2.1 The Three Clauses of the Theory

CPR is often characterized roughly as claiming that beliefs are justified if and only if they are the output of a reliable cognitive process. To be more precise, Goldman (Reference Goldman and Pappas1979) can be portrayed as using three clauses in relation to (doxastic) justification: the first deals with the justification of beliefs caused by belief-independent cognitive processes; the second deals with the justification of beliefs caused by belief-dependent cognitive processes; and the third with defeat.Footnote ¹ The distinction between belief-independent and belief-dependent processes has to do with whether the relevant cognitive process takes beliefs as inputs in producing its outputs: if it does, then the cognitive process is belief-dependent and will have different conditions for producing justified beliefs.

But let's start with taking a closer look at the first clause related to belief-independent processes:

Reliability here consists in a cognitive process's tendency to produce true rather than false beliefs. It appears most plausible not to cash out this tendency in terms of the actual truth-ratio of a cognitive process: if a cognitive process is used only once and happens to produce a false belief, then we should not conclude that the process is unreliable. It certainly makes sense to think of a very reliable process that, unfortunately, delivers a false belief the one time it is actually used. So we are better off in thinking about the reliability of a process in counter-factual or modal terms, even if it is still unclear how to cash this out precisely: a cognitive process is reliable if it would produce mostly true rather than false beliefs. Although there is much more to say about this matter, we can use this rough notion for the purposes of this paper.

Goldman realizes that things are somewhat different for belief-dependent cognitive processes that take beliefs as their inputs. Take reasoning and memory as examples of such processes: if one starts reasoning from false premises, or if one had once accepted false beliefs, then reasoning and memory respectively will not produce mostly true beliefs even if there is nothing defective about these processes themselves. For these processes, what matters for justification is not unconditional reliability, but rather conditional reliability: the processes should tend to produce true rather than false beliefs given that the input beliefs are true. This brings us to the second clause of CPR:

Note that this clause not only says that belief-dependent processes should be conditionally reliable if they are to produce justified beliefs, but also that the input beliefs to this process should be justified themselves. This does not lead to a vicious regress, as the first clause can explain how some beliefs can be justified without the need for additional justified beliefs. Indeed, Goldman explicitly acknowledges that his view can be taken as a kind of foundationalism, as long as it's clear that it is not the classical Cartesian variety of foundationalism (Goldman Reference Goldman and Pappas1979: 14–15).

But even with the distinction between belief-independent and belief-dependent processes, clauses 1 and 2 will not be sufficient for ultima facie justified beliefs. One can easily construct counterexamples in which a subject holds a belief that is reliably caused, but in which that same subject also has defeating evidence against the truth of this belief. The subject of such a scenario would not be justified in his belief because of his defeating evidence, even if the belief was the output of a reliable cognitive process. That is why Goldman specifies that the justification of a belief is not only a matter of the cognitive processes actually employed, but also of other cognitive processes that could have been employed (Goldman Reference Goldman and Pappas1979: 20). This brings us to the third clause of CPR:

Although, as Goldman recognizes, there are still some problems left with this way of cashing out defeat, I think these three clauses will, for now, give us a sufficient grasp of CPR. Let's now turn to the challenge raised by Clairvoyance-style cases.

2.2 The Clairvoyance Challenge

Two classical counterexamples have been raised against CPR's claim that the reliability of a belief-independent cognitive process is sufficient for the justification of the resulting beliefs. The first is that of Clairvoyance, presented by Laurence BonJour:

In this case, it seems that Norman's belief is produced by a reliable belief-independent process (clause 1). Furthermore, it's not clear that Norman has any alternative available process that would have resulted in his not believing that the president is in New York City (clause 3). Note that I will return to this second point in section 3, where I will argue that there is, in fact, another process that one can appeal to. But for now, let's accept that Norman's beliefs are not defeated and that clause 3 is satisfied.

If clause 1 and clause 3 are both satisfied, then CPR gives the verdict that Norman has acquired a justified belief. Yet there seems to be something wrong with this verdict: intuitively, Norman appears to be unjustified. So CPR appears to be mistaken.

The second classical counterexample against CPR's sufficiency claim is provided by Keith Lehrer:

Again the problem is that even though CPR would give the verdict that Truetemp's belief is justified — assuming that clause 3 is satisfied on the basis that there is no clear alternative reliable process available to Truetemp — intuitively Truetemp's belief is unjustified. So CPR appears to be mistaken.

BonJour and Lehrer provide a similar diagnosis of what happens in their cases: from Norman's and Truetemp's subjective perspective, it is an accident that their beliefs are true (BonJour Reference BonJour1985: 43–4; Lehrer Reference Lehrer1990: 163). Norman and Truetemp both lack good reasons to believe that they are using a reliable process, and their beliefs cannot be justified in the absence of such reasons.

