Epistemology of Memory

We learn a lot. Friends tell us about their lives. Books tell us about the past. We see the world. We reason and we reflect on our mental lives. As a result we come to know and to form justified beliefs about a range of topics. We also seem to keep these beliefs. How? The natural answer is: by memory. It is not too hard to understand that memory allows us to retain information. It is harder to understand exactly how memory allows us to retain knowledge and reasons for our beliefs. Learning is largely a matter of acquiring reasons for changing views. But how do we keep reasons for the views we keep? The epistemology of memory concerns memory’s role in our having knowledge and justification. This branch of epistemology, unlike nearly all other branches, addresses our having knowledge and justification over time.

1. The Nature of Memory

Traditionally, philosophers have likened memory to a storehouse or a recording device. In the Theaetetus, Plato claims that the mind is analogous to a wax tablet. To perceive is to make an impression on the tablet, leaving behind an exact image or representation of what was perceived. Memory keeps the images and forgetting is a matter of losing them. In his Confessions, Augustine says perception deposits images of objects into the storehouse of memory and the process of recalling is the process of retrieving these deposits. Locke and Hume tell much the same story, as do many other philosophers up through the 20th century.

On this storehouse view, memory stockpiles experiences and beliefs. Stored items may eventually degrade or become hard to access, but otherwise do not change (see Audi (1994: 420-1), Burge (1997: 321) and McGrath (2007: 13)). This view is commonsensical. It explains how it is that we are able to represent the past accurately in our thoughts and recollective experiences. It also explains why each of us, over time, tends to believe the same thing occurrently more than once. Yesterday, Maria believed that she went to high school in Santa Fe and she believes that today too.

During the 20th century psychologists generally abandoned the storehouse view (see for example Bartlett (1932) and Schacter (1996, 2002)), though still thinking that memory stores information. They believe human memory processing is much more complicated than the mere depositing of items and later withdrawing them. Memory selectively stores information, expands part of it, combines it with background information and adds data from the context, in which the subject later retrieves the information. In other words, memory generally alters significantly what enters it. As a result, recollecting is not the retrieving, but rather the generating of representations of the past. Recollecting actually generates new beliefs about the past. Empirically minded philosophers of memory also have generally abandoned the storehouse view in favor of this generative view (see, for example, Debus (2010) and Michaelian (2011a, 2011b)), but epistemologists have been slower to shift models. Since this article covers the epistemological discussion of memory up to the beginning of the 21st century, the storehouse view will generally be implicit.

Setting aside how exactly memory works, it will aid our epistemological discussion to get clearer on what memory is of or for. At least as far back as Henri Bergson (1896/1994) and Bertrand Russell (1921/1995), philosophers have recognized that there are different kinds of memory, or different memory systems and 20th century psychological research has confirmed the philosophers’ distinctions. Talk of ‘memory’ simpliciter, as if there were a single, uniform faculty, can obscure it. Distinct memory systems allow us to do different things and consist of different networks of rule-governed psychological processes. Two memory systems that are important to distinguish are declarative memory and procedural memory. Declarative memory is memory of information and events. Procedural memory is memory for skills and of how to perform actions. Different parts of the brain house, on the one hand, our data about bicycle riding and our riding experiences and, on the other hand, our acquired talent for riding. This helps explain the familiar phenomenon of finding it easy to do something, yet hard to state instructions for doing it (think of swimming, playing a flute, or tying a shoe), or vice versa.

Declarative memory divides into semantic (or propositional) memory and episodic (or experiential) memory. Semantic memory is memory for propositions and episodic memory is memory for events, one has experienced. To see this distinction, consider how these types of memory can come apart. You remember that Plato taught Aristotle, but you do not remember Plato teaching Aristotle. How could you remember it? You were neither there nor did you witness it. I can remember that I was born in a hospital, but (mercifully) I cannot remember being born in a hospital.

Semantic memory underlies memories with propositional content; semantic memory claims are often of the form “S remembers that p”. Episodic memory underlies memories with a kind of non-propositional content; episodic memory claims are often of the form “S remembers x”. Semantic memory is by far the most discussed memory system in epistemology. This is understandable, since epistemology centers on states that have propositional content. Epistemologists primarily discuss what it is for S to know that p, or what it is for S to have justification for believing that p, or the like. They focus less on non-propositional knowledge and justification. The epistemology of memory, as a result, has chiefly been the epistemology of semantic memory.

But it is worth noting that neglecting to consider other memory systems can render our epistemological theories vulnerable. Some philosophers have objected to certain theories of propositional knowledge on the grounds that they do not accommodate the role that episodic memory plays in our believing (see, for example, Shanton (2011)). And deeper reflection on procedural memory may advance other debates in epistemology, such as debates concerning knowledge-how. Knowledge-how is a practical knowledge, what you have when you know how to swim or how to tie a shoe. There is debate about whether knowledge-how is reducible to knowledge-that. That is, there is debate about whether practical knowledge can be fully understood in terms of knowing various propositions. But procedural memory seems to ground our knowledge-how and it differs importantly from declarative memory (see Michaelian (2011a)). In fact, psychological research suggests that sophisticated procedural memory can be retained even when semantic memory is crippled (one artist entirely lost his knowledge of language due to brain-damage, having to relearn his native tongue altogether and yet he remembered how to paint! See Schacter (1996: 140-2)). Investigating procedural memory may help reveal that knowledge-how is not reducible to knowledge-that.

2. Memory and Knowledge

Most of the interesting features of memory’s relationship with knowledge originate in memory’s relationship with justification. Knowledge requires justification. As a result, when justification connects in interesting ways with a topic, knowledge shares those connections. This section covers what is perhaps the only unique connection between memory and knowledge.

a. The Epistemic Theory of Memory

Semantic memory is responsible for our remembering that something is true. Much philosophizing in the 20th century tried to state necessary and sufficient conditions for propositions of the form S remembers that p. The theory that dominated that discussion is especially important in epistemology: the epistemic theory of memory (see, for example, Anscombe (1981), Ayer (1956), Audi (2002), Locke (1971), Malcolm (1963), Moon (2013), Owens (2000), Pappas (1980) and Williamson (2000)). Roughly put, the epistemic theory states that remembering is a kind of knowing. If S remembers that p, then S knows that p. Many philosophers go even further: if S remembers that p, then S knows that p because S previously knew that p. You remember that Plato taught Aristotle, and this is because in the past you came to know that Plato taught Aristotle, and because that past knowledge has contributed to your present knowledge. (Incidentally, Plato might even agree; he appears to endorse the epistemic theory of memory in the Theaetetus.)

