnote ¹⁴ Moreover, either belief would be rational provided that it would be rational for an agent who arrived at that belief to use that process (Reference Weisberg2020: 9). Weisberg illustrates this point by way of a case in which an agent is trying to determine whether more than half of her friends wear glasses. Her body of evidence includes memories of her friends’ faces, and she processes this evidence by taking a random sample of those faces. On this occasion, many of her glasses-wearing friends come to mind, so she believes that more than half of her friends wear glasses. However, the random sampling could have gone differently for her in that she could have instead recalled more friends that do not wear glasses and thereby believed that less than half of her friends wear glasses. Nevertheless, assuming the agent is rational in using this method, she is in a permissive case because given her evidence, she could either rationally believe that more than half of her friends wear glasses or rationally believe that less than half of her friends wear glasses (Reference Weisberg2020: 9).

If Weisberg is correct in thinking that evidence-processing is stochastic (i.e., that not all reliable processes are deterministic), it is easy to see how process reliabilism is an intrapersonally permissive view.Footnote ¹⁵ Suppose there are cases where a reliable process can yield different outputs given the same inputs. More specifically, suppose that Sanjay’s inputs and reliable belief-forming processes yield the belief that Elijah is not guilty 90% of the time. The other 10% of the time, however, the same inputs and belief-forming processes yield the belief that Elijah is guilty.Footnote ¹⁶ Recall that on the process reliabilist account under consideration, rational belief is determined by the reliability of the belief-forming process. So, if there are cases where the same reliable process can yield p on some occasions and ∼p on others, an agent using that process is permitted to believe either p or ∼p. Of course, this does not entail that the agent is permitted to believe p and ∼p. Rather, their belief that p is rational if the reliable process yields the belief that p, and their belief that ∼p is rational if that process yields the belief that ∼p. Nevertheless, this can be classified as a version of intrapersonal permissivism because each of p and ∼p can be rational for the same agent depending on the outcome of their input processing. Thus, in cases where Sanjay’s inputs and reliable belief-forming process produce the belief that Elijah is not guilty, that belief is rational. In cases where those same inputs and reliable processes produce the belief in Sanjay that Elijah is guilty, that belief is rational.

Similar reasons hold for thinking that if some reliable processes are indeterministic, process reliabilism is a form of interpersonal permissivism. If not all reliable processes are deterministic, there are cases in which different agents with the same inputs and processes can rationally adopt different doxastic attitudes. Suppose again that given the same inputs, a reliable process yields the belief that p 90% of the time and the belief that ∼p the other 10% of the time. Given a set of inputs, it is possible for one agent who uses that process to rationally believe p, but for another agent who uses that process to rationally believe ∼p.Footnote ¹⁷ With respect to the jury case, suppose Sanjay and Maria have the same inputs, and they use the same reliable belief-forming process. Nevertheless, these inputs and processes sometimes produce the belief that Elijah is guilty and other times produce the belief that he is not guilty. On this occasion, Sanjay’s process given his inputs yields the belief that Elijah is not guilty. When Maria uses that same process given the same inputs, however, it produces the belief that Elijah is guilty. Since Sanjay and Maria’s beliefs are both the product of a reliable belief-forming process, their beliefs are both rational according to process reliabilism.

Suppose, on the other hand, that all reliable processes given a set of inputs are deterministic. Under these conditions, any reliable process given a set of inputs will always produce the same doxastic attitude toward a proposition p. One might think this would make process reliabilism an impermissive view. If all reliable processes are deterministic, the only way process reliabilism can be permissive is if there are cases where multiple belief-forming processes are available given a set of inputs. If all reliable processes are deterministic and every set of inputs is such that there is only one rational way of processing them, it follows that every set of inputs determines at most one rational doxastic attitude – which makes process reliabilism impermissive. But I will argue that a process reliabilist should not think there can only be one reliable process available to an agent given a set of inputs. Rather, she should think there are cases where there is more than one reliable process available to an agent given their inputs, and they may rationally adopt a different doxastic attitude depending on which process they use.Footnote ¹⁸

