1 Introduction

It is rare that we find epistemologists (as opposed to philosophers of mindFootnote 1) defending specific views about the metaphysical nature of knowledge-apt representational states. However, mainstream work in epistemology—especially on the nature of knowledge—almost invariablyFootnote 2presupposes a certain background picture of cognition that fits very well with the idea that knowledge is necessarily stored and generated in the head.Footnote 3

In the philosophy of mind and cognitive science, this background picture has a name—cognitive internalis—the view that, necessarily, cognition supervenes on brainbound, biological properties of the cogniser.Footnote 4 Cognitive internalism is, as Fred Adams and Kenneth Aizawa (2009) put it, tantamount to a dictum of commonsense—viz. that the ‘mind is in the head’.

When articulated as a view about cognitive processes (i.e. memory storage and retrieval), cognitive internalism is usually read as maintaining that, necessarily, cognitive processes (e.g. memory storage and retrieval) play out entirely inside the headFootnote 5; alternatively, such processes are materially realised exclusively by physical processes in the brain. When framed as a view about states of cognition (e.g. beliefs), the view implies that your beliefs are literally in your head, in the sense that the physical subvenient bases of your beliefs are all and only intracranial subvenient bases.Footnote 6

Here is perhaps the most straightforward picture of how cognitive internalism is so easily ‘smuggled in’—and uncritically so—as a presupposition in epistemology. Propositional knowledge—of central interest in epistemology—is assumed ex ante to entail belief, truth and justification, as per the traditional JTB analysis.Footnote 7 The project of analysing knowledge—which dominated the second half of twentieth century epistemology—aimed to work out how these three (and perhaps other) conditions relate to each other when one has knowledge. Belief relates to the other two conditions—at least, in a way that matters for analysing knowledge—in so far as beliefs are propositional attitudes with a representational (i.e. mind to world) direction of fit. That is, after all, what really matters in the analysis of knowledge, because it is exactly this kind of a thing that is capable of being true and justified and thus, as the thought goes, capable of being known. And furthermore, in at least paradigmatic cases of propositional knowledge (think of simple perceptual knowledge—viz. your knowledge that there is a hand in front of you, which you generate and then retain in memory), it seems plain enough that brainpower is going to be both necessary and sufficient to (i) generate the (occurrent) propositional attitude with a representational direction of fit (i.e. that there is a hand in front of you right now) and then to (ii) store it, as a dispositional belief, in memory.

Thus, the ‘map’ to unearthing the cognitive internalist presupposition in mainstream thinking about knowledge is accordingly a pretty direct one, with two key ‘links’ in the chain: the first that gets us from knowledge to belief (via the ‘entailment thesis’ that knowledge entails belief) and the second that gets us from belief to cognitive internalism (where the latter is the assumed picture about how the former is realised and maintained).

In what follows, I am going to challenge both links of this chain. Or, more carefully, I am going to show that despite initial appearances, there is really no plausible construal of the first link (that knowledge entails belief) that should pressure epistemologists who care about knowledge to assume a cognitive internalist picture of the mind.

Here is the plan. §2 clarifies the sense in which knowledge entails belief, and in doing so defends a dispositional reading of the entailment thesis. §3 then puts pressure on cognitive internalism generally, and then more specifically on the idea that such a view would explain any better than cognitive externalism the datum that knowledge entails dispositional belief. Taken together, §§2–3 challenge some of the dogmas of traditional epistemology that seem to block progress on certain questions in extended epistemology—e.g. epistemological questions whose answers make reference to transcranially supervenient cognitive processes and states.

2 Knowledge and belief

Whilst the idea that knowledge entails belief is widely assumed,Footnote 8 it is rarely argued for positively (apart from being defended against objectionsFootnote 9), with two notable exceptions being G. E. Moore (1962) and Keith Lehrer (1968). Moore famously tried to show that knowledge entails belief via a (albeit somewhat odd) linguistic test, and Lehrer (1968) opted for a proof aimed at showing that knowledge formally entails belief. Neither is promising.

According to Moore:

There certainly is a common use of belief in which ‘I believe’ entails ‘I don’t know for certain’. Is there another in which ‘I know for certain’ entails ‘I believe’? One reason why it seems so is because ‘I thought I knew’ entails ‘I believed’ (1962, 115).