If the epistemic theory of memory is correct, we might not remember as much as we think we do. Remembering requires knowing and the standards for knowing are not low. In particular, it is generally accepted among philosophers that S knows that p just in case p is true, S believes that p, believing that p is justified for S and it is not accidental that S’s justification for p gives S a true belief that p. Knowledge is a kind of justified true belief, a kind where the truth of the belief is tightly connected to its justification. When the connection is not tight, the belief might be “Gettiered” or true by sheer accident. If you see someone walking down the street dressed as a postal worker, you might justifiedly believe that your mail will be delivered soon. Suppose the person you see is not in fact a postal worker, but is merely testing out a Halloween costume. And suppose that, nonetheless, the mail will indeed be delivered soon; your regular postal worker is just around the corner, delivering mail to your neighbor. Your belief that the mail will be delivered soon is justified, but true only by coincidence. So you do not know that the mail will be delivered soon.

If remembering requires knowing, then remembering requires everything required for knowing. If any requirement is not met, one does not remember, but at best merely seems to remember. In other words, if you seem to remember that the keys are on the dresser, but they in fact are not there, or you have no reason to believe that they are there, or you simply deny that they are there, then you do not remember that they are there.

Why endorse the epistemic theory of memory? A main reason is that it fits our ordinary uses of “remembers” and “knows” (see Moon (2013)). Consider the following conjunctive claim: Sally remembers that she has visited Rhode Island, but she does not know that she has. This conjunction sounds odd and one plausible explanation of the oddness is that remembering requires knowing. The second conjunct denies something the first conjunct asserts, so the conjunction seems incoherent.

Here is a closely related reason for endorsing the epistemic theory. Remembering requires knowing just in case all of the following are true: remembering requires believing, remembering requires justification and remembering requires non-accidental truth. And we can argue, one at a time, that remembering does indeed have these requirements. For example, the best explanation of the oddness of certain conjunctive claims is that remembering requires believing. Consider: Peter remembers that he owes Paul a dollar, but he does not believe that he owes Paul a dollar. At least at first glance, it is hard to make sense of this. How could Peter remember that without believing it?

Andrew Moon (2013) proposes another reason for supposing that remembering requires believing. He claims that if S remembers that p, then S can use p as a premise in certain justifying inferences. But, Moon adds, a premise is usable in justifying inference only if believed. If you do not believe that all tigers are mammals and that all mammals are animals, you cannot use these propositions as premises for reasonably inferring that all tigers are animals. So, remembering requires believing. Similarly, Moon claims that remembering requires justifiedly believing. This is because a premise is usable in justifying inference only if justifiedly believed. And inferences based on remembered propositions are justifying. So, remembering requires justified belief.

However, Moon’s argument faces worries. Suppose S remembers that p, but also remembers that all experts deny that p. Can S use p as a premise in any justifying inferences? Perhaps not. If S cannot, then not all we remember is usable as a justifying premise and Moon has not shown remembering requires believing. Or, suppose S justifiedly does not believe p. Couldn’t S nonetheless have reason to believe that if she uses p (rather than not-p) in her inferences, she will be more likely to arrive at the truth (if, say, p is a scientific theory that is likely ‘false but approximately true’)? If so, S might be able to use p as a premise in justifying inference, without believing p. Even if remembering that p allows justified inference from p, justified inference from p would not guarantee belief that p. It would not follow that remembering requires believing.

While the epistemic theory may make sense of certain conjunctive claims, it faces many objections. As noted above, if remembering requires knowing, then remembering requires everything required for knowing: belief, justification and non-accidental truth. Arguments against the epistemic theory have tried to show that remembering is possible even when at least one of these three requirements for knowledge is not met.

Martin and Deutscher (1966) give a well-known example, in which there (allegedly) is remembering without believing. A painter paints a detailed farmyard scene. He believes he merely imagined the scene. However, it turns out that the painting captures an actual farmyard scene that the painter saw as a child. Unwittingly, the painter simply reproduced that scene. Martin and Deutscher (1966) add that the painter “did his work by no mere accident,” suggesting that the painter’s childhood experience caused him to bring to mind the scene (even though he believes that he merely imagined the scene). They conclude that this is a case of remembering without belief. Since knowing requires believing, this would be a case of remembering without knowing.

Martin and Deutscher’s conclusion may in a sense be right, yet their example may also not pose any problem for the epistemic theory. We can agree that the painter does not believe that the scene occurred. But exactly what is it the painter is remembering? It is plausible that, if he is indeed remembering something, he is remembering the scene or his visual experience of it. It is less plausible that he is remembering that the scene occurred or remembering the scene as having occurred. In other words, Martin and Deutscher may have given a case of remembering without believing, but the remembering is not semantic. It is episodic or some other sort of memory. If that is correct, then the example is no threat to the epistemic theory of memory, since that theory concerns only semantic memory.

Audi (1995) and Bernecker (2010: 75-7) appear to offer cases of remembering without the sort of justification that knowledge requires. Knowledge requires fairly strong justification and this justification must not be defeated. If Billy knows that there is a cookie on the table, then Billy has strong reason to believe that it is on the table. Even if he has some reason to doubt that there is a cookie on the table (he may have reason to suspect that his sister shaped some clay to look like a cookie), these doubts do not defeat his justification, when he knows that there is a cookie on the table.

Audi and Bernecker offer the following kind of case. Suppose you remember that Plato taught Aristotle. However, your friends go on to play a prank on you and give you convincing reasons to think Plato never taught Aristotle–Plato never existed and Aristotle had no teacher. You retain your belief, but the prank defeats your justification. Your justification is no longer strong enough for you to know that Plato taught Aristotle. Nonetheless, Audi and Bernecker would think, you remember that Plato taught Aristotle. So, they conclude, remembering does not require justification.

But why suppose that, after the prank, you still remember that Plato taught Aristotle? The answer is unclear. Is it because you still have a true belief, which you acquired in the past, even though you lack overall reason for keeping it? Why would that be sufficient for remembering? Unless an explanation is offered, we may not have reason to count the case as a counterexample to the epistemic theory of memory.

Bernecker (2010) describes a case, in which there appears to be remembering without non-accidental truth–that is, the remembered proposition is true by mere accident: you justifiably, but incorrectly believe that your friend has borrowed a certain book from the library. Later, your friend indeed checks out that very book. As a result, your belief is true, but by coincidence alone. Bernecker thinks you still count as remembering that your friend has borrowed the book from the library. If this is a case of remembering an accidentally true proposition, it is a case of remembering without knowing. But is the antecedent here true? Some philosophers (for instance Moon (2013),) see no reason to suppose that it is. If Bernecker can persuade us that it is in fact true, he will have provided a genuine counterexample to the epistemic theory of memory.

We have seen several attempts to show that remembering does not require knowing. Each attempt faces a similar problem: when knowledge is absent, it is unclear whether semantic remembering is present. Support for the claim that semantic remembering is indeed present has typically involved an appeal to intuitions that some critics apparently lack. But there may be a less controversial way of showing that remembering does not entail knowing. If epistemologists discard the storehouse view of memory and adopt the generative view, they may discover clearer kinds of cases, where propositions are remembered, yet not known, or at least not known in the past by the subject.

