Knowledge of the Past

1. Some Preliminary Issues and Distinctions

1.1 Sources of Beliefs about the Past

What are some of the different ways in which one can arrive at beliefs about the past? One very familiar way is via memories of one's own experiences. But one also learns about the past through the present testimony of others. These two sources, however, on their own, can provide one only with beliefs about experiences that presently existing people have had, and events that they have witnessed.

A third source of beliefs about the past that can take one beyond information about the experiences that presently existing people have had, or about events that they have witnessed, consists of traces of the past. Here the idea is that beliefs about present states of affairs can be combined with information concerning scientific laws to enable one to draw conclusions about earlier states of affairs that have given rise to the present states of affairs.

The term "trace" is sometimes used to refer only to relatively complex and intrinsically improbable states of affairs that can best be explained by postulating appropriate past states of affairs. Consider, for example, a footprint on the beach, or fossils, or a photograph. But very simple states of affairs - such as the existence of an electron - can provide one with a good reason, given relevant conservation laws, for concluding that something with certain properties - such as a certain mass and electrical charge - existed at a slightly earlier time.

Fourthly, in addition to memories of one's own experiences, there are also memories that are not about one's own past experiences, and that do not at present rest upon any memories of past experiences. Thus one may remember, for example, that the United States once had a President named "George Washington", but have no memory at all of the experiences involved in one's learning this.

Fifthly, these different sources can be combined in a chain. Thus one may remember someone's telling one that they saw fossils of a certain sort that are evidence of the existence of a certain type of fish: one has a memory of first person experiences of testimony concerning a trace of the past existence of a certain type of animal.

A final source of beliefs about the past is perception, since given that causal processes are involved in perception, and that causal processes take time, any state of affairs that one perceives must be one that exists at an earlier time - and in the case of perception of astronomical objects, quite possibly at a much earlier time. Typically, of course, it is present-tense beliefs that we as a matter of fact acquire in perception, but once the question of what beliefs are justified is raised, it becomes clear that the basic beliefs here should either be beliefs about the past, or else indexical beliefs that are free of tense - such as the belief that that object is (tenselessly) round and very bright.

We have, then, the following sources of beliefs about the past:

(1) Memories of first-person experiences;

(2) Present testimony;

(3) Present traces of the past, including dramatic traces such as films and books;

(4) Memory beliefs that are not themselves memories of first-person experiences;

(5) Chains of the above, such as memories of first-person experiences of past testimony or of past traces, or present traces of past testimony;

(6) Perception.

1.2 Is Memory Primary in the Justification of Beliefs about the Past?

What role does memory play in the justification of our beliefs about the past? First, if we assume for the moment that skepticism is wrong, and we do have knowledge of, or at least justified beliefs about, the past, then it is clear, given the different sources of our beliefs about the past just set out, that it is not true that whenever one has knowledge of some past event, one remembers that event. For, in the first place, there are many past events that one does not oneself remember, but about which one has justified beliefs, because of the testimony of others. In the second place, one has justified beliefs about many events that no one remembers, since one can establish that certain scientific laws obtain, and then use those laws to determine what happened in the past, including both events were observed, but of which no one now has any memory, and events that no one ever observed.

But secondly, even if one can have justified beliefs about past events of which one has no memories, it might well be held that all justified beliefs about the past are either of events that one remembers, or else are beliefs that ultimately rest upon what one remembers. For, as regards justified beliefs whose justification rests upon the testimony of others, it might be argued that one is justified in accepting beliefs on the testimony of others only if one's past experiences supports the belief that the testimony of others is reliable. Similarly, in the case of beliefs that are justified by applying scientific laws to present states of affairs about which one has justified beliefs, it may well seem plausible that one can have reason for holding that the laws in question obtain only if one is justified in believing that events have taken place over time in patterns that conform to the law in question. For if one could not remember past experiments and observations, how could one ever be justified in believing that those scientific laws obtained?

These claims seem to me plausible, but they can be, and have been challenged. As regards the claim about testimony, for example, some philosophers have suggested that, in the absence of evidence either supporting the reliability of testimony, or supporting the unreliability of testimony, testimony possesses at least some prima facie credibility. That is to say, other things being equal, the mere fact that some other person claims that p is the case raises the probability that p is the case, at least slightly.

One reason for rejecting this claim is that I think that it can be show that all of the beliefs that one ordinarily takes to be justified can be shown to be justified without assuming that testimony possesses any prima facie credibility. But, in addition, a person's testifying to anything always takes time, and so the belief that a person has testified that p is the case must always presuppose, either directly or indirectly, beliefs that are justified by an appeal to memory.

If other sources of justification for beliefs about the past - such as testimony and the application of scientific laws - presuppose that memory claims about the past can be justified, can we not conclude that memory is primary with regard to the justification of beliefs about the past? This certainly seems plausible, in the light of the above considerations. However, while it is surely true that virtually all of our knowledge of the past is either identical with, or else based upon, memory knowledge, I believe that it is not the case, for example, that one cannot have any knowledge of the past at all unless one has some memory knowledge. For when we turn to the question of the justification of beliefs about the past, we shall see - in section 7.2 - that there are plausible grounds for rejecting the thesis that all knowledge of the past necessarily presupposes memory knowledge.

One final point. In suggesting that virtually all of our beliefs about the past either involve or presuppose memory claims, I am not suggesting that memory beliefs themselves must be epistemologically basic - that is, must be beliefs that are noninferentially justified. For, as we shall see, there is a very different alternative, according to which it is not memory beliefs, but beliefs about memory beliefs, that are epistemologically basic with regard to the justification of beliefs about the past.