It does seem plausible that a speaker who says ‘I thought I knew’ that p is committed in some way to accepting that they believed that p. But let us simply grant for the sake of argument that patterns like the one Moore mentions constitute linguistic evidence that knowledge entails belief (either occurrent or dispositional). Even on this charitable assumption, there is, as Carolyn Black (1971) has observed, also linguistic data that would seem to support the very opposite conclusion. Take for example, this case: ‘I say that my books are in my office.’ You ask ‘Do you believe that your books are in your office?’ I say ‘No. I know that my books are in my office’ (Black, 1971, 155–56, my italics). The felicitousness of this kind of exchange is a problem for arguments that attempt to establish that knowledge entails belief (of any sort) simply on the basis of our patterns of using the words ‘knows’ and ‘believes’.Footnote 10

So what about Lehrer’s (1968) proof? Here is the proof, which he takes to be sufficient to establish to a doubter that knowledge entails belief.

  1. 1.

    If S does not believe that P, then S does not believe that he knows that P.

  2. 2.

    If S does not believe that he knows that P, then, even though S correctly says that P and knows that he has said that P, S does not know that he correctly says that P.

  3. 3.

    If, even though S correctly says that P and knows that he has said that P, S does not know that he correctly says that P, then S does not know that P.

  4. 4.

    (Therefore) If S does not believe that P, then S does not know that P (1968, 498).

There are problems with both premises (1) and (3). The problem with (1) is that it is either false or at best questionbegging, given what Lehrer was attempting to do here. Just consider that the kind of opponent Lehrer is out to convince might very well think that “S knows that p” is compatible with the antecedent of (1). But then, (1) comes out false if S does not believe that p because S knows that p, given that, on that supposition, it is possible that S will believe that S knows that p. But even if this problem with (1) could be dealt with, there are independent problems with (3): just suppose your friend tells you they do not know all the lines of a certain poem by William Blake, but then (after telling you this) they proceed to recite the poem perfectly; this seems like a plausible case where—even though they do not know that they have correctly recited it—they nonetheless know the lines.Footnote 11 They had them mastered better than they had thought. This assessment of the case, however, is incompatible with (3).

Interestingly, we do not find many other attemptsFootnote 12 to positively establish the widely held assumption that knowledge entails belief (and, as it turns out, Lehrer himself abandoned his own proof later,Footnote 13 opting instead for a view on which knowledge entails not belief but acceptance).

Instead, what much of the literature on the knowledge-belief ‘entailment thesis’ concerns is whether outlying attempts to challenge the thesis are sound. The most widely discussed case on this score—also one that involves (contentiouslyFootnote 14) a kind of ‘knowledge-with-lack-of-confidence’ structure—is due to Colin Radford (1966):

UNCONFIDENT EXAMINEE: Kate is taking a history test. She had studied carefully and has been doing well on all the questions so far. She has now reached the final question, which reads “What year did Queen Elizabeth die?” As Kate reads this question she feels relief, since she had expected this question and memorized the answer. But before Kate can pause to recall the date, the teacher interrupts and announces that there is only one minute left. Now Kate panics. Her grip tightens around her pen. Her mind goes blank, and nothing comes to her. She feels that she can only guess. So, feeling shaken and dejected, she writes “1603”—which is of course exactly the right answer.Footnote 15

As David Rose and Jonathan Schaffer (2013) put it, ‘The case of Unconfident examinee represents the leading challenge to the orthodox idea that knowledge entails belief’ (2013, 20). Apart from this classic case from the mid 1960s—and the extensive critical response to it (on both sides), which fizzled out in the 1980s—the most notable recent lines of argument against the idea that knowledge entails belief, all in the past 10 years, are due to Blake Myers-Schulz and Eric Schwitzgebel (2013), Katalin Farkas (2015) and Susanna Schellenberg (2017a, 2017b). Myers-Schulz and Schwitzgebel present experimental evidenceFootnote 16 that the intuition in UNCONFIDENT EXAMINEEE that Kate has knowledge without belief (that Queen Elizabeth died in 1603) is robust, and on this basis, purport to give a ‘second wind’ to the old counterexample to the orthodox presumption that knowledge entails belief.Footnote 17 Farkas, on the other hand, uses cases of extended cognition (e.g. cases where one offloads one’s memory tasks to a notebook or a smartphone) as plausible cases where one has knowledge without belief, albeit, knowledge stored externally. Finally, Schellenberg’s tack is to cast doubt on whether the entailment thesis holds specifically in cases of perceptual knowledge, where (arguably) one knows simply via seeing that something is so, and regardless of whether one forms a belief.

Just as we have seen that Moore and Lehrer did not plausibly demonstrate that knowledge entails belief (either occurrent or dispositional), there is also a good case to be made that none of the above attempts aimed at establishing that knowledge does not entail belief succeed, at least in so far as none of these strategies plausibly demonstrate that knowledge does not entail dispositional belief. This point turns out to be relevant to the wider transition from ‘knowledge to belief, and then from belief to cognitive internalism’, given that occurrent belief rather than dispositional belief is more prima facie plausibly wed to a cognitive internalist picture of the mind.