For debates about the epistemic theory of memory, it matters significantly whether remembering entails knowing. And it matters significantly for another debate in epistemology. Timothy Williamson (2000) has influentially argued that the concept of knowledge is fundamental in our thinking. Having the concept of knowledge crucially allows us to understand quite a bit of psychology and epistemology and we cannot fully explain knowledge in terms of other psychological or epistemological conditions and relations.

In support of this, Williamson (2000: 34) claims that “knowing is the most general factive stative attitude.” He means roughly that, if the state of having a certain kind of attitude toward p (like hearing that p or seeing that p) guarantees that p is true then being in that state guarantees that p is known. Knowing is the most general factive stative attitude, in that there is no way that S could be in the state of having a truth-guaranteeing attitude toward p without also knowing that p. Now, many philosophers think that remembering that p guarantees that p is true, even if remembering that p does not guarantee belief that p, strong overall justification for believing that p or the non-accidental truth of p. If they are right and remembering does not require knowing, then Williamson’s claim is incorrect. Remembering is factive, but is not knowledge, so knowledge is not the most general factive stative attitude. As a result, his argument would weaken; it is less clear that the concept of knowledge is fundamental to our thinking.

A closely related claim of Williamson’s may also be challenged, if remembering does not require knowing. Williamson says that all and only evidence is knowledge. More precisely, he says that S knows that p just in case p is included in S’s total evidence. It is plausible that if S remembers that p, then S’s total evidence includes p. If this is right and if remembering does not require knowing, then not all evidence is knowledge. Some of what we remember is evidence, yet not known.

3. Memory and Justification

For most debates in the epistemology of memory it does not matter whether remembering entails knowing. This is because most debates ultimately concern the connections between memory and epistemic justification. So, even if remembering does not entail knowing, there remains much to discuss. One neutral way of proceeding is to think about cases of apparent remembering: cases, in which a subject has a memory experience that p, or recollects that p, or recalls p as known or as true and so on. Even if the subject is not in fact remembering that p, memory may still justify the subject in believing that p. But how? And in exactly what circumstances?

a. Problems

In debates about epistemic justification, philosophers have construed memory mainly as a source of challenges. A main way to test a theory of justification is to see if it has the right implication in cases, where memory plays some special role.  Philosophers apply this test most frequently in the debate about internalism and externalism in Epistemology.

It is controversial what these views even are, but here is a rough characterization. At a minimum, internalism states that mentally alike individuals are completely alike in their justification (see Conee and Feldman (2001)). Environmental differences by themselves make no difference to justification. So if, for example, you are justified in believing that there are boxes in the basement, that justification would remain even if your neighbor stole all the boxes from the basement. In order for your justification to change, your mental life would have to change–you would need to have a visual experience of an empty basement, or to seem to hear your spouse report that the basement is bare and so forth. You, and someone mentally just like you, are both justified in believing that there are boxes in the basement, even if only one of you has boxes, even if only one of you lives in a world with basements.

Externalism is the denial of internalism. It states that environmental differences can result in differences in justification, even if they do not result in mental differences. What is actually downstairs may matter. Or it may matter what is downstairs in nearby possible worlds. It may matter whether the particular way, in which you would form or keep the belief that there are boxes in the basement, tends to get at the truth.

Any theory of justification appears to face some challenge from facts about human memory. Externalists have argued that their view can overcome these challenges better than internalism can (see, for example, Bernecker (2008, 2010), Goldman (1999, 2009, 2011), Greco (2005) and Senor (1993, 2010)). A fine way to test a theory of justification is to check its implications about particular cases. A complete theory of justification will have implications about every particular case. The implications of a good theory of justification will also match our intuitive judgments about each case. The implications of a bad theory will not. That is, a good theory will typically imply that ordinary people, in ordinary circumstances are justified in believing what clearly trustworthy people tell them, in believing what their senses tell them about the world, in believing what seems to them to be the best explanation of what they have to go on and so on. A bad theory will not have all these implications and will imply that in some of these circumstances believing what is commonsensical is unjustified.

The circumstances of concern in this article all involve memory. The next sections cover particular kinds of circumstances that help test the implications of theories of justification. Think of each kind of circumstance as introducing a problem for these theories. If externalists are correct, and their view indeed has an easier time accommodating our intuitions and thereby solving these problems, then internalism is in bad shape. If, however, internalism can solve these problems easily enough, then it is much better off than many externalists suppose.

After introducing the problems we will consider the main responses to them. Of course, these are neither the only problems memory poses to theories of justification, nor the only responses. They are just the ones that have received the most attention.

i. The Problem of Forgotten Evidence

We are forgetful. We forget email passwords, where we put the car keys, anniversaries, acquaintances’ names and more. In some cases this gets us into trouble and in other cases it is harmless. Interestingly, when we do not forget and we keep beliefs about these things, we often nonetheless forget our original evidence for our beliefs. I cannot recall how I learned that Fred’s name is “Fred”–did a trustworthy friend tell me? Did Fred himself tell me? And you know that your email password is “iluvphilosophy,” but you cannot remember choosing it all those years ago. That password just seems familiar and using it works.

Forgetting is an epistemologically significant phenomenon. Here is one reason for that. In many cases, it seems that when we keep a belief, yet forget our original evidence for it, the belief remains justified. But this appears to conflict with certain theories of justification. In particular, it apparently conflicts with evidentialism, the view that the justified attitude for a subject toward a proposition is the attitude that fits the subject’s evidence (see Conee and Feldman (2008, 2011), Feldman and Conee (1985) and McCain (2014)). Understood broadly, your evidence is what you have to go on–your experiences, thoughts, feelings, background information and so forth. Evidentialism implies that if you are justified in believing that Fred’s name is “Fred”, then believing that fits what you have to go on. If you lose crucial evidence, however, believing that Fred’s name is “Fred” may no longer fit your evidence.

The Problem of Forgotten Evidence is the problem of accommodating our intuitions about justification, in cases where key-supporting evidence has been forgotten. There seem to be a lot of cases of this sort; we regularly forget our original evidence, while retaining the belief. Gilbert Harman (1986) is typically credited with developing this problem, though he never called it the “Problem of Forgotten Evidence”.

Which theories face the Problem of Forgotten Evidence? As mentioned above, evidentialism faces it. Traditionally, evidentialism has been understood to be a form of internalism. As a result, philosophers have understood the Problem of Forgotten Evidence to be a problem only for internalist theories of justification (for instance, Bernecker (2008, 2010)). But there are some evidentialist forms of externalism (see Comesaña (2010) and Goldman (2011)). These theories do not quite understand evidence to be all that you have to go on. Rather, evidence is understood more narrowly: it is just the stuff you have to go on, such that beliefs formed on its basis tend to be true (where contingent environmental factors partly determine what tends to be true).