This specific diagnosis of the case leads to two worries however. First, if one assumes that the same requirements for justification hold for all epistemic agents, then unsophisticated cognizers, like animals and small children, will be precluded from having any justified beliefs. These cognizers are not capable of having any higher-order beliefs about the (reliability of the) sources of their first-order beliefs, so they will lack the reasons that BonJour and Lehrer take to be required for having justified first-order beliefs. Second, requiring justified higher-order beliefs about the sources of one's first-order beliefs quickly leads to an infinite regress: if all justified beliefs require justified beliefs of a higher order, then this same requirement is true for the justified higher-order beliefs themselves. So we would be in need of infinitely more higher-order beliefs of increasing complexity to be justified in our first-order beliefs.

However, even if BonJour's and Lehrer's own diagnosis of the cases is problematic, the cases still present a challenge for CPR. The proposed clauses for justification appear satisfied, yet the intuition of unjustifiedness remains. What is going on?

2.3 Relativizing Reliability

One reliabilist response, the gist of which is already present in Goldman (Reference Goldman and Pappas1979: 16–17), attempts to answer the Clairvoyance Challenge by relativizing reliability to a specific world or specific set of worlds, such as the actual world or a set of ‘normal’ worlds. According to this account, what matters for justification is the reliability of the belief-forming process in a certain privileged (set of) world(s). Given that clairvoyance is not a reliable process in the actual world, Norman would, on a certain development of this account, not have a justified belief that the President is in New York City.

However, the relativization strategy faces a dilemma. On the one hand, one gets the correct result to the Clairvoyance Challenge when reliability is relativized to the actual world, but then this relativization is ad hoc.Footnote ³ There appears to be no reason to assume that the justification of Norman's belief depends on the reliability of his belief-forming process in the actual world. What appears to matter to the justification of his belief is the reliability of the process in his own world. This is also supported by the intuition that there could be worlds in which clairvoyance leads to justified beliefs – something which actual world reliabilism would have to deny.

On the other hand, when one relativizes reliability to a more motivated world, such as a subject's home world, as proposed by Majors and Sawyer (Reference Majors and Sawyer2005), then one no longer has an answer to the Clairvoyance Challenge. Norman and Truetemp are using belief-forming processes that are reliable in their respective home worlds, so they will still turn out to have justified beliefs, contrary to intuition.

One could try to salvage something of the relativization strategy by using it to explain why we mistakenly have the intuition that Norman and Truetemp have unjustified beliefs. The idea would then be that when we consider questions about justification, we mistakenly consider the reliability of the relevant belief-forming process in the actual world rather than its reliability in the world under consideration. Although this would go some way to explain our intuition for the case of Norman, it would be less plausible for the case of Truetemp, which is already pretty close to the actual world. In any case, an error theory like this should not be preferred if there is an alternative way of accommodating our intuitions. I will therefore go on to discuss some other contemporary reliabilist responses in the next sections.

3.1 Inferentialist Reliabilism

According to Jack Lyons' (Reference Lyons2009) inferentialist reliabilism, the Clairvoyance Challenge has to do with the fact that Norman and Truetemp's beliefs are non-basic. Even though the reliability of a belief-forming process is sufficient for the prima facie justification of basic beliefs, more is required for the justification of non-basic beliefs like those of Norman and Truetemp. Just as we typically need all kinds of additional justified beliefs about biology to justifiably believe that “bats are more closely related to primates than to rodents” (Lyons Reference Lyons2009: 135), so do Norman and Truetemp also need additional justified beliefs to have justification for their respective non-basic beliefs. To discuss this answer, we will have to look in more detail at the precise way in which Lyons distinguishes between basic and non-basic beliefs.

The distinction between basic and non-basic beliefs is closely connected to the foundationalist nature of reliabilism. Recall that Goldman provided two clauses about prima facie justification, one about belief-independent processes, and the other about belief-dependent processes. In CPR, beliefs resulting from belief-independent processes form the epistemic foundation on the basis of which other beliefs can be justifiably inferred. However, as Lyons points out, the distinction between basic and non-basic beliefs is not really captured by distinguishing between beliefs that arise out of belief-independent processes and beliefs that arise out of belief-dependent processes (Lyons Reference Lyons2009: 125–7). The epistemic basicality of beliefs has to do with whether they evidentially depend on other beliefs, i.e., with whether these beliefs are based on other beliefs. That this is not the same as being the output of a belief-(in)dependent process can be made clear with the following example.

Introspection could be taken as a belief-dependent process in which, say, the belief that I believe that p is partly dependent on the belief that p itself: the introspective process takes the belief that p as an input, and then outputs the belief that I believe that p.Footnote ⁴ However, the introspective belief that I believe that p can be justified even if the belief that p itself is completely unjustified. All that seems to matter to this introspective justification is that the introspective process is itself reliable, not that the beliefs it takes as input are justified.