Regarding the UNCONFIDENT EXAMINEE case, the pressure against the entailment thesis is really the strongest when we contrast (i) the observation that Kate’s lack of confidence in the proposition that the Queen died in 1603 does not seem to preclude her from knowing it, and indeed, manifesting that knowledge unconfidently, with (ii) the thought that Kate must believe and thus consciously endorse the content that the Queen died in 1603 at some time if she is to know it at that time. The force of UNCONFIDENT EXAMINEE against the entailment thesis lies in the fact that it leads us to embrace (i), and then on that basis reject (ii).

But importantly, a rejection of (ii) is compatible with the thesis that knowledge entails belief, so long as ‘belief’ is understood in a dispositional sense, where dispositional beliefs are merely available to mind for endorsementFootnote 18 even when the content of a dispositional belief is not (occurrently) consciously endorsed.Footnote 19 Rose and Schaffer (2013) support this rationale on the basis of two considerations. First, Kate’s memory traceFootnote 20 (viz. that Queen Elizabeth died in 1603) is not destroyed. Second, her guess is no accident.Footnote 21 On the second point, they write:

Indeed it seems as if her memory trace must still be not just present but actually operating in the background to guide her actions, even if she is unable in the moment to appreciate the fact. Putting these two reasons together—to the extent that it is useful to operate with the picture of a “belief box” in which various propositions are stored—we find it natural to think of Kate as having the proposition that Queen Elizabeth died in 1603 lodged in her belief box throughout. She stored it there during her studies and is still unconsciously guided by it when she “guesses.” Indeed we find it natural to imagine that—perhaps later that very day—Kate will recover from her panic and recall the information readily enough. She has the information stored in mind. She is merely temporarily blocked from accessing it normally (2013, 24–25).

It looks, then, as though we should deny that Kate has a dispositional belief only if we are prepared to say that her temporary block is permanent rather than temporary. But even if it were permanent, note that the kind of block she has just prevents her from accessing the information stored in mind normally. It does not prevent her from accessing it at all for the reason that this information stored in memory continues to guide her actions. Of course, were it to somehow be blocked off from even doing that, then we might then deny her the dispositional belief, on account that it is inaccessibly stored in memory. However, on that kind of a scenario, there would then be no pressure to attribute to her knowledge. If she guessed correctly, it would be by sheer luck.

The above considerations cast Myers-Schulz and Schwitzgebel’s (2013) experimental results in a different light. From the fact that folks are more likely to attribute knowledge than belief in UNCONFIDENT EXAMINEE, we have no good reason to reject the entailment thesis—at least not without a clearer sense of which sense of the polysemous ‘belief’ the participants took themselves to be withholding whilst at the same time attributing knowledge. Interestingly, as more recent experimental studies indicate,Footnote 22 when the same experiments are run whilst eliciting the dispositional reading of belief more so than it was elicited in the original experiments, people’s intuitions no longer disproportionately line up with attributing knowledge rather than belief.

The take-away lesson from the UNCONFIDENT EXAMINEE case seems to be this: the case (i) purports to show that it is not the case that knowledge entails belief; (ii) it plausibly does demonstrate that knowledge does not entail occurrent belief; but (iii) it does not succeed in showing that knowledge does not entail dispositional belief—on the contrary, we would plausibly be less likely to attribute knowledge in the case were dispositional belief not present.

Although Farkas’s argument against the knowledge-belief entailment thesis is ostensibly very different from the line of argument that proceeds from the THE UNCONFIDENT EXAMINEE case, an appreciation of Farkas’s wider argument shows that it ultimately slots into the very same kind of general (i, ii, iii) structure.

Her argument takes as its basis a case of cognitive offloading from memory to notebook. The case—involving the characters ‘Otto’ and ‘Inga’—was originally used by Clark and Chalmers (1998) as an argument against cognitive internalism, and in favour of the idea that cognition can extend beyond the boundaries of the skull and skin. The Otto and Inga case—and the thesis of ‘extended cognition’ more generally—will be discussed in some detail in §3. For our purposes now, though, let us focus squarely on how Farkas thinks the case supports a rejection of the orthodox idea that knowledge entails belief.

The key first step for Farkas is to take a queue from Edward Craig’s (1991) thinking about the purpose of the concept of knowledge, an understanding of which Craig thinks would help to illuminate what falls in its extension.Footnote 23 According to Craig: ‘[k]nowledge is not a given phenomenon, but something that we delineate by operating with a concept which we create in answer to certain needs, or in pursuit of certain ideals’ (1991, 2). On Craig’s view, we ‘create’ the concept of knowledge in order to meet the need we have to flag reliable informants. And so, on the Craigian view, the function of the concept of knowledge is to flag reliable informants, and relatedly, an appreciation of this function as the function it is should guide our thinking about what falls within the extension of the concept of ‘knowledge’.