1.3 Memory Beliefs and Memory Experiences

Perhaps the first distinction that should be drawn in approaching the question of memory knowledge is that between memory beliefs and memory experiences. A memory experience is the sort of mental state that one is in when one is consciously remembering something. But a person can also be said, at a given time, to remember something without having, at that time, any relevant memory experience, for he or she may have a memory belief that is not being consciously entertained at that time. In short, a memory experience is a conscious, occurrent state, whereas a memory belief is either a theoretical state or a dispositional state. (Some would construe a memory belief as a disposition to have memory experiences, others as a disposition to engage in relevant external behavior - possibly both verbal and nonverbal - and others as a theoretical state that is potentially causally related to possible experiences, or behavior, or both.)

1.4 Knowledge, and the Nature of Memory Experiences: Images Versus Thoughts

One question that philosophers have discussed is whether memory experiences are to be equated with (1) the mere having of certain images, or with (2) the mere having of thoughts about the past, or with (3) a state that involves both images and thoughts. Traditionally, many philosophers have held that the presence of images was essential. In part, the appeal was phenomenological, with its being claimed simply that one was introspectively aware of memory images. But the claim also reflected, at times, the view that memory experiences must involve images, on the grounds that the latter were necessary in order to provide a basis for memory knowledge.

This latter view reflected a very common tendency to treat memory knowledge in a fashion paralleling perceptual knowledge. Thus, philosophers who held that perceptual knowledge rested upon knowledge of sense data, or sense experiences, often held that memory knowledge rested upon knowledge of memory images. (This tendency is perhaps most vivid in the case of David Hume, though there it is part of Hume's more general view that both beliefs and concepts are to be identified with images.)

Recently, however, many philosophers have argued that even if memory experiences do involve the having of images, this is not of any epistemological relevance: any such images are a mere "phenomenological curiosity." I would agree with this contention that memory images, if they exist, are epistemologically unimportant with respect to knowledge of the past: thoughts about the past, completely unaccompanied by images, can have whatever force is had by the combination of thoughts and corresponding images.

1.5 Knowledge, and Memory Thoughts Versus Memory Beliefs

There is, however, a further question that needs to be addressed here - namely, whether there is an epistemological difference between memory thoughts and memory beliefs. If one is attracted to a foundationalist approach to the justification of belief, one is also likely to be attracted to the view that the foundational beliefs are beliefs about objects with which one is directly acquainted, objects that are immediately given. If so, then one is likely to conclude that memory thoughts must be foundational with regard to the justification of beliefs about the past, since thoughts are states of consciousness, and so are immediately given, whereas memory beliefs need not be objects of immediate awareness.

On reflection, however, this view seems deeply problematic. For on any given occasion, one is consciously remembering at most a very few things, and most of the time one has no memory thoughts, or experiences at all. If one held, then, that one's present memory knowledge is based upon one's present memory thoughts, the upshot would be that most of the time one would have no memory knowledge at all, and so no justified beliefs about the past. This would also imply, unless one adopted a direct realist view with regard to perceptual knowledge, that most of he time one has no justified perceptual beliefs. But even in the absence of the latter implication, the fact that most of the time one would have no justified beliefs about the past would on its own be a very unwelcome conclusion.

Thus it seems that, if one is not to be driven to something close to skepticism with respect to justified beliefs about the past, what one must say is that one's memory knowledge rests, not upon one's present memory thoughts or experiences, but upon one's present memory beliefs.

This conclusion may, of course, be troubling from a foundationalist point of view. Later, however, I shall try to argue that resting the justification of beliefs about the past upon states that need not be immediately given in consciousness at any given time is not incompatible with a satisfactory model of justification.

1.6 Memories of Experienced Events Versus Memories of Facts

Some philosophers feel that it is epistemologically important to distinguish between one's memories of events that one has personally experienced - or, perhaps more precisely: one's memories of one's own experiences - and one's memories of facts not connected with experiences, or experienced events. (Thus, for example, one may remember that Hume was born in 1711, even though one does not remember Hume's being born.) In particular, it has sometimes been contended that, on the one hand, if one apparently remembers having had a certain experience, then one is justified in believing that one did in fact have that experience, whereas, on the other hand, that if one seems to remember that some proposition, unconnected with one's experience, is true, one is justified in believing the proposition only if one can also recall the evidence that supports the belief in question.

I am inclined to agree that there is a difference between event memory and fact memory, but I have doubts about the view that the former, on its own, provides a justification for the relevant belief, while the latter does not. That there may well be some difference can, however, be seen as follows. It might turn out that when one works out the structure of the justification of one's beliefs about the past, it involves two stages:

(1) Given certain memory beliefs, including a memory belief that p, one is justified in believing that one had, in the past, a belief that p. (If 'p' expresses a tensed proposition, this will have to be put in a more complicated way, since a shift from the past tense to the present tense may be needed.)

(2) Given that one is justified in believing that one had in the past a belief that p, together possibly with other evidence, one is justified in believing that p is true.

If the chain of justification turned out to be of this two-step sort, there would be some epistemological difference between event memory and fact memory. For if the memory belief was a belief that one had an experience of a certain sort at time t, one would be justified in believing, according to stage (1), that one believed at time t that one was having an experience at that time. And then, given that one is rarely mistaken about one's experiences, stage (2) would be unproblematic. If, on the other hand, one were dealing with a case of fact memory, the inference at stage (2) would be much shakier, since one knows that one's beliefs in general are mistaken much more frequently than one's beliefs about one's present experiences.

Conclusion

If the structure of justification is something like that sketched above, memories of experienced events will at least provide a better justification for believing in the occurrence of the experiences than memories of facts will provide for the truth of the relevant propositions.

1.7 The Origin of our Concept of the Past

There is a final issue that should be briefly mentioned before we go on to consider the basic issue of the justification of our beliefs about the past. This is the question of the origin - or, more accurately, the analysis - of our concept of the past.

One very natural idea is that the concept of the past can be defined as the concept of what is earlier than the present. (Though this is a rather natural idea, most philosophers who favor a tensed view of the nature of time would reject this suggestion, since they almost invariably hold that the concept of temporal priority must itself be analyzed in terms of tensed concepts, especially those of past, present, and future.) But if the concept of the past is to be analyzed in that way, this immediately leaves one with the question of what account is to be given of the earlier than relation.