This case can be explained by stressing that epistemic basicality does not have to do with whether a process takes beliefs as input to produce an output. What matters is how these beliefs are used as an input. If the content of the beliefs is used as evidence for the conclusion, then the process leads to non-basic beliefs. If the content of the beliefs is not used as evidence for the conclusion, then the process can still lead to basic beliefs. This latter possibility is what goes on in the described case of introspection: although beliefs are taken as input, their content is not used as evidence for the outputted belief. My belief that I believe that p is, in the case described, intuitively not based on the content that p.Footnote ⁵

Given that epistemic basicality thus does not come down to being the output of a belief-independent process, Lyons presents an alternative way of cashing out the basic/non-basic distinction:

Let's take the notion of a cognitive system for granted as a functional entity which performs a specific cognitive task. To produce basic rather than non-basic beliefs, such a cognitive system should be, first, inferentially opaque. This means that the beliefs it produces should not be the result of an introspectible train of reasoning. Second, the system should not base its outputs on any doxastic inputs. In other words, the system should not use the content of any input beliefs as evidence for its output beliefs. This leaves open the possibility that the cognitive system uses the input beliefs in a different way, as, for instance, in the case of introspection.

Both these requirements seem well-motivated as requirements for the type of cognitive system that can produce basic beliefs. If I can introspectively ascertain the train of reasoning that leads to a cognitive system's output beliefs, then it seems that the resulting beliefs will no longer be basic for me. Similarly, if some input beliefs are used as evidence for a system's output beliefs, then the resulting beliefs do not appear to be basic even if I do not know that the input beliefs were used as evidence. However, the third requirement, that the cognitive system has resulted from learning and/or innate developmental processes, seems less well motivated. Why would it matter for a belief's being basic that the system which produced it resulted from learning and/or innate developmental processes?

Lyons attempts to support this etiological constraint by providing a variation of the original clairvoyance case in which the subject under consideration intuitively has a justified rather than an unjustified belief. Here's the case:

According to Lyons, Nyrmoon has a justified belief that his house is on fire. Further, the relevant difference between Norman and Nyrmoon, claims Lyons, is that Nyrmoon's belief is the result of a cognitive system which satisfies the etiological constraint, while the cognitive system responsible for Norman's belief does not satisfy such a constraint. If this diagnosis of the difference between the cases is correct, then Lyons' inferentialist reliabilism can solve the Clairvoyance Challenge. Both Norman and Truetemp lack a cognitive system that developed in the right way: it is somewhat assumed that Norman's clairvoyance system turned up out of nowhere, and Truetemp's temperature system has been artificially implanted. This would mean that their respective beliefs are non-basic rather than basic, and that we should look at a different set of conditions for their justification (e.g., conditional reliability, justified premise beliefs).

Unfortunately, Nyrmoon's case is so different from Norman's that it does not really warrant this conclusion. For instance, not only does Nyrmoon have a clairvoyant system that developed in the right way, he also is a member of a species that has such systems, and it is likely that his clairvoyant system has been selected for by evolution, a point made by Peter Graham (Reference Graham2011). Moreover, Nyrmoon has also used his clairvoyant system before, and presumably has some idea (if only implicitly) of when it works reliably. These other factors could just as well be mentioned as an explanation of the relevant difference between Nyrmoon and Norman (the latter factor even seems compatible with BonJour's and Lehrer's original diagnosis of the case).

Without some other motivation for the etiological constraint, inferentialist reliabilism thus at best gives us an undermotivated response to the Clairvoyant Challenge.

3.2 A Proper Functionalist Rejoinder

This other motivation for Lyons' etiological constraint might be found by looking at a proper functionalist framework as it is developed by, for instance, Peter Graham (Reference Graham2012, Reference Graham, Fairweather and Flanagan2014).Footnote ⁶ The idea of this framework is that a justified basic belief must be the result of a cognitive system that has reliably producing true beliefs as its function. To have such a function, the system must have a certain consequence etiology: an etiology that “explain[s] why something exists or continues to exist in terms of its consequences, because of a feedback mechanism that takes consequences as input and causes or sustains the item as output” (Graham Reference Graham, Fairweather and Flanagan2014: 18).

One example of such a consequence etiology comes from natural selection. In the envisaged case of natural selection, a system produces a result that is beneficial to the members of a certain species, thereby enabling these members to reproduce. In this way, the systems get replicated in the descendants with the function of again producing the same beneficial thing(s). Note that a system can have several functions simultaneously: many systems will have contributing to survival as one of their functions, but the way in which they contribute to survival (e.g., by pumping blood in the case of the heart) can also be one of its functions (Graham Reference Graham2012: 474).

But natural selection is not the only way in which a consequence etiology can arise. If this were true, then it seems that we could never acquire new ways of gaining justified basic beliefs. And we need not even think of far-fetched cases such as that of Swampman (Sosa Reference Sosa1993) to show that this really is possible: in some actual cases we have already succeeded in artificially providing subjects with new perceptual faculties (Beebe Reference Beebe2004).Footnote ⁷

It is therefore good to think of other ways in which consequence etiologies can arise. One simple example is that of trial-and-error-learning: a certain behavior produces a beneficial result, and so the subject gets reinforced in using that same behavior again with the function of getting the beneficial result again (Graham Reference Graham, Fairweather and Flanagan2014: 31). And the way in which new perceptual faculties are sustained also fits a consequence etiology: after adaptation and learning, these faculties reliably produce true beliefs and are continually used because they reliably produce these beliefs.