I already noted that there is disagreement among process reliabilists with respect to how reliable a process must be for a belief formed by way of that process to be justified. There is also disagreement about whether this threshold is vague or precise. Some process reliabilists – including Goldman (Reference Goldman and Pappas1979) – think that the threshold for justified belief is vague in the sense that there is no clear point at which a process type is reliable enough, but there are clear cases of processes that are reliable enough and ones that are not.^{Footnote 19,Footnote 20} Luckily, my argument does not require us to settle either of these disagreements. My argument will go through regardless of the answers a process reliabilist gives to the questions of how reliable a process must be and whether this threshold is clear or vague. The important issue for my purposes is not what the threshold is like, but whether a process r beats the threshold, because the latter is what determines whether a belief produced by r is justified. And I will argue that a process reliabilist should accept the claim that there are cases where multiple belief-forming processes cross the threshold. As a result, if one of these processes produces the belief that p, this belief is justified. And if another one of these processes produces the belief that ∼p, this belief is also justified.

One reason there will be cases where multiple processes are reliable enough to produce a justified belief is that different belief-forming process types can have an equal propensity to produce true beliefs and thus be equally reliable. For example, given a set of inputs, it is possible for there to be more than one reliable process available that is, say, 92% reliable. Moreover, there can be occasions in which these equally reliable belief-forming processes yield different outputs given the same inputs. Since reliability is a property of types, even if two processes r and r’ are equally reliable, there is still the possibility that on some specific occasion, given the same inputs, r will yield the belief that p, and r’ will yield the belief that ∼p.

Another reason there will be cases where multiple processes cross the threshold is that, regardless of whether the threshold is vague or precise, there will be justification-conferring processes with slightly different degrees of reliability. If the threshold is vague, there will be clear cases of processes that are reliable enough to produce a justified belief but differ in terms of their reliability. For instance, a process that is 91% reliable intuitively yields a justified belief, and so does a process that is 89% reliable. If the threshold for justification is precise, any process that crosses this precise threshold will produce a justified belief. And processes with slightly varying degrees of reliability will cross the threshold.Footnote ²¹ So, even if all processes are deterministic, there could still be situations where one reliable way of processing a set of inputs yields the belief that p while another reliable way of processing those same inputs yields the belief that ∼p.

Therefore, process reliabilism is still interpersonally permissive even given the assumption that all reliable processes are deterministic. It is interpersonally permissive because there will be cases where different agents with the same inputs can use different justification-conferring processes. And it could be that, given these inputs, one reliable process r produces the belief that p, but another reliable process r’ produces the belief that ∼p. As we have seen, one way these justification-conferring processes (r and r’) could be different is if they are equally reliable but are tokens of different process types. Another way that r and r’ could be different is if they have marginally different degrees of reliability. Either way, there will be cases where r yields the belief that p while r’ yields the belief that ∼p. Footnote ²²

The view that rationality is interpersonally permissive because different agents with the same evidence can use different truth-conducive processes bears some similarities to Simpson’s (Reference Simpson2017) proposal. Simpson illustrates his view via a hypothetical case involving fictional detectives Nancy Drew and Veronica Mars. In this scenario, Drew and Mars are investigating a crime together where X and Y are the main suspects. Upon carefully examining the same evidence, Mars believes it is 70% likely that X committed the crime and 30% likely that Y committed the crime. Drew, however, believes it is 30% likely that X is the culprit and 70% likely that Y is the culprit. On Simpson’s view, Mars and Drew can both be rational in holding different doxastic attitudes about X and Y’s guilt. This is because Mars and Drew’s respective doxastic attitudes are formed via epistemic standards (i.e., methods for forming beliefs) that reliably produce true beliefs given their cognitive abilities, which are the skills they implement when applying their epistemic standards (531).Footnote ²³ But Mars and Drew possess different cognitive abilities. Mars is skilled at correctly identifying insincere testimony, while Drew is very good at making holistic judgments about large bodies of evidence. So, Drew and Mars are cognitively well-equipped to apply different epistemic standards. While Mars examines each piece of testimonial evidence separately, Drew interprets all of the testimonial evidence as a set.