Some philosophers have been tempted to treat temporal relations - such as that of one event's being earlier than another - in a fashion paralleling a treatment of spatial relations - such as the relation of betweenness - that seems very natural. Thus, in the case of spatial relations, it is natural to view the relevant concepts as picking out relations that are immediately given, that are directly observable: one acquires the concept of what it is for one thing to be between two other things by acquaintance with things that stand in that relation - either physical objects, or sense data, or parts of one's visual field, etc. But can one plausibly maintain that temporal relations - such as the earlier than relation - are also immediately given, and thus that the relevant terms are ostensively definable?

I believe that this approach to the concepts of temporal relations is exposed to a serious difficulty. For it is natural to hold that a relation can be experienced at a time only if one also experiences, at that time, the things that stand in that relation. So if one is to have, at some time, an experience of the earlier than relation, one must also experience, at that time, the two events that stand in that relation. But won't this be impossible, given that if the events do stand to one another in the earlier than relation, they cannot exist at the same time? It would seem, then, that the earlier than relation is not one that can be given in immediate experience.

But if the concept of temporal priority is a concept that cannot be ostensively defined, how does one come to understand that concept? It must be analyzed, but how? What concepts could the concept of temporal priority be analyzed in terms of?

I do not want to pursue this issue in detail, but here are two possibilities:

(1) One might try to analyze temporal priority in terms of the concept of memory. Here the idea, very roughly, would be to say that event E is earlier than event F if there is a memory belief that is simultaneous with event F and which is a memory of event E - or else, there is some chains of events, each successive pair of which is linked together in the way just indicated.

This approach is exposed, however, to at least two difficulties. First, one wants to say that there are events in the past that are temporally ordered, but where no one does, or ever did, remember the relevant events. In order to assign any meaning to talk about the temporal order of such events, one would have to bring in talk about possible memories of sentient beings who might have existed, but didn't. This, however, seems very problematic, both because there are stages in the early development of our universe where the existence of sentient beings was nomologically impossible, and because, in the absence of events that are already temporally ordered, it seems hard to see how one can supply truth-makers for the counterfactuals involved in talk about possible memories.

Secondly, and more fundamental, how is the idea of a memory belief itself to be characterized without using the concept of the past? The answer would seem to be that there is no way of doing that. But if the concept of the past has to be used, what account can be given of that concept? A natural answer is that the past is by definition that which is earlier than the present. But then we have gone in a circle, as we are back to the concept of temporal priority that we were attempting to analyze in terms of the idea of a memory belief.

(2) A second idea is to try to analyze the idea of temporal priority in terms of the concept of causation. This approach presupposes that the concept of causation does not itself involve the concept of temporal priority - a not uncontroversial claim. In addition, causal analyses of temporal priority are exposed to a number of objections. I believe that those objections can be met, and that a causal account of temporal priority is correct. But that is an issue we cannot pursue here.

Suppose, however, that it turned out that no satisfactory analysis of the concept of temporal priority, or of the concept of the past, could be found. Then it would seem that one would need to re-examine the view that those concepts cannot be semantically basic, and there are at least two lines of thought that one might appeal to in order to make plausible the idea that the concept of the relation of temporal priority can be taken as semantically basic:

(1) First, one might appeal to the way in which one naturally describes some of one's experiences. For example, isn't it natural to agree with Bergson's claim that one is immediately aware of motion? If so, and motion is somehow given in a single experience, then one must be perceiving an object in different positions within a single experience. But since an object can only be in different positions at different times, this entails that a single experience can involve the perception of two events, one of which is earlier than the other.

(2) Secondly, one might appeal to physiological findings. Thus, it is apparently true that there are parts of the brain that change their state only in response to moving stimuli. If there are experiences that are either identical with, or correlated with, such brain states, shouldn't one say that the experiences which a change in this part of the brain gives rise to are experiences of motion? So once again, one would have reason for holding that one can have an experience of one event's being earlier than another.

2. Skepticism and Memory Knowledge

The general strategy underlying the skeptic's challenge to knowledge claims is familiar. Applied to the case of beliefs about the past, it runs as follows:

(1) Suppose it is granted that one knows, or is justified in believing, that one now has some beliefs about the past, or that one is having some thoughts about the past, or enjoying some memory images. (The skeptic may not view the first of these as by any means unproblematic, since it, unlike the second and the third, is not a matter of knowledge of a present state of consciousness.) The question then is how such facts about present states of oneself can possibly serve to justify claims about past states of affairs. How is one ever to bridge the gulf in logical type that exists between statements about the present and statements about the past?

(2) It is clear that no deductive reasoning will ever serve to bridge the gap. For one could be in exactly the same state that one is now in even if the world had only existed for five minutes, or even if it had just now popped into existence. (Compare the famous hypothesis that the world was created in 4004 B.C., but with fossils, etc., that make it look as if it has existed for a much longer time.)

(3) No inductive reasoning can bridge the gap. For in order to do so, one would have to establish a generalization, connecting events at different times, that one could then apply to one's present states in order to draw a conclusion about earlier states of affairs related to them. But in order to establish such a generalization, one would have to have information about states of affairs existing at different times, and one doesn't have such information until one has some memory knowledge. Before that, one's evidence consists entirely of information about one's own present state.

(4) The only legitimate methods of reasoning are deductive and inductive.

(5) Hence one has no knowledge of - nor any justified beliefs about - the past.

Most of us believe that we are justified not only in thinking that there was a past, but also in having a large number of very detailed beliefs about it. But how is one to answer the skeptic?

Many philosophers have despaired of finding a defensible answer. Don Locke, for example, in his book Memory, seems to conclude that all that one can say is that "one cannot question the possibility of memory-knowledge without shaking the entire structure of human knowledge to its foundations." But, as he immediately goes on to point out, "that, of course, is precisely what the philosophical skeptic means to do." (Page 137)

I want to argue, however, that there is at least one satisfactory answer to the skeptic. So let us now turn to an examination of possible answers to the skeptical challenge.