So, some forms of externalism face the problem; evidence, even on the narrower understanding, can be forgotten. The problem also challenges any theory of justification that states that S’s having evidence for p is necessary for S’s being justified in believing that p. Some non-evidentialist externalist theories state roughly this necessary condition (see Alston (1988)). And finally, while the Problem of Forgotten Evidence is stated in terms of forgetting evidence, there is a more general problem here: how do we accommodate our intuitions about justification in cases, where whatever it is that originally conferred justification (be it evidence or something else) is forgotten? It could be that most theories of justification face this more general problem, which is discussed prior to Harman (1986) by George Pappas (1980).

Any theory that faces, but cannot solve, the Problem of Forgotten Evidence is doubtful. It is important to consider, then, possible solutions to the problem and to consider which theories have solutions available. Before considering these matters, two related problems about memory and justification should be mentioned.

ii. The Problem of Forgotten Defeat

Unfortunately, we forget more than just our original reasons for believing. We also forget our defeaters, that is, our reasons for not believing, or for doubting our reasons for believing. Sometimes we remember our original reasons, yet forget our defeaters. You remember your original reason for believing that there are boxes in the basement: this morning you saw what looked to you like boxes, in what looked to you like the basement. But suppose your spouse tells you that the children have since taken all of the boxes out of the basement, in order to build a fort outside. Or, your spouse tells you that you did not in fact see boxes in the basement–you saw them in the attic. If you forget what your spouse told you, yet you retain your belief that there are boxes in the basement, you have forgotten a defeater for your belief. On some theories of justification, your belief can still count as justified.

Another kind of forgotten defeat is this. Suppose you never had any reason to believe that there are boxes in the basement, but you believed it anyways. Some theories will count this belief as justified, once you forget that you never had any reason for it. Some philosophers find this result unacceptable (see Annis (1980), Goldman (1999, 2009), Greco (2005) and Huemer (1999)). The Problem of Forgotten Defeat is the problem of accommodating our intuitions about justification in cases where key-defeating evidence has been forgotten.

Far more theories face the Problem of Forgotten Defeat than face the Problem of Forgotten Evidence and that is one of the reasons why it is worth distinguishing these problems. Often these problems are conflated–in fact, the former problem has never been given a name before. The reason that many more theories face the Problem of Forgotten Defeat is this. Just about every theory of justification–even theories that deny that some evidence can play a justifying role–grants that some evidence, understood broadly, can play a defeating role. That is, nearly all theories agree that, even if having evidence cannot by itself justify, having evidence can by itself eliminate justification. Your visual experience of the cookie on the table is part of your evidence that there is a cookie on the table. Non-evidentialists will deny that your evidence on its own justifies believing that there is a cookie on the table. But typically they would grant that your evidence at least partially defeats any justification you had for believing that there is nothing on the table. So, cases of forgotten defeat challenge both evidentialist and non-evidentialist theories, although philosophers (for example, Annis (1980), Goldman (2001, 2009), Greco (2005) and Huemer (1999)) have presented the problem as though only internalist and evidentialist theories face it.

iii. The Problem of Stored Beliefs

The final problem centers on beliefs that are merely stored. (Some philosophers instead call these beliefs non-occurrent or standing or dispositional.) These are beliefs that are in no way before the subject’s mind. The believer is not thinking about, reasoning from, acting from, or having an experience concerning them or their content. Contrast these with occurrent beliefs, which are before the subject’s mind. When you are remembering that Plato taught Aristotle, or are telling others about it, your belief that Plato taught Aristotle is occurrent. At most other times–when you are sleeping, driving, playing chess, washing dishes–that belief is merely stored in memory. (This seems true on the storehouse model of memory, at least; on a generative model you may lack the belief of these other times).

A belief can be both occurrent and stored, just as a song can be both playing and stored on your computer. A merely stored song is stored but not playing. Similarly, a merely stored belief is stored, but not occurrent. It is commonsensical to attribute countless stored beliefs to people, who are in normal circumstances. A few moments ago, you had beliefs about chemistry, the first U.S. President, your childhood, panda bears, the Indian Ocean, the Super Bowl and countless other topics. A few moments ago almost all of these beliefs were not just stored, but were merely stored. And it is plausible that many of these beliefs were justified a few moments ago. The Problem of Stored Beliefs is the problem of explaining how the merely stored beliefs that seem justified are indeed justified (for simplicity, the discussion below for the most part omits the ‘merely’).

Thomas Senor (1993) and Alvin Goldman (1999) influentially pose this as a special problem for internalism about epistemic justification (though George Pappas (1980) briefly discusses the more general problem even earlier). Goldman (2011) and Matthew McGrath (2007) target internalist evidentialism in particular. Our occurrent experiences, thoughts and feelings might justify some of our stored beliefs, but not nearly enough. Our active mental lives, at any given time, simply do not bear on most of our stored beliefs. As a result, internalism appears unable to explain how all justified stored beliefs are justified. The same goes for evidentialism, since our evidence is too constrained to fit all our justified stored beliefs.

Andrew Moon (2012) directs a knowledge-version of the Problem of Stored Beliefs toward an evidentialist view concerning knowledge. The evidentialist view is that S knows that p at t only if S believes that p on the basis of evidence at t. We have stored beliefs while we sleep and we know many of these believed propositions. But while we sleep, these beliefs have no evidential basis, so knowledge does not require an evidential basis. Though Moon’s argument concerns just knowledge, we can offer a parallel argument that concerns justified belief. If his original argument is sound, then the parallel argument is too and so justified belief does not require an evidential basis.

Of course, externalist and non-evidentialist theories also face the Problem of Stored Beliefs. But these theories can avail themselves of non-mental, non-evidential resources, so they appear to have an easier time solving the problem. The next section reviews some of these resources.

It is important to distinguish the Problem of Stored Beliefs and the Problem of Forgotten Evidence. The phenomenon of forgetting is essential to the latter problem, but not to the former. We can store in memory our original evidence for a justified, merely stored belief. So there is no relevant forgotten evidence here, but some questions remain: what evidence could justify the belief? Is it the evidence that is stored in memory? How could it justify when it is not accessed? And the phenomenon of having stored beliefs is not essential to the Problem of Forgotten Evidence, but it is obviously essential to the Problem of Stored Beliefs. We can forget the original evidence for a belief that remains occurrent: if I am distracted and exhausted when we meet at a bustling party and you tell me that you are from Santa Fe, I might form the belief that you are from Santa Fe, but immediately forget that you just told me so. I might even be slightly puzzled as to why I find myself believing that you are from Santa Fe. My belief was justified when formed, but what justifies it a moment later, when I have forgotten my evidence?