Although a process reliabilist version of permissivism (given determinism about processes) also involves different agents forming different beliefs via distinct truth-conducive processes, my proposal goes beyond Simpson’s in at least two ways. First, process reliabilism sets a minimum threshold on how truth-conducive a process must be to produce justified beliefs.Footnote ²⁴ However, Simpson’s account seems to leave open the possibility that an agent’s doxastic attitude can be rational if it is produced via epistemic standards that are not likely to yield accurate doxastic attitudes. It could be that an agent is employing the epistemic standards that are maximally truth-conducive for her given her cognitive abilities, but those standards are still bad at producing accurate doxastic attitudes.Footnote ²⁵ Second, Simpson’s account is interpersonally permissive, but he does not specify whether it allows for permissive cases at the intrapersonal level as well.Footnote ²⁶

I will explicitly argue, however, that process reliabilism is intrapersonally permissive even if we assume determinism about reliable processes. There will be cases where the same agent has different reliable processes (call them r and r’) available to them given their inputs, and r produces the belief that p while r’ produces the belief that ∼p. Again, r and r’ could be equally reliable but be tokens of different process types, or r could be slightly more reliable than r’. And if an agent uses r, they will justifiably believe p, but if she uses r’, she will justifiably believe ∼p. Consider Sanjay in the jury case. Suppose his inputs include things like information taken in from lawyers’ arguments and different hypotheses about what best explains this information. Suppose that in one variation of this scenario, there are two processes available to Sanjay, r and r’, that are both 93% reliable. In another variation, reliable processes r and r’ are available to Sanjay, and r yields true beliefs 92% of the time, while r’ yields true beliefs 90% of the time. In both variations, if Sanjay uses r, he will justifiably believe that Elijah is not guilty, and if he uses r’, he will justifiably believe that Elijah is guilty. So, Sanjay can justifiably believe either that Elijah is guilty or that Elijah is not guilty.

In practice, one way that agents could use justification-conferring processes with different degrees of reliability given a set of inputs is if reliability and other desirable epistemic features are balanced differently. To see why, note that process reliabilists are not committed to thinking that an agent must always use the most reliable process available to her. In fact, this stipulation would introduce idealizations into the reliabilist picture that may be unwanted insofar as the goal is a non-ideal account of justification. Rather, in addition to reliability, many process reliabilists may take into account features like what Hartry Field (Reference Field1982) calls “power,” which is something like the capacity to generate beliefs (565). Field gives an example of a maximally reliable belief-forming process that only generates beliefs in tautologies (Reference Field1982: 565). He points out that if a person used this process, they would have beliefs about very few things, which suggests that it is not epistemically good to focus exclusively on reliability. Although this example illustrates a bad way to balance reliability and other desirable values like power, perhaps there is more than one good way to make this tradeoff. Indeed, a process reliabilist might hold that there are multiple rational ways for the same agent and different agents to weigh epistemic goals.Footnote ²⁷ Additionally, different ways of making the tradeoff between reliability and other epistemic values will result in the use of processes with marginally different degrees of reliability. An agent who is more cautious on a particular occasion may use a process that is 91% reliable, but that same agent on a different occasion may use a process that is 89% reliable.

In the context of the jury case, it is certainly valuable to take precautions to avoid falsely believing that the defendant is guilty. Nevertheless, there does not appear to be just one epistemically good way of taking these precautions.Footnote ²⁸ Although agents should not gravitate toward the extremes of taking zero precautions or being so cautious that they never believe someone is guilty, there might be more than one reasonable way to make this tradeoff. To illustrate, suppose Sanjay and Maria have the same inputs, and their doxastic attitudes – including their input beliefs – are formed via reliable processes. However, their respective reliable belief-forming processes differ. Maria’s belief-forming process is an instantiation of a reasoning process type that is less scrupulous than Sanjay’s. Suppose the reliable reasoning process instantiated by Maria on this occasion is such that she makes good inductive inferences as well as deductive ones. The reliable reasoning process instantiated by Sanjay is such that he is slightly more hesitant to draw inductive inferences, instead requiring more information to do so. For instance, Maria infers that Elijah was at the crime scene based on the presence of his beanie and his car. Sanjay’s reasoning process, however, is such that he does not draw the same inference from that information alone. Moreover, Sanjay and Maria differ in terms of how trustworthy they perceive certain witnesses to be. Maria is slightly more confident than Sanjay that several witnesses testifying for the prosecution are credible. Suppose these witness statements include claims to have seen Elijah’s car fleeing the scene, psychiatric evaluations, and character testimonies. Given how many witnesses there are, it is not difficult to see how differences in perceived credibility could impact Sanjay and Maria’s beliefs about Elijah’s guilt. If Maria views the witnesses for the prosecution as a bit more trustworthy than Sanjay does, then she is more likely to form the belief that Elijah is guilty than Sanjay is even though they have the same inputs. Relatedly, although Maria and Sanjay both trust the lab results suggesting that the blood found on Elijah’s shirt belonged to the victim, Sanjay gives more weight to the possibility that the results are mistaken. We can imagine that due to these kinds of discrepancies, Maria’s reliable belief-forming process yields the belief that Elijah is guilty, but Sanjay’s reliable belief-forming process yields the belief that Elijah is not guilty. So, Sanjay and Maria have the same inputs, but their divergent reliable belief-forming processes lead them to adopt different beliefs. And, since both of their belief-forming processes are reliable, their beliefs are both rational by process reliabilist standards.