3. Possible Answers to Skepticism about Memory Knowledge

One common way of responding to skeptical challenges to knowledge claims is by advancing some appropriate reductionist claim. Thus, for example, phenomenalism, in viewing physical objects as constructions out of possible sense experiences, provides an account of how one can get from knowledge of sense experiences to knowledge of physical objects. Similarly, logical behaviorism, in holding that mental states are reducible to observable behavior, and behavioral dispositions, provides a possible solution to the difficult problem of knowledge of other minds.

Is a reductionist approach plausible in the case of beliefs about the past? Here such an approach would involve holding that propositions bout the past can be analyzed in terms of propositions about the future. This view has certainly been embraced by some philosophers. In particular, it was accepted by the Polish logician Jan Lukasiewicz and the New Zealand philosopher of time Arthur Prior. But the view that statements about the past are analyzable in terms of statements about the present does not really seem at all implausible in itself. And, in addition, it also has some rather unusual consequences. One of these, which was noted by Lukasiewicz - though he viewed it as a welcome consequence - is that the past is not fixed, at least in an indeterministic world, since in such a world there may be evidence, at one time, for the occurrence of some earlier event, but then no evidence at all at some later time:

Facts whose effects have disappeared altogether, and which even an omniscient mind could not infer from those now occurring, belong to the realm of possibility. One cannot say about them that they took place, but only that they were possible. It is well that it should be so. There are hard moments of suffering and still harder ones of guilt in everyone's life. We should be glad to be able to erase them not only from our memory but also from existence. We may believe that when all the effects of those fateful moments are exhausted, even should that happen only after our death, then their causes too will be effaced from the world of actuality and pass into the realm of possibility. Time calms our cares and brings us forgiveness.

One might be able to overcome the implausibility of this reductionist thesis if one could make out a strong case for a verifiability theory of meaning. For familiar reasons, which I shall not consider here, the prospects of doing that do not seem very promising. But, in addition, not just any verifiability theory would suffice: it would have to be one in which verifiability was verifiability now, and such a theory is even more implausible than verifiability theories in general.

If reductionism is set side, one is left with at least four ways of responding to the skeptic's challenge to our everyday belief that memory often provides us with knowledge, or at least justified beliefs, about the past, which deserve serious consideration:

Traditionally, philosophers have almost always held that if is true that one's memories are reliable, this is only a contingent truth, rather than a necessary one. This view can. moreover, be supported as follows. First, it is surely true in any particular case that there is nothing necessary about a given memory's being correct; it is surely conceivable that any apparent memory could be mistaken. But then, secondly, it is natural to suppose that if it is only a contingent matter whether any particular memory belief is correct, the same must be true with regard to any pair of memory beliefs: the falsity of one memory belief is surely compatible with the falsity of the other memory belief. But, then, if this is right, adding more memory beliefs will not, it would seem change things: the fact that all the memory beliefs in some set of n memory beliefs are false will surely not entail that some other memory belief is true. Consequently, while it might be true that memory beliefs are generally, or even always, correct, it is surely also possible that memory beliefs might be generally, or even always, incorrect. It is, therefore, simply an empirical or factual question whether memory beliefs are generally correct or generally incorrect.

This view can also be supported as follows. Couldn't one in fact construct "perceivers" most of whose memory beliefs would be mistaken? For one could construct a machine such that when it perceived a red object, that would at most sometimes (and possibly even never) result in a memory being stored of its having perceived something red. There could, for example, be a randomizing device in the machine's "brain" that generally led to the storage of a "memory" of having seen an object that was some color other than red. But if mechanical perceivers with such a characteristic could be constructed, must it not be a purely contingent fact if humans are such that most of their memories are correct?

Some philosophers have, nevertheless, challenged this view - most notably, Norman Malcolm and Sydney Shoemaker. They have contended that while it is a contingent matter whether any particular memory belief is correct, it is not a contingent matter whether memory beliefs are generally correct. On the contrary, they claim, it is a logically necessary truth that memory beliefs are generally reliable.

If this claim were true, would memory knowledge be inferential or noninferential? One might be tempted to say that it was inferential, on the grounds that one's justification for accepting any given memory belief was that one is justified in believing that memory beliefs are generally reliable. But the notion of an inferentially justified belief, in the case of contingent beliefs, should really be understood, I think, as the idea of a belief that is justified on the basis of other contingent beliefs. So if it turned out that it is logically necessary that memories are generally correct, I think that one should view the relevant memory beliefs as noninferentially justified. For if it is logically necessary that memories are generally correct, then, even when unsupported by any evidence, memory beliefs are, by their very nature, prima facie credible. Thus such beliefs would be noninferentially justified. The Malcolm/Shoemaker position thus represents one way of attempting to support a direct realist view of the justification of memory beliefs.

4.1 Shoemaker's Formulation of the Argument

The claim that it is logically necessary that memories beliefs are generally reliable is, in view of the considerations mentioned above, a surprising one. But an argument in support of it can be offered, along the lines of one set out by Shoemaker in his article on "Memory" in the Encyclopedia of Philosophy:

[The view that it is logically possible that memory beliefs are universally or generally false] rests on the idea that it can only be contingently true, if it is true at all, that memory beliefs are for the most part true. If this were so, we ought to be able to imagine a people whose memories were seldom or never correct. But supposing that there could be such a people, how could we identify any of their utterances as memory claims (as we would have to be able to do in order to find that their memory claims are mostly false)? We would not be satisfied that one of our own children had learned the correct use of the word "remember" and of the expressions that indicate past tense unless the sincere statements he made by the use of these expressions were normally true - just as we would not allow that someone knew the meaning of the word "blue" if he typically applied it to such things as grass and trees. In this case, as in many others, using an expression correctly necessarily goes together with using it to make statements that are (for the most part) true. It is only because this is true of many expressions that it is possible to decipher a strange language by seeing in what circumstances the expressions in that language are typically uttered. If the language of a people were translated in a certain way and it turned out that the utterances translated as memory claims nearly always had to be regarded as false, this would surely be conclusive grounds for saying that these utterances were not memory claims at all and that the language had been mistranslated. (Encyclopedia of Philosophy, edited by Paul Edwards, Volume 5, page 273)

