The new evil demon problem at 40

I shall use 'epistemic warrant' and 'epistemic justification' interchangeably for a normative property that provides a good route to true belief and knowledge.1 Epistemologists disagree over their relationship. In all possible worlds W, a belief is prima facie epistemically warranted in W if and only if (to the extent that) the psychological processes that caused or sustained the belief reliably produces true beliefs in W. So-called "internalists" disagree. They deny that being a reliable route to truth is necessary for warrant, relying on the "New Evil Demon" scenario to make their case.3 Here's the standard version:4 Norma is an ordinary human adult, with normally functioning perceptual and cognitive capacities, in her normal environment. Walking around the park, she forms reliably true perceptual beliefs about her environment. Perception, in her world, is a reliable route to truth and knowledge. Vicki is Norma's psychological duplicate in another possible world. From the inside, as it were, everything looks exactly to Vicki as it does to Norma. But Vicki is not walking around a park. Instead, she is floating in a vat of nutrients, where her sensory systems are hooked up to a massive super-computer. The computer deceives Vicki by inducing type identical perceptual representations to Norma's. That's why everything (in perception) looks the same to Vicki as it does to Norma, but nothing in Vicki's immediate surrounds is as it seems. Vicki, unaware of and unable to detect the deception, forms massively false perceptual beliefs. Perception, in her world, is not a reliable route to truth and knowledge. Internalists conclude that Vicki's perceptual beliefs, though massively in error, are nonetheless just as justified— "equally justified"—as Norma's. Taking perceptual experience at face value, given no reason to suppose it is not a good guide to truth in the circumstances, is just what Vicki is supposed to do. That's the correct or proper response to her perceptual experiences. If Vicki's perceptual states justify her perceptual beliefs, simple reliabilism doesn't stand a chance. That's why it's called the New Evil Demon Problem. Reliability has nothing to do with epistemic justification… Beliefs are justified because the believer is "reasoning correctly" (in a broad sense of "reasoning" where perceptual experience are reasons in an extended sense and so perceptual belief formation is a kind of reasoning). If one makes all the right epistemic moves, then one is justified regardless of whether the belief is false, or nature conspires to make such reasoning unreliable. (Pollock and Cruz, 1999: 113–114). If justification strongly supervenes on an individual's (nonfactive) mental states and relations, as the internalist supposes, but the reliability of perceptual beliefs is a contingent matter of fact, then justification as such has nothing to do with reliability or truth. Since our first fact is necessarily true, says the internalist, the second fact is irrelevant to epistemic justification. That's the moral, according to the internalist, of the New Evil Demon Problem. No wonder so many internalists so easily dismiss reliabilism with a one-liner. We owe the New Evil Demon Problem to Stewart Cohen. Cohen discovered the problem while a graduate student at the University of Arizona, at one time the center of the epistemological universe. Cohen first published the problem in a paper co-authored with Keith Lehrer in 1983 (Lehrer and Cohen, 1983) and then developed it further in his landmark paper "Justification and Truth" (Cohen, 1984; cf. Pollock, 1984; Foley, 1987). The New Evil Demon Problem is now over forty years old. Reliabilists remain undeterred. Here are two ways reliabilists have responded. I call the first, following Ernest Sosa (2003), the "heroic" response. According to the heroic response, internalists are just mistaken that Vicki has justified beliefs. This response starts with the force, role, or job of epistemic warrant: to be a good route to truth and knowledge. That's what makes epistemic justification epistemic. Justification as such then strongly supervenes on the reliability of the belief-forming process. But since perception is only contingently reliable (as the demon world case vividly illustrates), perceptual justification does not strongly supervene on an individual's (non-factive) mental states and their relations. Internalists, the idea goes, simply misunderstand the role or force of epistemic justification. Vicki's perceptual beliefs are then obviously not justified, for they don't stand a chance of being true. Philosophers taking this line include Sanford Goldberg (2012a),5 Thomas Senor (2013), and Jack Lyons (2013).6 I shall call these three philosophers, and their like-minded colleagues, "our heroes." Our heroes, though they deny that Vicki's beliefs are epistemically justified, usually grant that she is being rational in so believing, that her perceptual