It is clear, then, that the Problem of Stored Beliefs and the Problem of Forgotten Evidence are dissociable. Consequently it is a mistake to assume that they must share a solution. And it is possibly misleading to introduce the two problems simultaneously with a single example, as some philosophers do, without distinguishing them (see Goldman (2011), for instance). Doing so invites conflation of the problems.

b. Responses

The three problems discussed above are challenging. Tackling them has, however, helped inspire novel epistemological theses and observations about memory, some of which are general and may solve multiple problems, while others are more piecemeal and particular. This section looks first at the more characteristically evidentialist or internalist responses to the Problem of Forgotten Evidence and the Problem of Stored Beliefs and then at the more ecumenical responses. Replies to the Problem of Forgotten Defeat follow.

In answer to the Problem of Forgotten Evidence, Earl Conee and Richard Feldman (2001) point out that in ordinary cases, even when all of S’s original evidence for p is lost, S still has a host of evidence that could justify her in believing that p. This evidence could for example be rooted in induction, background information about memory or conscious recollection. You have forgotten why you originally believed that Fred’s name is “Fred”. But you have reason to believe that you tend to form beliefs with good reason, so you have evidence that you originally had good reason for your belief and this supports the belief. And you have reason to believe that your memory is fairly accurate. Since memory is supplying your belief about Fred’s name, you have justifying evidence for it. And if you are consciously recollecting that Fred’s name is “Fred”, then your experience is displaying that proposition as true, just as perceptual experiences display propositions about the external world as true. So, evidentialists of any stripe (internalist or externalist) can claim that there generally is justifying evidence in the central cases that motivate the Problem of Forgotten Evidence.

However, we do not usually have all of this evidence for a belief that is merely stored. A merely stored belief, by stipulation, is not being consciously recollected. Hence, the Problem of Stored Beliefs remains. Feldman (1988) and Conee and Feldman (2001) propose that justified stored beliefs can have “stored justifications”; S can recall some justifying evidence for p, when S has a justified stored belief that p. S’s evidence for p is stored (compare McCain (2014)). On this view, justified stored beliefs typically are not justified in the most fundamental sense, in the sense that justified occurrent beliefs typically are. When justified in the most fundamental sense, not all of the justifiers are stored, but rather some justifiers are occurrent: experiences, inferences and so on. If it is plausible that justified stored beliefs have the most fundamental kind of justification, then Conee and Feldman’s proposal will not solve the Problem of Stored Beliefs.

On a closely related proposal, the evidence and justifiers are occurrent. Call the proposal dispositionalism: dispositions of the right sort can justify (see Audi (1995), Conee and Feldman (2011), and Ginet (1975)). These dispositions can be memorial. Maria is disposed to recollect that she went to high school in Santa Fe. With the right cue, in ordinary circumstances, she will recollect that fact about her past. On dispositionalism, this disposition justifies her in believing that she went to high school in Santa Fe. The disposition is only occasionally manifest–she only occasionally thinks about where she went to high school–but she nonetheless has the disposition right now; it is not simply stored. As a result, the disposition can epistemically justify in the most fundamental sense right now. In some ways, dispositionalism parallels virtue ethics, which claims among other things that a virtue is a disposition that morally justifies certain actions, even when the disposition is not manifest.

Dispositionalism offers a promising solution to the Problem of Stored Beliefs. It also could solve the Problem of Forgotten Evidence: typically in cases, where S has forgotten her original evidence for her justified belief that p, S still has a disposition to recall p as known or as true. If this disposition justifies believing that p for her, then the Problem of Forgotten Evidence may disappear.

However, dispositionalism still needs crucial development. More must be said about exactly which dispositions justify believing exactly which propositions and how; and it would be good to have a principled way of determining which dispositions a given subject has, in order to see whether dispositionalism attributes to the subject justification for believing just the right propositions.

Conee and Feldman (2001) offer starting material for a final internalist, evidentialist-friendly solution to the Problem of Stored Beliefs. If we have stored beliefs, then these beliefs can justify other beliefs, including other stored beliefs. We can direct this proposal at the Problem of Forgotten Evidence too: stored beliefs can justify a belief, for which all original evidence has been forgotten.

A worry for this proposal is that we may not have enough stored beliefs to solve the two problems. We may not, in other words, have enough stored beliefs that could justify all justified stored beliefs and all beliefs that lack their original evidence. Goldman (2009) voices another worry: what ultimately justifies any stored belief? If a belief that p is justified by a stored belief that q, the latter belief should be justified too. It is hard to see how an unjustified belief can by itself justify another. But what justifies the belief that q? Does a stored belief that r justify it? If so, what justifies this stored belief that r? And so on.

A moderate form of coherentism could address Goldman’s worry: if S’s belief that p coheres with certain of S’s other beliefs, then S’s belief that p is justified. Coherence among stored beliefs can justify them. And coherence can justify belief in the face of forgotten evidence. Beliefs can have a special, mutually supporting relationship. However, coherentism has its costs; see Coherentism. But perhaps it could, if suitably defended, substantiate Conee and Feldman’s proposal. Still, if any stored belief is justifiedly based on something other than beliefs, then coherentism, even if correct, does not fully solve the Problem of Stored Beliefs (compare Moon 2012: 316-7).

The remaining responses to the problems are also available to externalists and non-evidentialists. A view nearly universally endorsed by discussants of the problems is what we might call preservationism (see Annis (1980), Bernecker (2008), Burge (1997), Goldman (2009, 2011) Naylor (2012), Owens (2000), Pappas (1980) and Senor (2010); some philosophers use ‘preservationism’ to refer to the view called ‘anti-generativism’ below). Roughly put, memory preserves the justification of the beliefs it preserves. More precisely, if S is justified in believing that p at t1, and retains in memory a belief that p until t2, then at t2 S’s belief that p is prima facie justified. (The ‘prima facie’ here allows that the belief may not be justified overall if there are defeaters for it.) Your belief that Plato taught Aristotle was justified when you formed it: a professor or some other clearly credible source told you that Plato taught Aristotle. And you have kept that belief ever since. So, your belief has ever since been justified.

Preservationism seems to provide a simple solution to the Problem of Stored Beliefs. Regardless of whether a belief is stored rather than occurrent, it can retain its justification as long as memory preserves it. A stored belief can inherit justification from the past and this appears to solve the problem. And forgetting evidence does not block the inheritance. So, preservationism appears to solve the Problem of Forgotten Evidence. In fact, a main motivation for preservationism is that it seems to solve these problems at no cost.

But is preservationism true? Externalists think that it is true only if certain features that are external to the mind obtain. Process reliabilists, for example, think that preservationism is true just in case memory is reliable. Process reliabilism is roughly the view that justification of a belief depends entirely on the reliability of the process that forms or retains the belief. According to preservationism, beliefs retain justification by being retained in memory. As a result, reliabilists think memory must be reliable, in order for preservationism to be true. Since it is contingent whether memory is reliable, on reliabilism it is contingent whether preservationism is true. Reliabilists, who appeal to preservationism, in order to solve the Problem of Stored Beliefs and Problem of Forgotten Evidence, bear the burden of showing that memory is reliable.