4.2 An Evaluation of the A Priori Argument

The above line of argument is an interesting one, but I believe that it is unsound, primarily for the following reasons:

(1) In translating a strange language into one's own, the demand for agreement in judgments varies radically, depending upon the content of the sentences that one is translating. Thus, for example, if one translates certain foreign expressions as color words in such a way that the color judgments of speakers of the other language are almost always in disagreement with one's own color judgments, then that is very strong evidence that one's translation is wrong. Suppose, in contrast, that one is translating the religious utterances of someone else into one's own language, and one winds up attributing to the other person magical beliefs, or theological beliefs, that one regards as false, or even absurd. One might, for example, find that if one translated a certain term in their language by the word "witch", it turned out that people in that society believed that there were in fact many women who had entered into contracts with the devil. Would that be strong evidence that the translation was mistaken? Surely not.

What, then, is the difference between the two cases? The relevant difference would seem to be that in the first case one is dealing with language of a low-level, observational sort, whereas in the second case one is dealing with sentences that are heavily theory-laden. The more observational certain sentences are, the more disagreement about the truth-values of those sentences counts against the translation in question. The more theory-laden they are, the easier it is to see how people might have a large number of false beliefs.

(2) Secondly, suppose that when the behavior of the individuals in question turns out to be unsuccessful - as it generally will if their memory beliefs are usually or always false - they utter words that we translate as, "That's strange, I was sure that I buried it here. My memory must have been mistaken." If they afterwards acknowledge in this way that their memory beliefs were ones that now appear to have been mistaken, does the fact that their memory beliefs are generally wrong still count against the translations we are offering of their utterances? It would seem to me that it does not. (This second consideration cannot be used if one is defending, not the view that memories could be generally wrong, but the stronger view that they could be always wrong, since to engage in the linguistic behavior in question, the individuals have to remember what they believed if they are to be able to say, in anything other than a random fashion, that the memory belief in question was wrong.)

(3) For at least a wide range of beliefs, the primary evidence - in a logical sense, not in a practical sense - for what a person believes is provided by his or her nonlinguistic behavior. Thus one may say that a dog believes that a bone is buried in a certain place because it is digging there, and has otherwise been behaving in ways appropriate to bone-seeking. So one may find individuals acting upon the memory beliefs that one is attributing to them, and such action will give one grounds for holding that the related utterances are being correctly translated, despite the fact that the resulting sentences are false.

Suppose, for example, that a person places a red object under an opaque cover on the right, and a green object under an opaque cover on the left. Asked where the red object and the green object were placed, they say something that we translate as the claim that they remember placing the red object under the cover on the left, and the green object under the cover on the right. Then, when asked to retrieve the red object, they do life up the cover on the left. Such behavior surely shows that the person believed that the red object was under the cover on the left, and this in turn supports the hypothesis that our translation of what they said as the claim that they remembered placing the red object under the cover on the left was a correct translation.

(4) Finally, suppose that when asked if they can define the term that we are translating "remember", they utter a sentence that we translate as follows: "To say that A remembers that p is to say that A believes that p, that A's present belief that p is caused by an earlier belief that p, where the earlier belief, in turn, was caused by a perception that p." Suppose, further, that when asked if they can define the term that we are translating "past", they utter a sentence that we translate as follows: "The past is what is earlier than the present." Suppose, finally, that when asked if they can define the term that we are translating "earlier than", they utter a sentence that we translate as follows: "The earlier than relation is a temporal relation that is transitive, asymmetric, and irreflexive, and that holds between the present and the future." Given such utterances, wouldn't we be justified in concluding that we had correctly translated certain sentences as sentences making memory claims about past events?

5. An Appeal to the Specious Present

A second way in which one might attempt to show that memory beliefs can be justified involves an appeal to the specious present. This approach was advanced by R. F. Harrod, in a book entitled Foundations of Inductive Logic, and as Harrod formulates it, it makes use of hypothetico-deductive method. But it may be helpful to start off by considering a simpler, and less satisfactory argument that also appeals to the idea of the specious present, and that involves only instantial generalization.

5.1 An Argument that Uses Instantial Generalization

Let us begin by considering, then, the simpler argument. Given the skeptical argument set out earlier, the idea that one can inductively establish generalizations to justify beliefs about the past may well seem, of course, quite out of the question. But, as we shall now see, there is a possible way of challenging that conclusion.

The basic line of thought is as follows:

(1) First, one's experience is not confined to that of an instantaneous state of affairs. For if it were, there would be no perception of motion. One's experience should be thought of, accordingly, as involving a specious present.

What is meant by the specious present? A way of understanding this is by comparing pictures of, say, a fast-moving racing car taken, on the one hand, by a camera with a slow shutter speed with those taken, on the other hand, by a camera with a very fast shutter speed. Pictures of the latter sort may be indistinguishable from pictures taken of a car that is at rest. One will be unable to tell whether the car is at rest, or moving forward, or moving backward. By contrast, if a picture is taken of a fast-moving racing car by a camera with a slow shutter speed, one will be able to tell from the picture, with the sharp image of one end of the car, and blurred image of the other end, that the car is moving, and the direction in which it is moving.

The situation is similar when we look at an object moving sufficiently fast. The image we form is not uniform. Instead, there is a residual image that represents slightly earlier positions of the moving object that we are watching.

(2) Secondly, if one's experience takes the form of a specious present, then one is experiencing, within a single experience, events that take place at slightly different times. One sees, for example, both where a moving object now is, and where it was at a slightly earlier time.