beliefs are rationally held (Meyers, 1988; Lyons, 2013). That, they think, often explains the internalist's "intuition" that Vicki's perceptual beliefs are just as "justified" as Norma's, for so many internalists tend to use 'justification' and 'rationality' interchangeably (e.g. Cohen, 1984: 283; Foley, 1987; Huemer, 2001: 19; Smithies, 2012: 274; Wedgwood, 2012: 280; Dogramaci, 2015: 777; cf. Siscoe, 2021; Sylvan, ms.). But once we separate epistemic justification (in the sense of a good route to truth and knowledge) from structural rationality, we should not be so quick to judge that Vicki's beliefs are epistemically justified.7 That's the first response. I call the second the "special-circumstances" response. According to this response, Vicki's beliefs are epistemically justified, even though her perceptual representations are not de facto—"in situ"—reliable. They are epistemically justified because her perceptual representations (her perceptual belief-forming processes) are reliable in a special set of circumstances. It's because they are reliable in special circumstances, the idea goes, that they continue to contribute to the epistemic warrant for perceptual beliefs even when the individual is not in special circumstances. Because Vicki's perceptual belief-forming processes are reliable in special circumstances, the idea goes, they produce epistemically justified beliefs even while Vicki is floating motionless and massively deceived in a vat of nutrients. I shall spend most of the paper on this response. For this response to work, it must at least specify "correct" necessary and sufficient conditions for epistemic justification. But it must also specify conditions that are appropriately deep or revelatory. In other words, this response must not only correctly predict that Vicki's perceptual beliefs are justified, but it must also explain why that should be so. Three of the most influential philosophers of the last fifty-years—Alvin Goldman, Ernest Sosa, and Tyler Burge—have all pursued versions of the special-circumstances response. It's the party line. In the first half of the paper, I will criticize their respective versions of this response. In the second half I will develop my own. The point of the second part is to develop the most plausible reliabilist account of warranted beliefs in demon world scenarios. I shall return to our heroes when concluding. Over a thirty-three-year period (1979-2012), Alvin Goldman published several different responses to the New Evil Demon Problem. I've spelled them all out elsewhere (Graham, 2017), and so I won't do so again here. Instead, I will focus on two of his versions of special circumstances reliabilism: Actual world reliabilism. In all worlds W, a belief B is prima facie justified in W iff formed or sustained on a process that is reliable in the actual world. This prevents demon-world cases, for perception is reliable in the actual world. (Goldman, 1979; 1999/2002; 2011/2012). Normal worlds reliabilism. In all worlds W, a belief B is prima facie justified in W iff formed or sustained on a process that is reliable in normal worlds. Normal worlds are worlds that share the general characteristics we believe true of the actual world. This rules out wishful thinking and clairvoyance counterexamples (they are not reliable in normal worlds) and the demon-world case (they are paradigm cases of non-normal worlds). (Goldman, 1986: 107, 113) Which general beliefs count for determining normal worlds? There seem to be too many choices. Whichever ones we select, it looks like dramatically different worlds might fall in the class of normal worlds. Does justification turn on reliability in all of these worlds? Who is the "we"? All humans ever? Only contemporary English-speaking epistemologists? Does the referent change over time? Perhaps it should come as no surprise that Goldman credits Stewart Cohen for imagining this example. After offering this case, Goldman concluded that "it seems wise to abandon the normal-worlds version of reliabilism" (Goldman, 1988: 62). What about actual world reliabilism? Does it avoid the demon world case? Extensionally it does, on the assumption that the actual world is not a demon world. But even so, like the normal-worlds theory, it won't do. For ALIEN shows it can't be right either. For the clairvoyance of these alien creatures does not exist in the actual world, and if it were it to exist in the actual world (per impossible), it would not be reliable. Surely there are possible warrant conferring processes that are not reliable in the actual world: possibility transcends actuality. Since there are possible belief-forming process that render beliefs warranted and knowledge even when those processes are not (and would not be) reliable in the actual world, it seems just as wise to abandon the actual-world reliability theory as it is to abandon the normal-worlds theory. Neither theory states correct necessary and sufficient conditions.8 J-APT (For all w) B is apt-justified in w (if and) only if B is acquired in w through the exercise of one or more intellectual virtues that in that world w virtuously would produce a high ratio of true beliefs. J-ADROIT (For all w) (B, in w, is adroit-justified in w (if and) only if B derives in w through the exercise of one or more intellectual virtues that in our actual world virtuously would produce a high ratio of true beliefs. "If one were a demon's victim, one's intellectual performance might still be adroit (adroit-justified). That is to say, in a world w in which one was a victim, one's beliefs acquired through our normal perceptual faculties would come out of processes that in our actual world are reliable, and hence those beliefs of ours in that world would be adroit (adroit-justified)." (2003: 161, emphasis added) What we've learned from the New Evil Demon Problem, Sosa concluded, is that we should "…distinguish between two sorts of justification: aptness and adroitness" (Sosa, 2004: 284; see also Comesaña, 2002; 2010; cp. Ball and Blome-Tillman, 2013). Adroit-justification is the sort of justification, "within truth-connected epistemology," that one "might still have even if unlucky circumstances make your belief-forming mechanisms unreliable" (2003: 161). He concludes that his "two sorts of justification" strategy is "secure against the new evil demon problem" (Sosa, 2003: 158). J-ADROIT uses the phrase 'in our actual world.' For Sosa it functions as an indexical. 'Our actual world' refers to the world where the phrase is uttered (or where the corresponding thought is thought). In any word where 'the actual world' or 'our actual world' is uttered (or the corresponding thought is thought), it refers to the world where the phrase is uttered (or thought), just as 'here' which (qualifications aside) refers to the location where the phrase is uttered (or thought) (Sosa, 2003: 157). Whether a belief is justified (adroit-justified) is then relative to the world of the attributor. When we say (or think) 'Vicki's beliefs are justified', we express a thought that is true if and only if Vicki's perceptual beliefs are formed by processes that are reliable in our actual world. And since those processes are reliable in our actual world, what we say is true, as already noted. That was Sosa's treatment of the problem prior to the development of his now well-known "AAA" virtue epistemology (Sosa, 2007; 2009; 2011; 2015; 2017; 2021). In his AAA virtue epistemology, Sosa sees the epistemic evaluation of beliefs as an instance of the more general normative evaluation of exercises of skills or competencies, of performances. Sosa defines skills or competences as dispositions to achieve a desired (or desirable) or valued (or valuable) effect. A competence is not just a disposition to produce an effect (that's why sugar does not have the competence to dissolve in water) but rather a disposition to produce an effect "aimed" at. A skilled archer is not only disposed to hit the target when she shoots, but she (literally) aims at hitting the target. She desires or values hitting the target. That's why her disposition to hit the target is a competence. Skills or competencies can then be evaluated along (at least) three dimensions: accuracy, adroitness, and aptness. Consider still the archer: Accuracy. An archer's shot is accurate iff it hits the target. Adroitness. An archer's shot is adroit iff it derives from a skill that is reliable (enough). Aptness. An archer's shot is apt iff its accuracy manifests adroitness (iff the shot is accurate because adroit). Sosa then applies this "AAA" structure to evaluations of beliefs. A believer's belief is accurate iff it is true (for belief or the believer, or the belief-forming competence aims at or functions to produce true belief). A believer's belief is adroit iff it derives from a belief-forming process (or virtue or skill or competence) that is reliable (enough). A believer's belief is apt iff it is true because adroit, only if its accuracy manifests adroitness (Sosa, 2009: 37). Sosa thereby repurposed his words. Apt now entails adroit, for an apt belief just is a true belief that manifests adroitness. Pre-2007 for Sosa, "apt" did not entail "adroit." Before, a belief that was apt-justified need not be adroit-justified, for a belief-forming process might be reliable in the world of its use without being reliable in our actual world. Post-2007, there is no such thing as an apt belief (in Sosa's jargon) that isn't also adroit. For Sosa post-2007, epistemic justification is adroit-justification, as defined. Despite this and many other developments (see note 24), Sosa's "indexicalist" answer to the Demon World Problem from 2001 to 2021 fundamentally remains unchanged (see Sosa, 2021: 198). Sosa thereby advances a version of "actual-world reliabilism" to solve the demon world problem. I just criticized Goldman's actual world reliabilism on the grounds that it can't handle the ALIEN