And, if the storehouse view of memory is indeed incorrect, then preservationism appears vacuous unless modified. If memory typically alters the information entering it, it is hard to see how memory could preserve many beliefs over time–the beliefs would seem to be destroyed, once their exact content is no longer represented in memory. Preservationists, who reject the storehouse view, must explain either of two things: first, how memory can nonetheless tend to preserve beliefs, even though it tends to modify the content that enters it; or second, how something other than memory preserves beliefs. Pursuing either option may require developing a novel theory of belief.

Now for replies to the Problem of Forgotten Defeat: Richard Feldman (2005) and Matthew McGrath (2007) in a sense deny that this problem exists. When a defeater is forgotten, it is no longer relevant to what one is justified in believing. Once you forget that your spouse told you that the children removed all of the boxes from the basement, your spouse’s testimony ceases to defeat; you are overall justified in believing that there are boxes in the basement, as long as you still have some support for believing that.

Feldman and McGrath press their point: some attitude toward the proposition that there are boxes in the basement must be justified for you. But which? Abandoning belief in the proposition seems unjustified, since you no longer have reason to abandon your belief. And suspending judgment in the proposition seems unjustified, since you still have some justifying support for believing it–for example, a vivid recollection of what looked like boxes in what looked like the basement. The only potentially justified attitude remaining for you is belief. It is hard to see a competing option. If that is correct, then forgetting defeaters poses no problem. Nothing, which is forgotten, can defeat. (Of course, something other than the original defeater can still defeat. If for example you recall that you have forgotten a defeater for p, but cannot recall what it was, then arguably you still have a defeater for p: you have reason to believe that you had reason to doubt p. Having reason to believe this is itself reason to doubt p.)

Why, then, suppose that there even is a Problem of Forgotten Defeat? Why suppose that forgotten defeaters remain at all relevant to justification? The main reason is this: many philosophers think that memory, unlike perception, testimony, rational intuition and reasoning, is not a generative source of justification. Memory cannot create or strengthen justification. Rather, memory at most preserves justification that has been acquired from some source (such as perception, testimony and so on). Call this thesis about memory anti-generativism. It is a “garbage in, garbage out” view of justification and memory. An unjustified belief that enters memory remains unjustified, unless new reasons for the belief are acquired from some faculty other than memory. Anti-generativism is traditional and popular (see Annis (1980), Goldman (2009, 2011), Owens (2000) and Senor (2007)), and so are variants of the view that concern knowledge or warrant (see Audi (1997), Burge (1997), Dummett (1994) and Plantinga (1993)). With respect to knowledge, many philosophers think memory and testimony are alike in this way: coming to know that p via testimony requires that the testifier knows that p; testimony does not generate knowledge from non-knowledge.

Sometimes anti-generativism is called “preservationism”, but this is infelicitous. Anti-generativism primarily states a limit on memory: memory does not generate justification, or knowledge, or anything similar. The theory does not centrally concern memory’s power to preserve anything (unlike the theory that is called “preservationism” above, which does centrally concern memory’s preservative power).

If anti-generativism is plausible, then the theories of justification that are compatible with it may avoid the Problem of Forgotten Defeat and theories that are incompatible with it may on that account face the Problem of Forgotten Defeat.

However, generativism, the view that memory can generate justification, is increasingly common (see Audi (2002), Bernecker (2010), Lackey (2005, 2007), Huemer (1999), Michaelian (2011a) and Owens (1996)). Arguments for this view reveal that it comes in several forms. Jennifer Lackey (2005, 2007) and Sven Bernecker (2010) join ranks with Feldman and McGrath in thinking that memory generates justification in cases of forgotten defeat. But notice that the justification generated in these cases is overall, not prima facie. That is, since memory is responsible for the loss of a defeater, memory results in a balance of justification that favors belief. This is not yet to say that memory is creating new reasons for belief. One generativist view, then, is that memory can generate overall justification, even if it cannot generate prima facie justification.

Lackey offers other support for generativism: a subject’s memory can store information, which, in the past, the subject never paid attention to. If the subject recalls and attends to the information afterward, the subject can use it to form justified belief. The basis of this belief would be memory. Lackey builds her support with an example. Suppose that Clifford has his mind on many things, while he is driving. Later, his friend Phoebe asks him whether construction on the freeway has begun. Clifford then recalls seeing construction on his recent drive and only then forms a belief that construction on the freeway has begun. His belief is justified and memory is its source. Generativism follows. Still, as Bernecker (2010) observes, if Lackey is correct, she has only supported the generativist view that memory can generate doxastic justification. She has not shown that memory can generate propositional justification. In other words, at best Lackey demonstrates that memory can generate a reasonable belief, not that memory can generate new reasons for believing. Memory merely based a belief on a reason that perception generated.

Huemer (1999) and Michaelian (2011a) endorse the stronger thesis that memory can generate new reasons for believing. Huemer thinks that S’s seeming to remember that p can produce reason for S to form a belief that p, regardless of whether S already had reason to believe that p. And Michaelian attacks the storehouse view of memory, arguing that memory can generate new content and new belief in that content. The belief can have justification when formed as long as certain external conditions are in place (and Michaelian thinks they are). Consequently, sometimes, when memory generates justified belief, it generates justification for believing.

Since anti-generativism is controversial, the severity of the Problem of Forgotten Defeat is unclear. Interestingly, although it is primarily externalists who find the problem to be severe, the mix of internalist and externalist advocates of generativism is fairly even.

4. Memory and Skepticism

So far the surveyed discussion has assumed that memory plays some role in our actually having justification and knowledge, and the discussants have simply debated the margins of that role. But many early and mid-20th century epistemologists worried about this assumption. Why believe memory has an important, or even any, epistemic role? Since this question may invite skepticism, call it a skeptical question for simplicity. Satisfactorily answering this sort of skeptical question about memory is a fundamental epistemological problem. In fact, according to Richard Fumerton (1985), answering it is the most fundamental epistemological problem. If memory has no epistemic role, then we have no reason to believe just about anything we ever learned, or think we learned, at any time in the past.

What is more, memory appears to be involved not just in our retaining what we have learned, but in our very learning. When Chloe tells you “I am changing the oil in my car today,” you use memory even to understand what she is saying–some memory system is responsible for your applying the concepts that make “changing” and “oil” and “car” (and so on) intelligible to you. And you use memory not just to grasp the meaning of words, but also of sentences. Memory is holding fixed in your mind the beginning of Chloe’s statement when she finally says the word “today,” allowing your mind to string concepts together in a way that yields in you a mental representation of what she has testified. Without memory there is no understanding of what is testified. If memory has no epistemic role, then it is hard to see how we could even learn from testimony in the present. Memory seems similarly involved in intuition, reasoning, introspection and perception. Accordingly, it is hard to see how we could learn from those sources if memory plays no epistemic role.