(3) Thirdly, given that a single experience does contain a representation of events that have taken place at slightly different times, such experience can serve to confirm generalizations relating those events that exist at different times.

(4) What generalizations can be confirmed in this way? Here are some possibilities:

(a) Things that are at rest tend to stay at rest;

(b) Things that are moving tend to continue moving;

(c) Things generally do not change their colors quickly;

(d) things generally do not change their shapes quickly.

(4) Then, fourthly, the generalizations thus confirmed can be used to arrive at conclusions concerning earlier moments. So, for example, if one is presently having an experience of a round, green object moving quickly to he right, one can conclude that that object was round and green at a slightly earlier time, and was moving quickly to the right, but located some distance further to the left than it presently is.

How much knowledge of past events can be justified in this way? The answer is that only a very small proportion of one's knowledge claims about the past can be justified via this route, since memories play no role at all in type of argument just set out: one is simply projecting backward from that part of one's sensory experience, or that part of the external world that one is perceptually aware of, making use of those generalizations that can be confirmed by those perceptual experiences that lie within the specious present in question.

What one needs, clearly, if one is to be able to establish that one's memories are generally reliable, is an argument that makes use both of one's memory beliefs and/or experiences, and of the idea that our sensory experience occurs in a specious present. To arrive at such a conclusion by means of instantial generalization, one would need to be able to recognize that some memories were correct simply by features of the experience that one was having at the time in question, and this would only be possible if the memories were restricted to ones about earlier phases of the experience in question. Perhaps some such argument might work, though it would clearly be more appealing, I think, if one could construct an argument in terms of memories that were not thus restricted. To do that, however, one will need, it would seem, to make use, not of instantial generalization, but of hypothetico-deductive method.

5.2 R. F. Harrod's Justification of Memory Knowledge

An argument of the latter type - that does make use of hypothetico-deductive method - was advanced by R. F. Harrod in chapter 8 of his book, Foundations of Inductive Logic. Here is a longish quote from Harrod's discussion:

I call attention to the point that all our confidence in the informativeness of memory must be derived from the experience of one specious present. To adduce in favor of memory, prior to any probability of its informativeness, evidence collected from specious presents that have already disappeared would be to argue in a circle, for it would imply that the memory of those specious presents that have already disappeared was informative.

You might at first think it hard to collect from one fleeting specious present strong evidence in favor of so momentous a generalization as the veridicity of memory. But this is not really so. Suppose that an observer was sitting quietly in a small room. This room, as it figures in his sensible experience, may be regarded as a collection of some ten thousand small patches. Let him predict that all these patches will remain quite stationary in the specious present except for four, located on what he calls the ceiling; for those he predicts motion within the specious present. There are some four hundred billion alternative ways of selecting four patches out of a collection of ten thousand. It was open to him to select any one out of those four hundred billion sets and, if memory were illusory, he would have no evidence in favor of one rather than another. Let us suppose his predictions are correct, the four patches selected being flies, which had been in incessant motion before the backward end of the specious present. (Page 193)

By taking a few such simple matters I believe that we could get supra-astronomical odds against the truth of one combination of predictions, chosen out of all possible combinations of predictions within the range of the imaginable, if there were nothing in favor of one combination rather than another. (Page 194)

. . . It follows that if memory is uninformative there are sometimes astronomical odds - even supra-astronomical odds - against having the success in prediction that one often does have in a single specious present. If, on the other hand, the memories are indeed informative the successful predictions are quite probable. (Pages 194-5)

In the first stage of the argument we can only establish probability in favor of the informativeness of the memories used. We can say that there is a high probability in favor of the informativeness of most of the memories used for prediction in the specious present, and these can be taken as a sample of the wider class of all memories used and unused.

But as soon as we can say that there is a high probability for most members of the narrow class of memories, viz. those used in the specious present for prediction, we are released from the case of the specious present. For if those memories are veridical we have access, on the basis of high probability, to some past experiences, which may contain other examples of successful prediction. Thus the sample of memories very likely to be (or to have been) veridical is widened. We then proceed by ordinary sampling to establish high probability for a sizable proportion of memories generally. (Page 195)

Harrod's argument can, I think, be summarized as follows:

(1) A person is aware that a certain state of affairs, of type S, exists at the beginning of a given specious present, P. (In Harrod's illustration, the state of affairs of type S involves 10,000 small color patches, including four particular black spots.)

(2) The person in question has a memory of a slightly earlier state of affairs of type S that had property T. (In Harrod's illustration, T is the property of its having been the case that the four particular black spots were in motion, whereas all of the other 9,996 color patches were at rest throughout the immediately preceding interval in question.)

(3) The person in question also has other memories of a set of generalizations, G, that apply to states of affairs of type S with property T, and which predict that a state of affairs of type S that has property T will be followed by other states of affairs of type S with property T. (In Harrod's illustration, G will involve two generalizations, one to the effect that moving objects tend to remain in motion, and the other to the effect that objects at rest tend to remain at rest.)

(4) Consider, now, the hypothesis, H, that the memories just mentioned are reliable. Hypothesis H, together with (1), (2), and (3), generates the prediction that the state of affairs of type S that the person is aware of within the specious present in question will also have property T. (So, in the case of Harrod's example, the prediction is that the four particular black spots will be in motion, whereas all of the other 9,996 color patches will be relatively motionless.

(5) The probability that the state of affairs of type S in question will have property T is very low if hypothesis H - the hypothesis that the memories in question are reliable - is not true. (Here Harrod offers a quantitative assessment: in predicting that the four particular black patches will move, while the other 9,996 patches will remain at rest, one is choosing one possibility out of about 400 billion alternatives.)

(6) First Conclusion: The hypothesis that the memories used in the prediction are reliable is justified, since otherwise it would be an incredible accident that the prediction that the person made on the basis of his or her memories has turned out to be true.

(7) There are a number of comparable predictions that one can make with regard to a single, specious present, and, accordingly, one can argue that, if the relevant memories are not reliable, the probability against any such combination of predictions must be "supra-astronomical".