Now, with these Craigian ideas assumed, Farkas encourages us to think about the case of Otto and Inga:

OTTO AND INGA: Inga would like to go to the Museum of Modern Art (MoMA); she recalls that the MoMA is on 53rd street, and she sets off accordingly. Otto suffers from severe memory loss and therefore he keeps all important information recorded in a notebook which he carries with him all the time. When he decides to go to MoMA, he looks up the whereabouts of the museum, finds it’s on 53rd street, and then he sets off. Many people agree that Inga had had the belief that the Museum of Modern Art was on 53rd street even before the issue came up in connection to her current visit. But Clark and Chalmers claim that if Inga has the belief, so does Otto, even before he looked up the information in his notebook. Otto has reliable, constant and easy access to the contents of his notebook, and he endorses the contents of his notebook automatically. This, according to Clark and Chalmers, is enough to qualify him as having the belief (2015, 190).

Farkas’s own idiosyncratic take on this case fits neither with traditional thinking (on which Otto neither believes nor knows that MoMA is on 53nd street in virtue of storing this information as he does in his notebook but not in his head) but nor does Farkas’s assessment lines up with the point Clark and Chalmers originally used the case to make, which is that—as they see it—Otto’s memory (and thus, his dispositional beliefs stored in memory) lies partly in the notebook, external to his biological brain. Farkas thinks—and we need not get in to the details just yet, but we’ll return to them—that we should agree with the traditionalist that the cognitive differences between Otto and Inga are substantial enough that, when it comes to attributing ‘belief’, we should do so disanalogously, to Inga but not to Otto. On the other hand, however, she thinks we should part ways with the traditionalist—and simply be guided by Craig—when it comes to assessing whether to attribute knowledge to Otto. Recall again the Craigian idea that the point of the concept of knowledge is to track reliable informants, and just consider in this light how we use ‘knowledge’ to track such informants in cases of, e.g. seeking phone numbers. As Farkas writes, in ‘some everyday contexts, it is very natural to attribute knowledge to subjects who are in Otto-type situations. You ask me if I know NN’s phone number, and I say “sure”, reaching for my smartphone’ (Farkas, 2015, 190).

Putting this all together, Farkas thinks we have compelling reason to think Otto knows but does not believe that MoMA is on 53rd street, and, a fortiori, that the knowledge-belief entailment thesis is false. Now, I have suggested at the outset that I think Farkas’s argument ends up slotting into the (i, ii, iii) structure that characterised the purported argument against the entailment thesis from UNCONFIDENT EXAMINEE. I now want to explain why.

First, consider that one tempting spot to challenge Farkas’s reasoning is her claim that Otto and Inga are different enough that we should not attribute dispositional belief across the cases symmetrically. Why not? Why are not Clark and Chalmers right about this, as opposed to the traditionalist? Fortunately, there is a way press back against Farkas without fully opening that can of worms (we will circle back to it in §3), which is to suggest that by her own lights we ought to attribute Otto a dispositional belief. The reasoning here is that attributing such a dispositional belief is the most promising way for Farkas to vindicate her claim that Otto has (extended) knowledge. Farkas’s rationale for attributing Otto extended knowledge, after all, is meant to be guided by the Craigian idea that we should use ‘knowledge’ to track reliable informants. The presumption here (which we may grant, ex hypothesi) is that Otto is such a reliable informant; ask him where MoMA is, he can reliably tell you (via a process that involves consulting his notebook rather than biomemory). Now, what is it that grounds Otto’s reliability about where MoMA is? It is hardly a brute fact that he is reliable. On the contrary, he is a reliable informant because he reliably stores the information (just like Inga does); his information is correct, easily available for endorsement, etc. Indeed, it thus looks quite a bit like the thesis that Otto knows where MoMA is (in virtue of what is in his notebook) would be explained—even granting the Craigian story—by his having something that looks an awful lot like a dispositional belief.

Now, a traditionalist has at this juncture might try to dig their heels in as a matter of principle: ‘Cognitive internalism is true and so, necessarily, all cognition plays out in the head; therefore, Otto simply can’t have a dispositional belief externally stored’. But—crucially—it looks like this kind of a principled reason is already out the window for Farkas, who explicitly allows knowledge outside the head. Farkas’s line that Otto’s case features knowledge without belief accordingly occupies a curious area of dialectical space: her claim that Otto has knowledge (that MoMA is on 53rd street) itself seems best explained by his having a dispositional belief, in virtue of how he stores the information he does, not in biomemory, but in the notebook. Farkas, of course, denies that he has a dispositional belief (by appealing to cognitive internalist thinking), but that denial would itself be principled denial only if Farkas were to also deny that he has extended knowledge (which she of course does not deny).