To put all the pieces of my argument together, a process reliabilist framework is a version of permissivism that would say something like the following about the jury case: given the same inputs, it is possible for Sanjay and Maria to rationally form different doxastic attitudes toward the proposition that Elijah is guilty. This could in principle happen in a scenario where their beliefs are the result of the same reliable process, but this process produces different output beliefs. More substantively, multiple justification-conferring processes may be available to Sanjay and Maria, and these processes may yield different doxastic attitudes toward the proposition that Elijah is guilty. These justification-conferring processes may either be different processes that are equally reliable or be processes with marginally different degrees of reliability. And, provided that all of Maria’s and Sanjay’s input beliefs are reliably produced, both of their beliefs would be rational.

Another important issue is the degree to which process reliabilism allows for cases where agents with the same evidence rationally hold different doxastic attitudes toward the same proposition while seeing each other’s doxastic attitudes as rational – which Smith (Reference Smith2020) calls “acknowledged permissive cases” (158).Footnote ²⁹ To illustrate, suppose the jury deliberates and Sanjay and Maria learn that they disagree about Elijah’s guilt. They also each recognize that the other person formed their belief via a reliable process. So, Sanjay sees that Maria’s belief is rational and vice versa.Footnote ³⁰ Some permissivists claim that in cases like this, Sanjay and Maria can maintain their beliefs while still viewing the other person as rational, which would constitute an acknowledged permissive case. But if agents in a situation like Sanjay and Maria’s must always revise their beliefs in light of the disagreement, there may be no acknowledged permissive cases.

The extent to which a process reliabilist posits the existence of acknowledged permissive cases depends largely on how they deal with defeat. When two (or more) agents learn that they disagree, it is plausible to think they have each acquired a defeater for their original belief. Namely, the disagreement indicates either that one of them used an unreliable process or that all of their processes were reliable, but one produced a false belief. Assuming none of the agents have reason to think that someone else is wrong or used an unreliable process, this seems to undermine their justification. Accounts of defeat in which justification is not undermined under these conditions allow for many acknowledged permissive cases because agents can rationally hold onto their original beliefs. But accounts of defeat that hold that justification is undermined in these cases only leave room for acknowledged permissive cases if they also allow agents who disagree to still end up with different beliefs after revising them.Footnote ³¹

Unfortunately, however, the topic of defeat is highly controversial among process reliabilists. There is no generally agreed upon process reliabilist account of defeat that sufficiently captures the notion while respecting the goal of providing a theory of justified belief solely in non-epistemic terms.^{Footnote 32,Footnote 33} Additionally, many process reliabilist accounts of defeat seem to presuppose uniqueness.Footnote ³⁴ But if what I have argued here is correct, that presupposition is false. These views would therefore need to be extended or revised to account for the full range of cases that process reliabilism must account for. A full treatment of process reliabilist defeat in permissive cases is beyond the scope of this paper, so I will leave it as a question for future research.

6. Other common objections to permissive views

Thus far, I have argued that standard process reliabilism is a permissive view and that it is an open question to what degree it allows for acknowledged permissive cases. It is also worth considering some common objections to permissivism in the literature to see whether a proponent of process reliabilism can successfully respond. I will examine two such objections and show that they do not ultimately pose a problem for process reliabilist permissivism.

7. A satisfactory response to the value problem

As I noted in Section 3, philosophers like Schoenfield and Horowitz at best establish a connection between rationality and expected accuracy. Expected accuracy is a matter of one thinking it is likely that their credences are accurate. Actual accuracy, on the other hand, is a matter of one’s credences lining up with the world. Insofar as actual accuracy is important to us, then, responses to the value problem that focus solely on expected accuracy fall short.