Philosophers have sharpened the general skeptical question about memory into more challenging related sub-questions. This section discusses responses to two of these sub-questions. Answering them is not easy, since they introduce foundational problems that do not arise with other kinds of skepticism. Yet, oddly, philosophers exploring contemporary skepticism have mostly neglected the issue of memory skepticism. Half way through the 20th century C. I. Lewis (1946) thought the issue was so significant that the level of silence on it even then was “a bit of a scandal.” And the times have not changed.

a. Memory and Accuracy

Consider

(MR) Memory is reliable.

MR states that memory tends to get things right and that it is generally accurate. It does not state that memory is perfectly accurate. A first skeptical question is: why believe MR? If there is no reason to believe MR, then memory may not provide (either by preserving or by generating) any support for what it represents as true in a given case. If you have no sense as to whether, say, a particular political blog tends to get things right, then there may be no sense in believing anything on the mere basis of the blog. Don Locke (1971) thinks that if we have no reason to believe MR, we have no knowledge via memory at all. If he is correct, it may be critical that we identify support for MR.

It is not at all clear that Locke is correct. But even if he is, our troubles are not as severe as they might seem. Suppose we have little to say positively in answer to the first skeptical question. We may still have reason to believe that memory is often correct (correct, say, around 40% of the time), and even that in the kinds of cases we care about it is usually correct. Further, having no reason to believe MR is not the same as having reason to believe MR is false. Having no reason to believe MR may just require us to be neutral about it. Granted, process reliabilists, who must be neutral about MR, may be in trouble. They may have to suspend judgment about whether any given belief that memory preserves is justified, since they must suspend judgment about whether such a belief is preserved by a reliable process. But on other theories of justification, perhaps we remain reasonable in thinking that memory justifies.

Still, it would be somewhat troubling if there were no reasons to believe MR. It would be strange for us to rely so heavily in our reasoning and behavior on something we have no reason to believe is typically accurate. It is worth considering MR’s status for us.

Locke considers the following line of support: doubting MR is self-defeating. To raise doubts about MR requires the use of memory. Raising relevant doubts requires citing examples, in which memory has erred. But memory alone can supply these examples. If these examples impugn MR, it is because memory supports believing something: the fact that it has erred in certain cases. So, the mere attempt to undermine memory itself vindicates memory in a way.

This result is not clearly worth celebrating. We have merely established that memory supports believing that it itself fails here and there. And even if this result is established, it yields no support for MR. Memory may sometimes support belief, but it could nonetheless typically fail to support belief and could be unreliable. But the self-defeat consideration reveals something unique about memory skepticism: using or contemplating arguments for or against it requires the use of the very faculty being scrutinized. We cannot help but use memory, in order to explore memory skepticism. Nothing parallel is true about, say, external-world skepticism. It is not the case that thinking about or offering an argument for and against it must occur via our perceiving something external. Thus, memory skepticism is especially thorny: addressing it uniquely, unfailingly involves some kind of circularity. Thomas Senor (2010) claims that there is no non-circular “demonstration” of MR. If this is true, any demonstration of MR may be suspect.

Richard Brandt (1955) offers an alternative line of support: MR is the best, and only, explanation of our data. What are our data? For Brandt it is our present experience and our having a host of cohering beliefs about the past and about science. According to Brandt (1955: 93), we have these beliefs, because our brains have over time interacted with the world in a truth-conducive way and “the only acceptable theory is one which asserts that a large proportion of our memory beliefs are veridical. No alternative to such a theory has been proposed; nor can one imagine what one would be like.” If MR is the only explanation of our data, MR is by default the best explanation and it may thereby be credible. Contra Senor and others, Brandt thinks this support for MR is non-circular, since it does not take for granted that any recollections are accurate.

But there is reason to think Brandt’s support for MR is indeed circular. To support MR, Brandt makes an explanatory inference based on our data. But why suppose we have the very data he thinks we have–why suppose we have a host of cohering beliefs? That we have them is not wholly manifest to us at one time. We must use memory, in order to appreciate it. We think about our various beliefs, how they fit together, how snug that fit is and we make an inference about how our beliefs cohere. This thinking and inferring is not instantaneous. It unfolds over time and (we presume) memory holds fixed and supports the parts that (we also presume) have already unfolded. So there is a kind of circularity: memory is used in establishing the data MR allegedly explains (see BonJour (2010: 169-171) and Plantinga (1993: 61-4)). If this circularity is vicious, Brandt’s argument yields no new reason for believing MR.

Another objection to Brandt’s argument is that MR is not the only explanation of our data. Bertrand Russell (1921/1995) provides a famous rival hypothesis: we and the world came to exist only five minutes ago and it merely appears that everything is much older. In each of us is a package of cohering beliefs about the past. And we find rings in trees, rust on cars and ruins in Rome. All of this is misleading. Everything is new. As unpalatable as this hypothesis is, it is not easy to disprove. At any rate, it is a rival explanation of our data. Oddly, Brandt actually considers a Russellian hypothesis, but dismisses it as a fantasy, wholly lacking “evidential foundation”. But there is no need for evidence for Russell’s hypothesis, beyond this: it fits the data. Since it does, MR has an explanatory rival. We cannot assume MR is the best explanation. We must do the hard work of showing it is better than Russell’s hypothesis.

b. Memory and the Past

Our target has shifted from defending MR to defending something more basic. Memory could be massively misleading. For any view about the past, why suppose it is even approximately right? That is our second skeptical question. The first skeptical question challenges our view about how memory performs overall. It still allows that memory provides reason to accept some appearances about the past. The second question goes further, probing each appearance. It challenges our view about memory’s performance in each given case. Answering this question well is especially demanding.

Senor (2010) claims that most philosophers agree that Russell’s hypothesis has not been refuted. Regardless of whether Senor and these philosophers are correct, note that the demand here is greater than just refuting the particular hypothesis that Russell offered. Russell’s exact hypothesis may be bad: it seems ad hoc and uninformative. The present demand is to show why all hypotheses like Russell’s are inferior. One hypothesis similar to his is that the world and its inhabitants all popped into existence six minutes ago. Another is that the world is as old as it seems, but just its inhabitants popped into existence five minutes ago. In order to reasonably hold our commonsensical beliefs about the past, we must have reason to reject each skeptical hypothesis that is incompatible with the truth of our commonsensical beliefs. Moreover, we must have reason to think that what we commonsensically believe explains better our data than the entire disjunction of skeptical hypotheses does.

Russell proposes a pragmatic answer to the second skeptical question. Taking memory appearances at face value is extremely practical. We cannot help but do it and it works. Skepticism, therefore, poses no genuine threat. This answer appeals to something like the practical rationality of believing that the past really is how it seems. But this answer tells us nothing about the epistemic rationality of believing anything about the past. Even if Russell is right, we do not have on that account a key ingredient for knowledge of the past: epistemic justification.