Putting this all together, then, it looks as though—as with UNCONFIDENT EXAMINEEE—the case of Otto and Inga (at least, as Farkas is using it) exhibits (i, ii, iii) structure; it is a case that Farkas (i) purports to use to show that it is not the case that knowledge entails belief; (ii) the case plausibly does demonstrate that knowledge does not entail occurrent belief (given that Otto clearly lacks such a belief, no less than Kate does in UNCONFIDENT EXAMINEE), but (iii) it does not succeed in showing that knowledge does not entail dispositional belief, and, if anything, only serves to positively reinforce this idea.

Let us round out our discussion of the knowledge-belief entailment thesis with a brief look at Susanna Schellenberg’s domain-specific dismissal of the idea that knowledge entails belief. The line she advances is ‘domain specific’ because it is meant to apply exclusively to perception, and thus to perceptual knowledge. According to Schellenberg’s view of perceptual knowledge, capacitivism, a subject (S), has perceptual knowledge that p by seeing that p, which requires that S employ ‘a capacity to single out what she purports to single out’ (2017, 318) and S’s mental state (whereby S sees that p) must have ‘the content it has in virtue of S having successfully employed her capacity to single out what she purports to single out’ (2017, 318).

As the reader will have noticed, ‘belief’ does not feature in the above story. This, Schellenberg thinks, is just as it should be. She writes:

Orthodoxy has it that one cannot know that p without believing that p. Capacitivism is neutral on whether there is any such belief condition on knowledge. This is attractive, since arguably, we know that p simply in virtue of seeing that p. By contrast, we do not believe that p simply in virtue of seeing that p. After all, I can see that p without forming any beliefs (2017, 318).

Of course, even if Schellenberg is right, she will have been right about a story of perceptual knowledge acquisition. What about perceptual knowledge retention? Suppose you see that p at t1. At t2, you are no longer thinking about p. But, if someone asks you at t2 what p looked like, you remember and can tell them. But this would turn out to be mysterious if at t2 you did not retain this information about p in a way that was then later available to mind for endorsement.Footnote 24 But that is just the mark of a dispositional belief. To the extent that Schellenberg’s capacitivism is a correct story of perceptual knowledge acquisition, this story looks to be compatible with the version of the entailment thesis that has seemed most plausible so far—viz. that knowledge requires at least dispositional belief.

Recall now that the ‘map’ to unearthing the cognitive internalist dogma in mainstream thinking about knowledge had two key ‘links’ in the chain, one from knowledge to belief (Link 1), the other from belief to cognitive internalism (Link 2).

This section—critically examining Link 1—reveals that the most charitable way to unpack Link 1 is as:

  • Link 1dispositional: propositional knowledge \(\to\) (entails) dispositional belief rather than:

  • Link 1occurrent: propositional knowledge \(\to\) (entails) occurrent belief:

However, from Link 1dispositional, we most plausibly get to cognitive internalism only by way of:

  • Link 2dispositional: dispositional belief \(\to\) (is best explained by) a cognitive internalist picture of the mind rather than:

  • Link 2occurrent: occurrent belief \(\to\) (is best explained by) a cognitive internalist picture of the mind.

But this is where the overarching story—from mainstream thinking about knowledge to the cognitive internalist assumption that tacitly underlies it—begins to show some real cracks. Just consider that, whereas Link 2occurrent is prima facie very plausible (if not obvious to many), Link 2dispositional really is not.

The reason Link 2occurrent seems platitudinous is that occurrent belief is usually taken to involve consciously entertaining (and subsequently endorsing) a proposition, and a biological brain is plausibly (though this point is of course debatable) necessary and sufficient for this kind of conscious activity. Accordingly, it is prima facie plausible that cognition of the sort that is realised exclusively as the cognitive internalist countenances is what furnishes us with whatever occurrent beliefs we have. Crucially, however, a biological brain is—though obviously sufficient—not necessary for realising the kind of thing that hosting a dispositional belief is generally taken to involve, which is the storing of information that is available to us for conscious endorsement. If anything, the ubiquity of cognitive offloading suggests that even though biological brains suffice for storing information available for conscious endorsement, they are obviously not necessary because we use them for this very purpose increasingly less—especially when it comes to practical information of the sort we rely on to structure our lives. It is, then, at best prima facie plausible that cognition of the sort that is realised exclusively as the cognitive internalist countenances furnishes us with only some of our dispositional beliefs. But this means, then, that the phenomenon of dispositional beliefs is best explained by a picture of the mind that allows for storage of information available for conscious endorsement to sometimes be handled intracranially, sometimes (and increasingly often) not.