(8) If the memories in question are reliable, then those memories provide one access to information about other specious presents in addition to the one that one started from, and in which successful predictions may also have been made.

(9) The resulting justified beliefs about different specious presents in which successful predictions have been made justify one, in turn, in concluding that it is likely that one's memories in general are reliable.

(10) Final Conclusion: One is justified in concluding that one's memories in general are reliable, and thus that one's beliefs about the past that are based upon one's memories are justified

5.2 Possible Objections to Harrod's Approach to the Justification of Memory Beliefs

(1) Does one want one's justification of memory beliefs to presuppose a certain thesis about the nature of sensory experience - namely, that it involves a specious present? One might think that this is harmless, on the ground that the existence of a specious present is an undeniable phenomenological fact. But I want to suggest that there are two things here that are problematic. There first, and the less serious, is this. Even if it is a fact about one's experience that it involves a specious present, would it be acceptable if it were the case that if one's experience were instead instantaneous, one would not be able to justify any memory claims? That is to say, if one's experiences were not of the specious present variety, but one still had essentially the same memories - except for the fact that those memories were instead of instantaneous experiences - would it be plausible that one could not then show that one had any justified beliefs about the past?

If that is not plausible, the conclusion is at least that there must be some other way of justifying memory beliefs. This, of course, does not show that there is anything wrong with the method proposed by Harrod.

(2) The second, and much more serious problem arises when one asks what account is to be given of the specious present. Suppose, for example, that the correct description of one's sensory experience were as follows. First, the purely sensory state that one is in at any given time is similar in qualitative nature to what one has when one looks at a picture taken of a moving object by a camera with a slow shutter speed. That is to say, there is a sharply focused representation of one end of the moving object, but the image of the other end of the moving object is connected to a faint tail representing, so to speak, earlier locations of the moving object. One can therefore tell, from the image, that the object is moving, and the direction in which it is moving. Secondly, one's 'experience' in the broad sense, at any given time, involves both the sort of image just described, along with memories of qualitatively similar images in which the moving object is in the locations it was in at very slightly earlier times.

If this is the correct account of the mental state that one is in when one is watching a moving object, then the idea that one can, simply via awareness of the content of that experience, know directly, for example, that something is moving, seems problematic. For on the one hand, although there are the memories of the object's having been in different positions at slightly earlier time, one cannot treat those memories as accurate in the context of an argument whose goal is to show that memories are in general reliable. On the other hand, there is the sensory image itself, but while that can be a ground for concluding that one is watching a moving object, that conclusion presupposes, it would seem, prior information about how a moving object looks: the detailed characteristics of the image only enable one to conclude that the object is moving because one knows that certain parts of the image are traces of the object's having been in other locations than the one that it is presently in.

In short, there are two ways in which one can move from the nature of one's total 'experience' at a given time to the conclusion that an object is moving. The one involves reasoning from a trace. Any such inference always presupposes, however, the prior confirmation of a relevant generalization, which in turns requires memory. The other way of arriving at the conclusion that something is moving, given the 'experience' in question, makes uses of the memories that are part of the 'experience' in the broad sense. Either way, then, one can only get to the conclusion that something is moving by appealing, directly or indirectly, to memories. But if this is right, then Harrod's argument is implicitly circular.

(3) A final point about Harrod's account is that, even if it were tenable, it could not be a complete account. The reason is that Harrod's account cannot provide any basis for one's having any justified beliefs about the past at times when one is not having any sensory experiences of the requisite sort. Suppose, for example, that you have your eyes closed in a quiet setting, and that all you are experiencing are a few tactile sensations. Then there would not seem to be the possibility of the dramatic predictions that Harrod was able to appeal to in the case of visual experiences. Or, more radically, suppose that you were in a state of complete sensory deprivation. Wouldn't you still have, in either case, virtually all of the justified beliefs about the past (with the exception of beliefs about the recent past states of one's immediate physical environment) that you now have?

6. Direct Realism

Let us now turn to two accounts of the justification of memory beliefs that correspond to two very important strategies that can be employed in responding to the general skeptical argument set out earlier. The one, which will be considered in the next section, involves the attempt to show that one can make use of hypothetico-deductive reasoning, or inference to the best explanation, to establish that memory beliefs are inferentially justified. The other, which we shall begin to consider in this section, involves what is often referred to as a direct realist approach to memory, and here the central claim is, as a first approximation, that memory beliefs are noninferentially justified.

Let us consider, then, a direct realist view of memory. One way of formulating this view is the one just mentioned - that is, as the view that memory beliefs are noninferentially justified. There is, as we shall see shortly, a different, and somewhat more modest way of formulating the direct realist view. But let us begin with the version according to which memory beliefs are noninferentially justified.

When it is claimed that some belief is noninferentially justified, one always needs to go on to ask what the basis is - that is, what state of affairs is the ground of one's being thus noninferentially justified in accepting the belief in question. In the case of memory beliefs, there are at least three different answers that might be given:

(1) The basis of A's being noninferentially justified in accepting a certain belief, p, about the past is that A has a corresponding memory image;

(2) The basis of A's being noninferentially justified in accepting a certain belief, p, about the past is that A has the belief that p;

(3) The basis of A's being noninferentially justified in accepting a certain belief, p, about the past is that A has the thought that p.

The first two versions of the direct realist theory of memory knowledge, parallel the two different versions of direct realism with respect to perceptual knowledge advanced by Sellars and Armstrong, and discussed in an earlier chapter.. Thus, just as Sellars accepts the existence of qualitative, non-intentional, perceptual experiences in setting out an account of perception, so the first version of direct realism involves reference to memory images. But this approach, while holding that there are memory images, maintains that one's beliefs about the past are not inferred from beliefs about those memory images - just as Sellars maintains that one's perceptual beliefs about the external world are not inferred from beliefs about one's sensory experiences. In addition, it may be that the memory images may cause one to have certain beliefs about the past, so that one has - in Sellars' terminology - causal mediation, just as Sellars holds that one's sense experience may cause one's perceptual beliefs. But there is no epistemic mediation: one's beliefs about the past are noninferential, and noninferentially justified.