One family of replies to the second skeptical question uses transcendental arguments to reject Russell’s hypothesis. A transcendental argument is of this form: A obtains; A is impossible in the absence of B; (therefore) B obtains. Norman Malcolm (1963) and Sydney Shoemaker (1967) offer the following transcendental argument: we know how to make past-tense statements; this competence requires that most of these statements are true; (therefore) most of these statements are true. Since these statements express our beliefs about the past, most of these beliefs are true. Not only does MR follow, but it also follows that the past tends to fit our expressed views about it.

The general idea behind this argument is that one’s having skill at using a kind of statement is incompatible with one’s systematically misusing it. If Elmer sincerely refers to toasters, clouds and orange things as “rabbits,” then Elmer must not be using that word to talk about rabbits. There must be an alternative way of understanding his “rabbits” expressions, such that they tend to be true. Now, we are competent at making statements about the past. It follows that most of these statements are true and so are our corresponding beliefs.

Don Locke (1971: 135-7) offers a transcendental argument for MR (compare Lewis (1946)), which may also answer the second skeptical question. The fact that we have knowledge at all and that we inquire, requires that we have memory knowledge. And we in fact know things and we in fact inquire. (In support of this claim we might note that it seems readily proven: are we inquiring? Yes!). So there is memory knowledge. And, as noted earlier, Locke thinks that if there is memory knowledge, then MR is true. So, he concludes that MR is true. If Locke is right, it follows that Russell’s hypothesis is false. The world did not come into existence five minutes ago. Many hypotheses like Russell’s will also be false. This may answer the second skeptical question. Our reason to suppose that a given belief about the past is true is that it is of a class of beliefs that tend to be correct.

Some philosophers doubt that transcendental arguments can rationally support any anti-skeptical conclusions. But even if some can, the transcendental arguments covered here are questionable. Malcolm and Shoemaker take it as a datum that we know how to make past-tense statements. But why accept the datum? In answer, we can at best cite the kinds of statements we can recall ourselves competently making. And why suppose that the past resembles those recollections? If we popped into existence five minutes ago, those recollections are misleading. If the transcendental argument simply assumes that the recollections are accurate, then the argument fails to generate support for believing that they are accurate.

Similarly, in reply to Locke: why suppose we inquire? You might blush with embarrassment and note that to ask that question is to inquire. But why suppose a question has been asked? Observing inquiry may rely on memory. Perhaps we cannot even think at all or observe a case of inquiry all in one moment. Perhaps thought and observation are always extended in time and we may need to use memory, in order to observe the temporal extension of anything.

As noted, a transcendental argument is of the form: A obtains; A is impossible in the absence of B; (therefore) B obtains. The replies to the transcendental arguments here question the first premise. To support anything as data, we may need to use memory. If that is correct, it may seem viciously circular then to use this data, in order to support either memory or beliefs about the past.

However, one might think that this reveals that memory skepticism is indeed self-defeating. Merely raising a skeptical challenge to MR or to views about the past uses some data about memory or the past. This data may include the fact that observing inquiry requires memory, or that Russell’s hypothesis is compatible with one’s having a given recollection. But if we need to use memory in order to support any data, then raising a skeptical challenge about memory uses memory. So, anyone who offers such a challenge undermines her own position. If memory truly supported nothing, skepticism could have no support.

This line of reasoning notes a conflict between an activity (supporting memory skepticism) and a theory (memory skepticism). Unfortunately, even if there is a conflict, the theory may still be correct (compare Bernecker (2008: 130-1) and Fumerton (1995: 52)). Why believe memory skepticism is false? Even if supporting memory skepticism is self-defeating, it may still be true. And, we may still be justified in believing memory skepticism, but simply unable to demonstrate its support.

Finally, Sven Bernecker (2008: 131-3) attempts to “disarm” Russell’s hypothesis and skepticism about the past by taking a relevant alternatives approach (see Contextualism in Epistemology). Bernecker thinks memory can provide us with knowledge about the past, even if we do not know that there is a past and even if we do not know that Russell’s hypothesis is false. Here is why: a table can be flat, even if it appears bumpy under a microscope. The table is not relevantly bumpy, so it counts as flat. Bernecker thinks knowledge is similar to flatness. S’s knowing that p does not require that S is able to know every alternative to p to be false. S might know that p and yet be unable to rule out some situation in which not-p is true. All S must be able to rule out are the relevant alternatives to p–the relevant situations in which not-p is true.

For example, in order for you to know that Plato taught Aristotle, you must be able to rule out the relevant alternatives to that fact. One relevant alternative is that Socrates alone taught Aristotle. And you can rule this out: you have reason to believe that Socrates swigged his poisoned hemlock years before Aristotle’s birth. Although Russell’s hypothesis is an alternative, to what you believe about the past and you may be unable to rule out, Bernecker thinks it is ordinarily an irrelevant alternative. So memory can provide knowledge of the past even when you cannot rule out Russell’s hypothesis.

Bernecker’s reply faces difficult objections. It is not obvious that knowledge is sufficiently like flatness. Supposing it is, it is unclear that Russell’s hypothesis is ordinarily irrelevant. And, supposing it is, why agree that we can rule out the alternatives that are relevant? What enables you to rule out that Socrates alone taught Aristotle–evidence from memory? The strength of this evidence should be in question if Russell’s hypothesis is not yet ruled out. But, supposing we can rule out the relevant alternatives, Bernecker’s reply may leave us unsatisfied. At best it secures for us bits of knowledge about the past, yet it does not secure knowledge that the past exists or knowledge that Russell’s hypothesis is false. The latter two results seem simply to concede victory to an unpalatable skepticism. And they pair oddly with the former result–how could we simultaneously have knowledge about the past from memory and yet lack knowledge from memory that the past exists? Whatever ultimately explains the one, suggests the other is false.

It is clear that satisfactorily answering the skeptical questions is not easy. There have been other attempts to answer them, but none more promising or developed than those mentioned here (for additional discussion, see Locke (1971) and Bernecker (2008)). Since memory skepticism threatens most of our knowledge and justification, failing to rule it out would be uncomfortable. Still, for two reasons it would be premature to despair. First, even if we cannot show that memory skepticism is false, it is unclear what is thereby threatened or what we are thereby required to believe, if anything. This is because even if memory skepticism is true, it is unclear what we can conclude (compare BonJour (2010: 170-1)). If memory must support any justifying inference or data about the past and memory cannot support, then what can we are justified in inferring from the truth of memory skepticism? It is hard to say. Second, we should not confuse our failing to disprove memory skepticism with our having no reason to believe anything about the past or having reason to deny MR. It could very well be that memory is reliable and justifying, but that we simply have a hard time showing it.