At this juncture, the proponent of cognitive internalism might simply double down as follows: ‘even if the sense in which knowledge entails belief is best understood as Link 1dispositional rather than Link 1occurrent, and indeed even if it looks as though we can make sense of many of the dispositional beliefs we have without assuming anything like cognitive internalism, it remains that cognitive internalism stands up as an independently and overwhelmingly plausible ‘pillar’ in the philosophy of mind. It establishes the bounds of cognition in a way that aligns with centuries of philosophical thinking, and we are better placed simply accepting the implications of cognitive internalism wherever they lead us, even where they don’t align so well with our other commitments (at least, when these other commitments lack the kind of ‘bedrock’ status that cognitive internalism enjoys). And so, despite initial appearances to the contrary, we should not accept but resist the temptation to think that the process of storing information available for endorsement in a notebook or iPhone (rather than in biomemory) is a genuine cognitive process, and thus, we should resist attributing ‘beliefs’ and ‘knowledge’ on the basis of such storage.’

Does the proponent of cognitive internalism here have a point? This really depends on whether cognitive internalism is (or deserves to be) the kind of ‘pillar’ in our theorising that the above reasoning suggests. As it turns out, pillars fall, and, lately, old ‘internalist pillars’ in particular have been falling right and left.

3 Objection and reply

Thus far, the line advanced is that a cognitive internalist picture of the mind is much more dispensable in epistemology than has been assumed. However, let us now take seriously the following kind of rejoinder: regardless of whether cognitive internalism is indispensable (or not) in our epistemological theories, it independently enjoys a kind of sacrosanct status as a theory of the metaphysics of mind; thus, as the thought would go, we should be taking it for granted anyway in epistemology, given the strong independent reasons for thinking that cognitive internalism is on entirely solid ground.

Let us think through this line of critique. Until relatively recently, the study of knowledge was—following a tradition inherited from DescartesFootnote 25—a thoroughly ‘internalist’ enterprise in three key ways.

First, it used to be taken for granted that the content of our thoughts is determined entirely by the inner workings of the mind—viz. content internalism.Footnote 26 On this way of thinking, your intentional attitudes (e.g. your beliefs and other attitudes that are about things) are about the things that they are about (rather than about other things) in virtue of your psychological states and nothing else. Any two people in the same psychological states, then, must be thinking about the very same thing. For those (like Descartes) who are aligned with this kind of thinking, it is easy to see how ‘rigorous philosophical inquiry must proceed via an inside-to-out strategy’, and of course, as was apparent in the Meditations, from this kind of methodological starting point, the challenge of (non-circularlyFootnote 27) defeating the sceptic becomes especially difficult.Footnote 28

Even so, content internalism is not itself an epistemological thesis (even if it has some epistemological ramifications); it is a thesis about how the content of our thoughts and words is individuated. An importantly different kind of internalism—also inherited from Descartes and widely assumed until around the 1970sFootnote 29—is epistemic internalism.Footnote 30 Epistemic internalism is not a thesis about what our thoughts and words refer to, but about what kinds of things justify our beliefs in a way that matters for knowledge. It is in principle compatible with either content internalism or content externalism.Footnote 31 What the epistemic internalist maintains is that epistemic justification is solely determined by factors that are internalFootnote 32 to a person.Footnote 33 Such factors include, e.g. what mental states one is in and what is accessible to one via reflection alone. A simple reason why this kind of view (a centrepiece of Descartes’ epistemology, but with origins as early as the Theatetus) has plausibly enjoyed the support it has is that we tend to think of the kind of justification that matters for knowledge as being associated with reasons and evidence, and the matter of what reasons and evidence one has seems—on the face of things—to be determined by factors internal to one (e.g., what your mental states are).Footnote 34

Rounding out the three internalist ‘pillars’ of Cartesian epistemology is our old friend cognitive internalism on which what is claimed to be ‘internal’ to a thinker is not the content of their thoughts (content internalism) or what matters for justifying their beliefs (epistemic internalism), but rather the material realisers of their cognising, including whatever thoughts and beliefs they have, justified or not.

The suggestion—canvassed in the previous section on behalf of the traditionalist—that cognitive internalism is some kind of unalterable ‘pillar’ that must not be dislodged is really not very compelling in the context of appreciating that—of these three internalist ‘pillars of Cartesianism’—the first two have already fallen, and both within just the past 50 years.