example. Does Sosa have a way out? Prima facie he does. For on Sosa's view, when one of the aliens says (or thinks) 'my clairvoyance beliefs are (mostly) justified' what they say (or think) will be true. For, according to Sosa's proposal, the thought the alien expresses will be true if and only if their clairvoyance beliefs are reliably true in their actual world. And since ex hypothesi they are reliably true in their world, they are justified. Counterexample diffused. But Sosa's "indexicalism" comes at a price. For consider what we express when we say (or think) 'their clairvoyance beliefs are epistemically justified'. On Sosa's view, we express a thought that is true if and only if their clairvoyance beliefs are reliably formed in our actual world. But since, ex hypothesi, their clairvoyance beliefs are not reliably formed in our actual world, the thought we express is false. So on Sosa's view when they say 'our beliefs are justified' the thought they express is true, but when we say 'their beliefs are justified' the thought we express is false. Does that sound right to you? I think when we say (or think) 'the alien's clairvoyance beliefs are justified' we express something true, not false. I think Sosa's view makes the wrong prediction about the truth value for our talk and thought about the aliens, even if Sosa's view makes the right prediction about the truth value of their talk and thought about themselves.9 Or consider the thought that Vicki would express on Sosa's view if Vicki were to say (or think) 'my perceptual beliefs are justified.' On Sosa's view, when Vicki says 'my perceptual beliefs are justified' the thought she expresses is false, for Vicki is in a demon world, and perceptual beliefs are not reliably formed in her actual world, the world where she actually is. Does that sound right to you? It does not sound right to me, nor to other (non-heroic) reliabilists like Goldman or Burge. I think when Vicki says (or thinks) 'my perceptual beliefs are justified' she expresses something true, not false, even if Sosa's view makes the right prediction of the truth value of our talk and thought about Vicki's perceptual beliefs.10 Even though, in one way, Sosa gets the right results about the demon world case and the alien case, in another way he gets the wrong results in both cases. Out of the frying pan and into the fire, as the saying goes. I now turn to Tyler Burge's treatment of the problem from his paper "Perceptual Entitlement" (2003), a treatment he returns to in "Entitlement: The Basis of Empirical Epistemic Warrant" (2020). Burge's view involves a switch from reliability in the actual world (as in Goldman and Sosa) to reliability in normal conditions. What do we all already know about normal circumstances? Intuitively, normal circumstances for our perceptual capacities are just a fraction of the actual world; not all actual circumstances are normal circumstances. For the actual world is a vast Universe or even a Multiverse, possibly infinite in space and time. So just because you are in your actual world, it does not follow that you are in normal circumstances. Relatedly, normal circumstances for our perceptual capacities transcend the actual world. A human being could then, as it were, be in normal circumstances for perception even if she were not in our actual world but in some other possible world. Normal circumstances are both less than and more than actual circumstances. Reliability in normal conditions is not the same thing as reliability in the actual world. Burge has a theory about what makes circumstances normal for our perceptual states and perceptual systems. On his view, normal conditions for our perceptual states are those conditions in explanations for how our perceptual state types acquired their representational contents (Burge, 1993; 2010). According to his perceptual anti-individualism, a fundamental way perceptual systems acquired their representational states was in virtue of patterns of causal interaction between the perceptual system (partly through its influence on the individual's behavior), on the one hand, with instances of the properties and relations that the perceptual states types of the perceptual system represent, on the other, interactions that help explain how the individual successfully navigates its broader environment. For Burge, normal conditions for our perceptual systems are then those conditions, and conditions of a similar type, that enter explanations for how perceptual systems acquired their perceptual state types and so their capacities to represent various properties and relations in the broader physical environment. Normal conditions are formative conditions.11 In all possible circumstances, a normally formed perceptual state token contributes to a prima facie epistemic warrant to a perceptual belief appropriately derived from that token perceptual state if and only if the normally formed token is of a type that is reliably accurate in normal