This proper functionalist framework thus appears to provide Lyons with what he needs to motivate his etiological condition: only systems that have developed from innate conditions or learning are systems that have consequence etiologies, and so only these systems have proper functions capable of grounding justified basic beliefs. In that case, the Clairvoyance Challenge might be properly answered after all.

However, there is reason to doubt that this proper functionalist story will do the trick. First, there is a question as to why the proper function of reliably producing true beliefs is necessary to produce justified basic beliefs. Graham's response to this question is that it only makes sense to think that a cognitive system has satisfied a certain epistemic norm if that cognitive system also has a proper function on the basis of which one can decide whether the system has functioned in the way it should (Graham Reference Graham2012: 467–8). In the absence of proper function, there is no way a system should behave, and so no sense can be made of the idea that the system has performed in the epistemically right way. And a basic belief can only be justified if that belief is the result of a cognitive process that has performed in the way it epistemically should have.

But one can question whether epistemic norms can only arise from these kinds of teleological considerations. Perhaps we simply apply epistemic norms that we derive from things we value epistemically: e.g., we value truth, so we evaluate belief-forming processes that reliably produce true beliefs positively. This might be sufficient to provide an epistemic norm for such belief-forming processes without appealing to the teleological considerations that Graham favors. The difference would be that the epistemic norms are now no longer fixed by the history of the cognitive systems themselves, but are rather applied to those cognitive systems by us, the people who care about systems that reliably produce true beliefs. Perhaps one might worry that this makes the property of justification ‘unnatural’ or ‘constructed’, but I don't think that such worries are really grounded when one carefully considers what is going on. Even if the epistemic norms for cognitive systems derive from things we value as epistemic agents instead of coming from the cognitive systems themselves, this does not make the properties having to do with these norms any less real.

So, as a first point, there is reason to be sceptical of the necessity of proper function for justification. But if proper function is not necessary for justification, then Norman and Truetemp could still turn out to have justified basic beliefs. More importantly though, there is also independent reason to doubt that the intuitions which ground the Clairvoyance Challenge are really sensitive to considerations having to do with proper functions. Take the following variation on the cases of Norman and Nyrmoon:

Norbert's clairvoyant faculty has a consequence etiology, because its existence is explained by the fact that it has provided a certain benefit to his ancestors. Furthermore, it is reliable, has developed from innate conditions and supposedly is inferentially opaque. Yet it seems that Norbert's belief is still unjustified for exactly the same reason as before: from his own subjective perspective he simply has no reason to trust the visual image that flashes before his eyes. This means that, even if one assumes that a proper functionalist story is correct in general, its diagnosis of the Clairvoyance Challenge remains problematic.

3.3 Other Alternatives

Not only does the case of Norbert work against a proper functionalist diagnosis of the Clairvoyance Challenge, it also works against other contemporary diagnoses, such as those presented by evidentialist reliabilism or agent reliabilism.

Let's begin with evidentialist reliabilism as it has been presented by Juan Comesaña (Reference Comesaña2010). According to this view, roughly, a subject's belief that p is justified if and only if it is based on evidence e, and the type producing a belief that p based on evidence e is reliable. With this account in hand, Comesaña presents the following diagnosis of the original Clairvoyance case:

The case of Norbert is, however, relevantly different from that of Norman. Norbert has a certain experience – a brief visual image flashing before his eyes – on the basis of which he believes that the President is in New York City. Moreover, the connection between having this experience and the President's being in New York City is reliable (at least for Norbert), so he seems to have excellent evidence. Yet we still have the intuition that he should have been more hesitant in accepting the belief he does when considered from his own perspective.Footnote ⁸

Contrast this with an ordinary perceptual case in which we, say, have a visual experience of a table and on that basis believe that there is a table in front of us. Here we have no hesitation in ascribing justification, even though the evidence in this case appears to be similar to that in Norbert's case. So we are in need of a different answer to this modified Clairvoyance Challenge.

In fact, Comesaña's evidentialist reliabilism is not even as good in dealing with the original Clairvoyance Challenge as he himself believes. The case of Norman is somewhat underdescribed, but at least for Truetemp it's clear that there are unconscious representational states (those of the tempucomp) that are reliably connected to the temperature of the environment and which cause Truetemp's beliefs. So why not suppose that Truetemp also has excellent evidence on the basis of which he believes that it is now 104 degrees? More would have to be said about the notions of ‘evidence’ and ‘basing’ to exclude this possibility, and it's not clear that this can be easily made to work without also excluding some of our ordinary perceptual beliefs.Footnote ⁹

A second alternative diagnosis of the Clairvoyance Challenge is provided by Daniel Breyer and John Greco, who believe that clairvoyance cases have to do with the fact that the clairvoyant abilities under discussion are not grounded in an agent's cognitive character:

According to Breyer and Greco, justified beliefs can only be produced by an ability that is properly attributable to an agent, and for that to be the case, the ability needs to be properly integrated in the agent's cognitive character. However, the case of Norbert shows that more needs to be said about the nature of this cognitive integration, for, given that Norbert's clairvoyance faculty has developed from innate conditions inherited from his parents, there seems to be no reason to assume that Norbert does not have a well-integrated cognitive faculty.