Horowitz herself raises the worry that on her view, the connection between rationality and truth is subjective rather than objective (Reference Horowitz2014: 47). She points out that her account leaves open the possibility that a rational agent could be completely wrong about the world. Horowitz says that although this situation is possible, it is subjectively unlikely (Reference Horowitz2014: 47).Footnote ⁴⁰ Namely, a rational agent will believe it is unlikely that she is vastly mistaken about the world. Even if we grant that a rational agent will believe such a scenario is unlikely, this does not mean it is objectively unlikely. Assuming we want reassurance that a rational agent is unlikely to be completely wrong about the world, this response does not dispel the worry.

On views like Schoenfield’s and Horowitz’s, agents merely believe they are likely to form true beliefs. But the process reliabilist view I have presented makes it the case that an agent is actually likely to form true beliefs. Instead of focusing on expected accuracy, externalist views like process reliabilism build actual accuracy into the picture. Under process reliabilism, the reliability of a belief-forming process is determined by its propensity to produce doxastic attitudes that map onto the world. So, a process reliabilist has a straightforward response to the value problem: since there are cases in which multiple reliable belief-forming processes are available given a set of inputs, it is possible for agents with the same inputs to use different reliable processes that produce different beliefs. Thus, even if different agents can rationally hold different doxastic attitudes, there is still a strong connection between rationality and truth – namely, the rational doxastic attitudes are those that are the product of belief-forming processes that tend to produce true beliefs. The connection I propose, then, is between what is rational and what is truth-conducive, not just what one expects to be truth-conducive.Footnote ⁴¹

One might object, however, that process reliabilism may not really secure a connection between rationality and truth. To see why, note that a process may be truth-conducive in some environments but not others. For example, it seems like many perceptual processes have a propensity to yield true beliefs in the actual world. But if these same processes took place within a brain in a vat, they would have a propensity to produce false beliefs a significant percent of the time. And if a process reliabilist takes the reliability of a process to be determined by its propensity to produce true beliefs in the actual world, they would say that the beliefs formed via these perceptual processes are reliably produced and thereby rational. But since most (if not all) of these beliefs are false in the brain in a vat world, there is no direct connection in this world between what is rational and what is true.

I will make two points in response to this objection. The first is that many versions of process reliabilism do not take the reliability of a belief-forming process to be determined by its truth-conduciveness in the actual world. As I noted earlier, I stipulate for the sake of this paper that a process is reliable just in case it has a propensity to produce true beliefs rather than false ones in the world in question.Footnote ⁴² With respect to the brain in a vat case, this means that since perceptual processes are unreliable in that world, the beliefs that result from them are not justified.Footnote ⁴³ On this view, there is a robust connection between rationality and truth because the rational beliefs are those that are produced via processes that tend to yield true beliefs in the world in which the process is being used. Thus, there is a direct link between what makes a belief rational (i.e., the tendency of a process to produce true beliefs in some world w*) and what makes it true (i.e., w*).

Secondly, a process reliabilist who adopts the view that the reliability of a process is determined by its propensity to produce true beliefs in the actual world still has a more satisfying response to the value problem than an internalist. As I have already illustrated, internalist accounts of rationality cannot secure anything beyond a connection between rationality and expected accuracy in any world, including the actual world. Even in the actual world, it could turn out that our non-factive mental states are not guides to the truth (even if we think they are). On the other hand, consider a process reliabilist picture under which beliefs are justified if and only if they are produced by processes that are truth-conducive in the actual world. This picture guarantees a connection between rationality and truth in the actual world because it holds that the rational beliefs about the actual world w just are those that are formed via processes that tend to produce true beliefs in w.

You might worry that this view does not capture something important about the value of rationality. On a process reliabilist framework, agents can be rational even if they are unaware of the reliability of their belief-forming process. But it is not clear why one should care about rationality if they are not always in a position to know that they are rational.Footnote ⁴⁴ Granted, the process reliabilist picture I have developed does not explain – and is not trying to explain – why rationality matters from the perspective of a rational agent. Nevertheless, it does explain why epistemologists who want their theory of rationality to be connected to truth should care about rationality. In particular, under process reliabilism, a rational belief is likely to be a true belief.Footnote ⁴⁵ Moreover, if process reliabilism is true, any agent who sees that it is will be able to appreciate this strong link between rationality and truth.