circumstances. Normal circumstances reliabilism marks an advance over the accounts offered by Goldman and Sosa. Consider the demon world case. Norma's perceptual states are reliable in normal conditions. Hence, normal circumstances reliabilism deems that they contribute to a prima facie warrant for her perceptual beliefs appropriately derived from those states. Vicki's perceptual states are also reliable in normal conditions, even though she is, unbeknownst to her and undetectably so, no longer in normal conditions. Even so, normal circumstances reliabilism deems that her current perceptual state tokens contribute a prima facie warrant for her current perceptual beliefs. So normal circumstances reliabilism gets the right result for the demon world (Burge, 2003: 507, 509, n. 6). And since what counts as normal conditions for a perceptual system does not shift as the circumstances of assessment shift, this view avoids exactly what seems to be problematic about Sosa's view. Consider next the aliens. These possible beings possess a psychological capacity of clairvoyance that we assume to be reliable in its normal circumstances and they currently reside in such conditions. Normal circumstances reliabilism deems their beliefs so formed prima facie warranted. We can even imagine one of these aliens whisked away to non-normal conditions–possibly by a deceptive scientist or an evil demon–where the alien would have no idea, and could not detect, that they were no longer in normal conditions. Normal circumstances reliabilism deems their beliefs in abnormal conditions–just like Vicki's perceptual beliefs–prima facie warranted as well. Again, this view avoids what seemed problematic about Sosa's view of our assessment of their beliefs, whether in their normal conditions or transplanted into abnormal conditions. Goldman and Sosa get the extension wrong; normal circumstances reliabilism gets it right. If we had to choose between the responses from Goldman, Sosa and Burge, Burge's view has the upper hand.12 But we are not done yet. Even if reliability in normal circumstances is the right kind of reliability to get the extension of warrant right across possible cases (stating correct necessary and sufficient conditions), we still face the explanatory challenge. Even if we know that a normally functioning perceptual competence contributes to prima facie perceptual warrant in all possible circumstances if it is reliable in normal conditions, we can still ask why this should be so. For, provided we conceive of warrant as a good and so reliable route to truth and knowledge, we might naturally ask why any perceptual belief should be warranted at all in demon world scenarios. After all, in a demon world, the individual's perceptual states are systematically inaccurate and so the individual's perceptual beliefs are systematically false. How could an individual's perceptual states serve as good routes to truth and knowledge while the individual is massively deceived, floating in a vat of nutrients and hooked up to a massive supercomputer? Why shouldn't we conclude with our heroes (that because perception is not a good route to truth and knowledge in the vat) that the "inhabitants of the demon world are unwarranted in most of their beliefs"? If our heroes are mistaken, if perceptual warrant persists in demon worlds because the perceptual states are reliable in normal conditions, we should want to know why this is so. Why does reliability in normal conditions explain why warrant persists outside of normal conditions? One can ask why should reliable success in … the normal environment bear on the contribution of perceptual states to entitlement in abnormal environments that the individual may have landed in … for it is the individual at a given time for which warrant is supposed to provide a route to truth … In many such environments, a perceptual state or competence … would not be a good guide to veridicality or truth … Why does the contribution of the normal environment carry over to other environments that the individual might form perceptual beliefs within? (Burge, 2003: 533, emphasis added) In the rest of this section, I shall discuss two answers to this question that I find inadequate before developing my own. The first I've heard many times in conversation and can be found in print (e.g., Sosa, 2015: 103; the first quote from Sosa in section three; and Gerken, 2018: 116, 119). The second answer is from Burge (2003; 2020). Here is the first: the reason why a perceptual state contributes to warrant in the demon world, when the perceptual state type is reliable in normal circumstances, is because if the perceptual state were in normal conditions, then the perceptual state would be a good, reliable route to truth–the perceptual state would then and there objectively raise the probability of the truth of the belief appropriately derived from the perceptual state. In short, perception contributes to warrant in the demon world because