Breyer and Greco provide several ways in which one could develop the notion of cognitive integration further, the clearest of which appeal to a subject's reflective endorsement of his beliefs as stemming from a certain reliable cognitive disposition, or to a subject's taking responsibility for his beliefs as his own (Breyer and Greco Reference Breyer and Greco2008: 179–81). Now it's true that Norbert appears to lack such integration, as he does not know where his clairvoyant images stem from: from his perspective, it would be a matter of luck that his clairvoyant beliefs are true. But note that these higher-order requirements are similar to the ones that were presented by BonJour and Lehrer themselves, and so face similar worries: given that unsophisticated agents are unable to have higher-order capacities, these agents would be precluded from having any justified beliefs. If we also assume that justification is a necessary condition for knowledge, unsophisticated agents would thus be precluded from having any knowledge.

This leads me to another of Breyer and Greco's suggestions, which is less clear but at least circumvents the over-intellectualization problem. According to this structuralist suggestion, cognitive integration results either from the coherence of beliefs, or from “the cooperative causal interaction of relevant cognitive dispositions” (Breyer and Greco Reference Breyer and Greco2008: 183). With regard to the latter, it's not really clear why Norbert's cognitive dispositions would not be in cooperative causal interaction. For instance, one can imagine that Norbert starts listening more attentively, or starts moving in a certain direction, whenever he has a clairvoyant image of there being a dangerous situation some distance away. But that does not appear to make his clairvoyant beliefs any more justified.

With regard to the former idea of cognitive integration as arising from coherence, there is the difficulty that this condition might be too weak or too strong, depending on how it is cashed out. Some memorial beliefs might be independent from almost all other beliefs, such as the belief that Hector was slain by Achilles. If coherence requires strong support from other beliefs, then this belief will not cohere very well with other beliefs. This would make the belief cognitively unintegrated, and thus unjustified. On the other hand, if coherence merely requires no inconsistency, then the memorial belief counts as being cognitively integrated and is therefore perfectly capable of being justified. But then so are Norbert's, Norman's, and Truetemp's beliefs. Their beliefs are not inconsistent with other beliefs, they just lack independent support from other beliefs. Yet the account was supposed to show that Norbert's, Norman's, and Truetemp's beliefs were unjustified. This means that agent reliabilism does not provide a good answer to the Clairvoyance Challenge, or at least needs to explain more about the relevant notion of coherence.

I think the most plausible account of coherence, in relation to cognitive integration, will have to say something about how first-order beliefs should cohere with higher-order beliefs to count as being cognitively integrated. For instance, memorial beliefs should be backed up by higher-order beliefs which claim that the memorial beliefs actually stem from memory, a source that is recognized by the agent as one of his own. But if we make such a notion of cognitive integration a requirement for justification, then we are left with the over-intellectualization problem again. In the next section I hope to show how the Clairvoyance Challenge is related to higher-order capacities, without ending up with the over-intellectualization problem. For this to work, I propose that we should think of these higher-order capacities as being able to defeat justification rather than being a requirement for justification in general.

4.1 Defeat in CPR

Let's begin our discussion of a defeat account of clairvoyance by looking at such an account proposed by Goldman himself:

Goldman's idea seems to be that Truetemp acquires a higher-order (undercutting) defeater for his temperature beliefs because he has reason to believe that they are unreliably caused.Footnote ¹² But it's not so clear how this should be reconciled with his own account of defeat, present in clause 3 of CPR. According to this account, defeat requires that Truetemp has an alternative available (conditionally) reliable process that would have led to his not accepting the temperature beliefs. The previous quote seems to cite introspection in combination with some sort of reflection as the relevant alternative conditionally reliable process: if Truetemp were to introspect his temperature beliefs, he would notice that they come out of the blue, and, by reflecting on this fact, he would conclude that they are probably unreliably caused, and, hence, he would not accept them. But if such an extensive reflective process is allowed to provide defeat, then too many of our justifications would count as being defeated. For instance, by reflecting on a certain set of beliefs one might be able to derive a contradiction, thereby being forced to give up a specific belief in the set. But if the derivation is difficult, or just contains many steps, then it certainly is not obvious that the justification is defeated before the derivation is actually carried out.