Content externalists in the 1970sFootnote 35 and 1980sFootnote 36 have shown how our environments play a crucial role in individuating meaning and mental content, and to such an extent that content internalists are nearly extinct in 2021. As Juhani Yli-Vakkuri and John Hawthorne (2018) put it, in a recent monograph purporting to be the final nail in the coffin of this kind of internalist thinking, ‘entanglement of our minds with the external world runs so deep that no internal component of mentality can easily be cordoned off’. With the exception of Hawthorne and Yli-Vakkuri’s purported final takedown, content externalism is now so popular it is rarely taken to need any additional argument. Essentially, philosophical thinking has ‘flipped’ almost completely since the mid 1970s, and on a position fundamental to our grip on the very nature of thought.

What about epistemic internalism, then? It has slowly but steadily (since the 1960s) been heading the way of content internalism. According to results from a PhilPapers Survey published by David Chalmers and David Bourget in 2014, only about a quarter of 931 philosophers surveyed (246/931 (26.4%)) self-identify as epistemic internalists.Footnote 37 This is so even though internalism captured the default position in epistemological theory from Plato, to both rationalists (Decartes) and empiricists (Locke and Hume)Footnote 38 all the way up to Gettier (1963). Whilst debates between epistemic internalists and externalists remain contentious, one thing that is clear is that epistemic internalism is no longer the default view but rather the exception.Footnote 39

So is cognitive internalism then the only ‘Cartesian pillar’ that should be thought of as ‘safe’ from the externalist wave—and as such, permanently fixed? The short answer is ‘no’ for the reason that this final internalist pillar has at least partially (arguably: mostly) fallen already, as the past 20 years of the philosophy of cognitive science suggests. It is just that—put simply—this news has not quite spread to mainstream epistemology.

The cracks in cognitive internalism started quite small.Footnote 40 Forget iPhones and the like for a moment, and just think about your hands, and how you move them around, gesturing as you talk. This kind of gesturing not only facilitates communication, but also helps language processing (McNeill, 1992). Likewise, consider the baseball outfielder (e.g. McBeath et al., 1995) trying to catch a fly ball, by running in a direction that makes the ball appear to follow a straight line. In doing this, the outfielder is solving a complex problem not just by perception, but by a kind of ‘perception–action coupling’—viz. by using perceptual information to guide movement and then using movement to hold the perceptual information constant.

The above are just some representative example cases—others (many of which have appeared just since the 1990sFootnote 41) involve visual consciousness,Footnote 42 concepts,Footnote 43 memory,Footnote 44 moral cognition,Footnote 45 etc.—which have been taken to favour the view that cognition is best understood as not only taking place in the brain, but more widely, as embodied. Wilson and Foglia (2017) articulate the core of the ‘embodied cognition’ thesis as follows:

Many features of cognition are embodied in that they are deeply dependent upon characteristics of the physical body of an agent, such that the agent’s beyond-the-brain body plays a significant causal role, or a physically constitutive role, in that agent’s cognitive processing.

What the evidence for embodied cognition suggests is that cognition is not merely (as traditionalists would have it) ‘sandwiched between, while segregated from’ perception and action. The dependence of the former on the latter is simply too deep to separate in the clean way the traditionalist/internalist would want.Footnote 46

Getting down to brass tax: if any part of an agent’s non-brain body has ever played a physically constitutive role in cognitive processing, then strictly speaking, cognitive internalism is false.Footnote 47 And as more evidence has come in that validates this very idea, embodied cognition has increasingly taken over as the ‘default’ position in cognitive science. As Fred Adams (2010)—a committed traditionalist—concedes: ‘The view that cognition is embodied […] is rapidly gaining prominence in the world of cognitive science, and is aiming for dominance (2010, 619). According to Lawrence Shapiro (2014), embodied cognition is”now one of the foremost areas of study and research in philosophy of mind, philosophy of psychology and cognitive science’.

It is hard to see how cognitive internalism should deserve any kind of sacrosanct status when the tide in cognitive science is now generally against it. But if that is right, then is not it just a clear mistake for epistemologists to cling tacitly to cognitive internalism?

Maybe—in a sense—not. Consider this line of argument: ‘Let’s assume cognition is embodied – granted! Even so, this is a far cry from suggesting that you can have beliefs in your phone. Your phone is not part of your brain or your biological body!’ This kind of rejoinder suggests that perhaps the best candidate for a plausibly sacrosanct thesis in the neighbourhood of cognitive internalism is not strict cognitive internalism after all, but rather the more permissive cognitive bio-internalism—the thesis that cognition is essentially biologically realised.

However, even if we shift the goal posts of sanctity from cognitive internalism to cognitive biointernalism, we still fail to capture anything properly sacrosanct. Two straightforward challenges on this score come from cognitive neuroscience over the past 6 years alone: (i) the 2015 creation of the first artificial neuron (Simon et al., 2015) and (ii) the first successful case (in 2019) of creating artificial memories from scratch and implanting them in mice, where the artificial memories guided behaviour indistinguishably from non-implanted memories (Vetere et al., 2019). Note that in neither of these cases is cognition realised entirely as the biointernalist would have it.