perception, if it were in normal circumstances, would be reliable. Perception contributes to warrant here (in the demon world) because it would be reliable there (in normal circumstances). I don't find this answer satisfying. I have a variety of bills in my wallet. Some are from the United States. Some are from my last trip to Glasgow. And some are from South Korea from my time teaching there. If we were in Scotland, I could buy you coffee with my RBS notes. But here I am in New Jersey, where the cafes won't take them as legal tender. (I'm told taxi drivers won't take them in England either.) Does the fact that I could buy you coffee in the Dear Green Place make my notes valuable in the Garden State? I don't see how. Or suppose the notes in my wallet were in a currency that no longer existed, and there were no collectors who would buy them from me in a currency that I could use. Do those notes have monetary value, just because, if I were to travel back in time, I could use them to buy lunch? Hardly. Being valuable elsewhere doesn't suffice to explain why they should be valuable here. So, I don't see why the fact that if a belief-forming process were in circumstances where it would be reliable explains why it contributes to justification in the world where it is, especially when the process is not reliable where it is. I don't see why reliability in normal circumstances explains why warrant persists in demon worlds. How is this first answer supposed to answer our question? Indeed, according to our heroes, all this move amounts to is that our beliefs would be justified if we were in normal conditions, not that they are justified outside of normal conditions. More needs to be said to make the appeal to normal circumstances work. This privilege entails the irrelevance to perceptual warrant of reliability and unreliability in other conditions. So it extends to indiscernible abnormal environments that the individual might contingently find himself in. (Burge, 2003: 536) The application of these ideas to demon world scenarios is straightforward. Individuals who are moved unawares from normal environments to matrix-type situations still have warranted perceptual beliefs, even though they are reliably mistaken in their new environment. Being moved from one situation to another does not in itself alter epistemic warrant. (Burge, 2020: 87) That's Burge's answer to our question, why should warrant persist in demon worlds? Unfortunately, this conclusion based on (1)-(5) strikes me as a complete non-sequitur. I just don't see it. Because warrant requires reliability in normal circumstances, warrant persists even outside of normal circumstances where reliability lapses? Why does that follow? Imagine two interlocutors for Burge. The first internalist interlocutor denies that reliability in any conditions matters to warrant. Truth and reliability, this philosopher thinks, has nothing to do with warrant. (Recall Pollock and Cruz.) The demon world case shows there are circumstances where reliability does not matter to warrant, so this philosopher concludes that reliability does not matter to warrant in any circumstance. The second interloctuor agrees with Burge that reliability is not just necessary for warrant but reliability in normal conditions is required for warrant. This interlocutor agrees that normal circumstances are privileged in the explanation of perceptual warrant. But this interlocutor takes the heroic line and denies that warrant persists in demon worlds. Perceptual warrant, this interlocutor thinks, is then "bounded" or "restricted" to normal conditions. Outside of normal conditions, this interlocutor says, warrant lapses. Outside of normal conditions, perceptual states are not a good route to truth and knowledge, and so cannot contribute to warrant. That's the idea. Does the explanation just offered by Burge answer this second philosopher? I don't see how. That's why I say his reasoning strikes me as a complete non-sequitur. Burge's reasoning makes more sense to me as a reply to the internalist's denial of the relevance of reliability. "Look," they say, "here in the demon world there is warrant but no reliability. Hence reliability is irrelevant to warrant in any circumstance." "No," Burge replies, "all you have pointed out is that a process that warrants belief need not (or even cannot be) reliable in all possible circumstances, especially abnormal circumstances. That does nothing to show that reliability in some circumstances–especially normal circumstances–is not necessary for warrant." There's clear textual evidence that Burge has this first interlocutor in mind when giving his answer to our question. Just before his answer he writes that "a motive for…specifying what…normal situations are…was to answer the claim that…demon scenarios refute the claim that reliability is a necessary condition on epistemic warrant" (Burge, 2020: 85). And then just afterwards he concludes that "demon world scenarios should not