Although the suggestion by Goldman is on the right track, we need a better way of spelling it out.Footnote ¹³ Instead of using a combination of introspection and reflection as the alternative reliable process, it seems better to assume that we, cognitively sophisticated agents, have largely unconscious monitoring mechanisms in place which check, e.g., the origins of the information we receive and its coherence with other information we possess.Footnote ¹⁴ The idea is that these monitoring mechanisms will reject information that stems from an untrustworthy or unknown source, or information that does not cohere well with other well-founded beliefs. Moreover, the existence of such monitoring mechanisms could also be used to explain why we normally know that we see, hear, or even remember that p in situations where we actually do see, hear or remember that p – something that the reliabilist will have to account for anyway. Higher-order knowledge would on this account be nothing more than the application of a specific sort of cognitive mechanism, i.e., a monitoring mechanism, available only to sophisticated epistemic agents.Footnote ¹⁵

Two things are important to stress about these monitoring mechanisms. First, the cues they use to ascertain the trustworthiness and source of information need not be cues that are accessible to the subject. For instance, the fluency of the cognitive processing involved in retrieving a memory might be used as a cue for the trustworthiness of the retrieved memory (Dokic Reference Dokic, Beran, Brandl, Perner and Proust2012). Just so might a monitoring system be able to identify information as stemming from imagination rather than perception on the basis of the causal antecedents of the information (e.g., which neurons were active, were one's eyes open or closed, etc.) rather than the content of the information itself. Second, the monitoring mechanisms need not lead to higher-order beliefs directly. Perhaps the formation of higher-order beliefs is mediated by noetic feelings of a certain kind, such as feelings of (un)familiarity or feelings of (not) knowing. In the normal case, these feelings could be all that is present to consciousness up until the point when one considers a certain higher-order question, making higher-order beliefs dispositional rather than either occurrent or unconscious.Footnote ¹⁶

Applied to the cases of clairvoyance under discussion, the subjects' monitoring mechanisms should find that the relevant clairvoyant information is not strongly corroborated by any other available information, and that the information also lacks any recognizable trustworthy source. Even though Norbert's belief is based on a visual image, this visual image itself is not well integrated with his other perceptual experiences – the visual image just briefly flashes before his eyes – and it does not stem from any recognizable source, such as perception, memory, or even imagination. This should be enough for the monitoring process to reject the information presented by that visual image.Footnote ¹⁷

However, even if this story about monitoring mechanisms is correct, then it will still be difficult to integrate with CPR's third clause. First, with regard to the reliability of the monitoring mechanisms, given that the mechanisms are supposed to reject beliefs, reliability cannot be accounted for in terms of a tendency to produce true rather than false beliefs. Perhaps this can be solved by focusing on the fact that monitoring mechanisms prevent the formation of false beliefs, while not preventing the formation of true beliefs. Monitoring mechanisms can then be said to be reliable if they prevent sufficiently more false beliefs from occurring than they prevent true beliefs from occurring. But even if such a suggestion is on the right track, there is still a second and third problem left.

The second problem with combining CPR and monitoring mechanisms is that it's not really clear whether the monitoring mechanisms should be seen as additional mechanisms or as parts of our perceptual mechanisms. Either option will lead to difficulties in combination with CPR. If we suppose that the monitoring mechanisms are additional to our perceptual mechanisms, then, given the set-up of Truetemp's case, we can suppose that the monitoring mechanism is being overridden by the tempucomp. This means that Truetemp's belief is the output of a reliable mechanism with no available alternative reliable process to defeat its justification. In other words, Truetemp still comes out as having a justified belief.Footnote ¹⁸ If, on the other hand, we suppose that the monitoring mechanisms are part of our perceptual faculties, then Truetemp, Norman, and Norbert should be seen as using entirely different processes to get at their respective beliefs. But in that case, there would again be no alternative available mechanism that would lead to the rejection of their beliefs.

Now, there might be a way around this second problem by using a rather different notion of availability according to which a cognitive process is available if a subject is generally able to use it (even if the subject is not able to use it in this particular instance). But there is still a more fundamental problem with CPR's defeat clause left: it's not clear that it can be properly motivated from within a process reliabilist framework. Recall that CPR aims to provide a reductive account of justification in terms of reliability. Supposing that justification really is reducible to, or supervenient on, the reliability of a belief-forming process, then why does it matter that beliefs might have turned out differently if another reliable process was used? Consider the following example: I have two textbooks available that are both equally and highly reliable, and I happen to pick one of them to find out more about a certain topic. If the textbook is reliable about its subject matter, and justification is reducible to reliability, then it shouldn't matter what the other, equally reliable and available textbook, would have told me. So it seems that one cannot uphold CPR's defeat clause if one also wants to maintain, with CPR, that justification really is reducible to the reliability of belief-forming processes.

One response on behalf of CPR is that its claim is not that justification is reducible to just the reliability of the actual belief-forming processes, but also to the reliability of processes “… that could and should be employed” (Goldman Reference Goldman and Pappas1979: 20). But this does not yet answer the question as to the motivation of this addition. One such motivation might be that an epistemic agent has not done her epistemic duty if she neglects to use a very reliable method at her disposal, but this just raises a question as to whether reliabilism can then still count as having provided a fully reductive account of justification. Epistemic duties should be explained in terms of reliability, rather than using them as primitives to explain which forms of reliable processes (actual and potential) should be used to account for justified beliefs. What's more, this account of epistemic duty also seems to become overly demanding if an agent is always expected to use two methods if they are equally reliable, instead of just using one of them.