A larger elephant in the room, however, comes from the philosophy of cognitive science, where researchers are increasingly open and explicit in their denial of even cognitive biointernalism. It is here where it will be useful to circle back to the case of Otto and Inga, originally due to Clark and Chalmers (1998). Both the cognitive internalist and the more permissive cognitive biointernalist are going to diagnose Inga and Otto asymmetrically when it comes to whether they count (respectively) as remembering—prior to accessing this information from storage—that the Museum of Modern Art is on 53rd street. In Inga’s case, we attribute to her a paradigmatic dispositional belief, in virtue of storing this (previously endorsed) information in biomemory. In Otto’s case, we—according to the cognitive internalist and biointernalist—deny this dispositional belief attribution, simply because the information is not stored in biomemory; it is stored somewhere else.

But how important should this be, really? Proponents of extended cognition think that giving this kind of theoretical weight to the material constitution and location of our memory storage is outdated and unprincipled—and as David ChalmersFootnote 48 puts it—a form of bioprejudice. A more egalitarian approach to the bounds of cognition would have us focus, when assessing whether to include something as part of a cognitive process, less on what it is made of and where it is, and instead on what it does. If something is doing the same thing as something that is part of a cognitive process, then why not—in the spirit of parity of treatment—rule it in?

This is the central (then-)radical idea from Clark and Chalmers’ (1998) The Extended Mind, which is summed up (1998, 8) in their ‘parity principle’:

Parity Principle: [I]f, as we confront some task, a part of the world functions as a process which, were it to go on in the head, we would have no hesitation in accepting as part of the cognitive process, then that part of the world is part of the cognitive process (1998, 8).

For those willing to reason in accordance with this principle, it will follow that cognition is not merely embodied, but also that it can in some circumstances be extended—viz. in the sense that not just extracranial but extraorganismic things in one’s environment (e.g. notebooks, smartphones, tactile vision substitution systems,Footnote 49 eyeborgsFootnote 50) can in certain circumstances partially constitute that agent’s cognitive system. On this way of thinking, we not only can, but should—viz. with reference to the Parity Principle—treat Inga and Otto symmetrically in terms of memory.Footnote 51 Not only Inga, but also Otto, remembers—prior to accessing this information from storage—that the Museum of Modern Art is on 53rd street. Additionally, in Inga’s case, we attribute a dispositional belief (that the Museum of Modern art is on 53rd street), in virtue of her storing this information in biomemory. And by parity of reasoning, in Otto’s case, we attribute an (extended) dispositional belief with this same content, in virtue of Otto’s storing the same information in (extended) memory.

Of course, for the champion of extended cognition, not everything that one causally interacts with whilst engaging with a cognitive task is going to get ‘ruled in’ as part of one’s ‘extended’ cognition. Far from it. One of the key research problems in the contemporary literature on extended cognition is how exactly to distinguish cases like that of Otto, where the parity principle is plausibly satisfied, from cases where we should think it is not—e.g. as when one consults a phone book, or just happens to use a device for a one-off task.Footnote 52

Let us zoom out for a moment. We have been responding in this section to an envisioned critic of the argument from §1. The argument maintained that a cognitive internalist picture of the mind is much more dispensable in epistemology than has been assumed. The anticipated rejoinder was that even if that is right, cognitive internalism independently enjoys a kind of sacrosanct status. Granted, the epistemologist (like anyone else) should accept an internalist picture of the mind, and whatever is implied by it, if the cogitive internalism deserves to be treated as a kind of theoretical pillar—one such that we should alter what comes into conflict with it, rather than to alter the pillar itself. However, we have now seen in this section that this is hardly the case—and on the contrary—that the tide in recent cognitive science is moving against not only cognitive internalism (e.g. embodied cognition) but also even cognitive biointernalism (e.g. extended cognition).

4 Concluding remarks

I have argued here against a certain kind of presumptive picture in epistemology—one that takes cognitive internalism for granted and in doing so de facto forecloses the possibility that epistemic evaluations of good and bad cognition are anything other than intracranial evaluations. This way of thinking artificially—and without satisfactory theoretical motivation (§2)—restricts the way epistemologists approach questions posed by digital information storage and generation, and the truth-directed (epistemic) evaluations we make in response to such questions. When asking whether information stored digitally can be digital knowledge, we should feel free to ask this literally, and without concern that doing so requires giving up any underlying commitments we need to make in order to pursue traditional epistemological questions, or (§3) which are otherwise sacrosanct.