It might be helpful to compare this complaint to one raised by Thomas Grundmann (Reference Grundmann2009). According to Grundmann, CPR's defeat clause is ad hoc as it “fails to explain why internally rational counterevidence removes justification” (Grundmann Reference Grundmann2009: 70). Grundmann seems particularly worried about the fact that this counterevidence can be false and unreliably produced, thereby defeating beliefs that are themselves true and reliably produced. Goldman responds to this worry in part by saying that he doesn't think “it's the responsibility of a principle to explain why satisfaction of the defeater-sensitivity condition should be correlated with improvement in overall truth-ratio, as long as it is so correlated” (Goldman Reference Goldman2009: 248). Now Goldman is probably right in thinking that satisfaction of his defeater-sensitivity condition is actually correlated with an improvement in overall truth-ratio. This could provide a general motivation for the defeat-clause even though forming beliefs in accord with this clause could lead to a lower truth-ratio in particular instances (e.g., cases in which the counterevidence is itself unreliably produced). But this still would not answer the worry I have raised. This worry has to do with the fact that CPR appears committed to holding that a certain amount of process reliability is sufficient for justification, irrespective of whether other processes would increase this reliability even more.

These difficulties provide us with sufficient reason to look at a different account of defeat. I propose that an account in proper functionalist terms goes a long way in circumventing the above problems: such an account can supply a specific proper function for defeat mechanisms, has no problems with accounting for the availability of these mechanisms, and can be properly motivated from within the proper functionalist framework.

4.2 Proper Functionalist Defeat

The idea of this proper functionalist account of defeat makes use of Plantinga's suggestion that we have a “defeater system” (Plantinga Reference Plantinga1993: 40–2) which is aimed at forming true beliefs when functioning properly:

One of the problems with Plantinga's account is that it makes use of the notion of a “design plan” to explain the proper function of cognitive systems. Not only does that create difficulties with the case of Truetemp, where the designed Tempucomp reliably outputs temperature beliefs which are nevertheless unjustified, it also has problems with accounting for justification by newly acquired cognitive processes that lack any design plan. Instead, I prefer the framework provided by Graham (Reference Graham, Fairweather and Flanagan2014) in which proper function has to do with how a system performed when it produced the beneficial effects that led to that system's continuing existence.

Now, as we saw above, the specific proper function that Graham takes to be necessary for producing justified beliefs is the proper function of reliably producing true beliefs. It seems that a similar reliability-related function should be connected to defeat, although it should allow for the fact that defeater systems are in the business of preventing false beliefs rather than producing true beliefs. A good suggestion here is that defeater systems are those systems that have the proper function of reliably preventing the formation or maintenance of false beliefs (without thereby preventing the formation or maintenance of true beliefs).Footnote ¹⁹

In contrast to CPR, proper functionalism can give a good account of the relevance of such defeater systems. Given that the output of our cognitive faculties are influenced by what the defeater systems have assessed as trustworthy, our cognitive faculties are no longer selected purely on the basis of their own merit, but in combination with the merit of the defeater systems. Thus, the consequence etiologies for our cognitive faculties by now seem to include our defeater systems in an important way. This means that these defeater systems cannot be ignored when considering the full justification of our beliefs.

A proper functionalist account of justification can thus make a motivated appeal to defeater systems that have their own proper function relevant to justification. But a question that still remains is how this defeat should be spelled out exactly. According to Michael Bergmann, no beliefs are justified if a subject's defeater systems are not functioning properly (Bergmann Reference Bergmann2006: 170). This, however, seems too strong to work. Recall that on the proper functionalist proposal under consideration, one's perceptual systems have the proper function of reliably producing true beliefs. So even if one's defeater systems are not functioning properly, then one's perceptual systems will still reliably produce true beliefs if they are functioning properly. On Bergmann's proposal, if one's defeater systems are somehow malfunctioning, then such reliably produced true beliefs will not be justified — even if a properly functioning defeater system would not have rejected them. And this appears to be an unwelcome result: perceptual faculties are now deprived of their justificatory power even if the beliefs they produce are perfectly fine. That is why I propose a slightly weaker Proper Functionalist Defeat (PFD) clause, in line with suggestions by both Grundmann (Reference Grundmann2009) and Bedke (Reference Bedke2010):

Note that this proper functionalist defeat clause no longer encounters the availability problem of CPR's defeat clause. A belief will be defeated if there is a defeater system present that would have resulted in the rejection of this belief if it had been working properly. So even if the defeater system is not actually available due to circumstances that prevent it from doing its work, what matters is what the defeater system would have done if it had been functioning properly.

PFD also allows for cases in which beliefs are both true and reliably produced, and yet still defeated because they would not have been formed if the defeater-system had been working properly. Indeed, as we will see in the next section, this is exactly what goes on in Clairvoyance-style cases. But according to a proper functionalist account, this is exactly how it should be, given that defeater systems have become a part of our total cognitive system. Even though defeater systems might get it wrong in a particular case, their contribution to our overall cognitive structure is still important enough to validate their decisions even in those rare cases.

Before going back to the diagnosis of Clairvoyance-style cases, let me also mention that the current approach to defeat is one that Jonathan Kvanvig (Reference Kvanvig and Baker2007) has labeled as a ‘backdoor theory’ of defeat. Instead of characterizing defeat in terms of propositional contents that arrive at the doorstep of the cognitive system, such a backdoor theory characterizes defeat in terms of beliefs that are kicked out of the backdoor. Kvanvig specifically targets Plantinga's (Reference Plantinga2000) account of defeaters, beliefs or experiences that defeat the justification of other beliefs. According to Kvanvig, this account will have difficulties accounting for, first, defeater-defeaters, and, second, the several rational possibilities one has in responding to defeaters, as a backdoor approach only looks at belief rejection. Instead, the propositional approach to defeat is supposed to fare better, as this can provide simple and illuminating definitions of defeaters and defeater-defeaters in terms of what is (no longer) supported by the evidence:

However, process reliabilists and proper functionalists have good motivations not to accept this picture of defeat. Propositional relations by themselves won't provide a satisfying theory of defeat, as what counts as a defeater (or evidence for that matter) for a belief that p for one agent is not the same as what counts as a defeater (or evidence) for a belief that p for another. One agent might see many consequences of his beliefs more easily than another; this means that this former agent will encounter inferential defeat more easily than the latter. That is why CPR focuses on available reliable processes, and why proper functionalism should also focus on possibly individually developed proper functions of defeater systems: to account for the possible differences in defeat between different epistemic agents. So process reliabilists and proper functionalists have good reason to focus primarily on the cognitive processes by means of which beliefs are defeated, rather than the propositional contents of defeating and defeated beliefs. This will make the notions of ‘defeater’ and ‘defeater-defeater’ only of secondary interest as they tend to abstract away from the specific cognitive processes of specific agents. What's more, if a subject S's defeater system would not have led to the rejection of a certain belief B (but would have opted for a different rational response), then B simply is not defeated for S according to the theory, even if the defeater system of another subject S′ would have rejected B. This again underlines the idea that defeat can be different for each subject, and that it is an idealization to speak of defeaters independently of specific subjects.

4.3 Back to Clairvoyance

Let's apply the foregoing considerations to the discussed cases of clairvoyance. As a first point, notice that the earlier proposed monitoring mechanisms count as defeater systems as they have been explained above. The described monitoring mechanisms check incoming information for trustworthiness and coherence, and have presumably been selected because this reliably prevents false beliefs from occurring. They thus have what it takes to constitute a defeater system in the proper functionalist sense.

Second, I have argued above that appealing to a lack of proper function is not adequate to explain Clairvoyance-style cases, as, first, it is not so clear that proper function is absolutely necessary for justification, and, more importantly, there can be Clairvoyance-style cases (like that of Norbert) in which proper function is present but justification is still intuitively absent.Footnote ²¹ Fortunately, an appeal to defeat in the proper functionalist sense appears to get us exactly what we want.Footnote ²²

Given that monitoring mechanisms are a specific kind of defeater system, to see whether Norman, Truetemp, and Norbert's beliefs are defeated we need to look at what their monitoring mechanisms would have done if they had been functioning properly. It certainly appears plausible that their monitoring mechanisms would have rejected their respective beliefs had they been functioning properly: the information presented by their special senses is not corroborated by any of their other senses, nor does the information stem from a recognizable trustworthy source. This should give their monitoring mechanisms sufficient cause to prevent the information from rising to the status of belief. Now, it's true that in the case of Truetemp it's stipulated that the tempucomp leads to the acceptance of his beliefs. In the current framework that would be interpreted as the tempucomp's overriding Truetemp's monitoring mechanisms. But in that case the proper functioning of these monitoring mechanisms certainly seems to be impacted negatively: in such a case the monitoring mechanisms are definitely not functioning in the way they were when they got selected. And if this is true, then what is relevant is what they would have done had they been functioning properly, namely: preventing the occurrence of these precise temperature beliefs.

Finally, note that this answer to the Clairvoyance Challenge does not have the unwanted consequence that unsophisticated animals or small children lack justification for their perceptual beliefs – although the explanation of this fact is slightly different for each of these groups of agents. Given that monitoring mechanisms are not necessary for perceptual justification (as they are only necessary for a specific kind of defeat), even unsophisticated animals which lack such mechanisms can still have justified beliefs. Small children, in contrast, presumably have monitoring mechanisms that have not yet developed fully. This means that some beliefs that are defeated for adults will not be defeated for these small children. Indeed, if Norbert's clairvoyant faculty would already have been active during his early childhood, then these early clairvoyant beliefs might well be justified.Footnote ²³ However, this is not a weakness but rather a strength of the theory, as this accords with our intuitions about such matters: given that adults have more sophisticated cognitive capacities than small children, adults have to satisfy higher demands to have ultima facie justified beliefs.