The second version of direct realism with respect to memory knowledge does not assign any role at all to memory images. The reason may be in part to avoid any temptation to think that one's beliefs about the past are inferentially justified on the basis of beliefs about memory images. But I suspect that the main reason involves a point made earlier - the realization, namely, that if one restricts memory knowledge to cases in which one is having memory experiences, it will turn out that one has very little memory knowledge: if one is not to be driven into a close approximation of skepticism, one must maintain that memory beliefs, in the absence of memory experiences, can provide the basis for justified beliefs about the past.

The third version of direct realism treats the basis states as neither images nor beliefs, but thoughts - where a thought, here, is a certain type of state of consciousness. The appeal of the view that it is the thought that p that is the basis of one's being noninferentially justified in accepting a certain belief, p, about the past is, I think, twofold. In the first place, the having of images does not seem necessary for one to have memories. In the second place, it is tempting to think that, since thoughts are states of consciousness, one can be directly acquainted with thoughts, and there is, I think, considerable appeal in the idea that the basis states that ground any noninferentially justified beliefs should be states with which one can be directly acquainted. So thoughts are more appealing candidates for the role of basis states than beliefs are.

But I think there is a very strong objection to taking thoughts to be the basis states - namely that, just as in the case where one takes the having of memory images to be the basis states, one is confronted with a very unwelcome skeptical consequence, to the effect that when one is not having any thoughts about the past, one cannot have any justified beliefs about the past, and this will mean, for example, that one is often acting on the basis of beliefs that are not, at the time of the action, justified.

It seems reasonable to conclude, then, that if direct realism is to be sustained, it will have to be the second version.

So far I have been treating direct realism as the view that memory beliefs are noninferentially justified. It is possible, however, to formulate direct realism in a slightly different way, which involves a more modest claim. Consider, for example, the following passage from C. I. Lewis's An Analysis of Knowledge and Valuation (LaSalle Illinois, Open Court Publishing, 1946):

". . . it is only essential that the fact of present memory afford some presumption of the fact which is memorially presented. All that is needed is initial assumption that the mere fact of present rememberings renders what is thus memorially present in some degree credible. For the rest, the congruence of such items with one another and with present sense experience will be capable of establishing an eventual high credibility, often approximating to certainty, for those items which stand together in extensive relations of such congruence." (See Meaning and Knowledge, edited by Ernest Nagel and Richard Brandt, (New York: Harcourt, Brace & World, 1966), page 514.)

The expression "in some degree credible" is not entirely clear. Is Lewis saying that the probability that the belief in question is true must be at least greater than one half, so that the belief is more likely to be true than to be false? If so, the belief would be noninferentially justified, and we would not have a distinct version of direct realism. The answer, however, is that this is not what Lewis means here by the expression "in some degree credible". All that he requires is that the probability that the proposition about the past is true must be greater, given that one accepts it, than it would be if one did not accept it. In short, the basis state must raise the probability of the proposition in question above its a priori probability, but it need not make it more likely to be true than to be false.

Lewis's idea is then that if one has a number of memory beliefs that are, in that sense, in some degree credible, then if those beliefs stand in certain relations of mutual support - which he refers to as relations of congruence - then the probabilities of all the beliefs will be increased, with the result that the probability that any given belief may be not merely greater than one half, but possibly very close to one.

This line of thought generates three additional versions of direct realism, corresponding to the three earlier versions:

(4) If p is some proposition about the past, then the basis of p's having a probability for person A that is higher than its a priori probability is that A has a corresponding memory image;

(5) If p is some proposition about the past, then the basis of p's having a probability for person A that is higher than its a priori probability is that A has the belief that p;

(6) If p is some proposition about the past, then the basis of p's having a probability for person A that is higher than its a priori probability is that A has the thought that p.

Which of these last three versions of direct realism does C. I. Lewis accept? The following passage provides, I believe, at least a partial answer:

"The indispensable item is some direct empirical datum; the actually given reports, the facts of our seeming to remember; and without that touchstone of presentation, relations of congruence would not advance us a step toward determination of the empirically actual or the validly credible. However important this relation of congruence in the building up of the structure of empirical beliefs, the foundation stones which must support the whole edifice are still those items of truth which are disclosed in given experience." (Meaning and Knowledge, pages 513-14.)

Expressions such as "direct empirical datum", "the facts of our seeming to remember", "touchstone of presentation", and "disclosed in given experience" seem to suggest that the basis state is some state of consciousness, which suggests either position (4) or position (6), rather than position (5). Moreover, the expression "the facts of our seeming to remember" tends to suggest, I think, that it is thoughts, rather than images, which are, for Lewis, the relevant basis states.

But both version (4) and version (6) of direct realism are open to the same objection that was mentioned in connection with versions (1) and (3) of direct realism - namely, that it follows that at times when one is not having any memory experiences, one has no justified beliefs about the past. The plausible option as regards positions (4) through (6), then, is position (5).

One final point needs to be made about he formulation of direct realism, and it concerns the logical form of the basis rules that specify the conditions under which a given belief about the past is noninferentially justified. In the case of some noninferentially justified beliefs of other types, it may well be that the relevant basis rules specify conditions that are sufficient to make it the case that a given belief is noninferentially justified. Consider, for example, beliefs about one's own present experience. It seems very plausible that such beliefs are noninferentially justified, and in the case of such beliefs, one might very well think that the following sort of basis rule is correct:

In view, then, of the fact that one's memory claims about the past can be mistaken, and that, in addition, one can be presented with evidence that provides one with good reasons for concluding that some of one's memories are mistaken, the noninferentially justified beliefs in question will have to rest upon basis rules of a different logical form, such as the following:

7. Indirect Realism: A Hypothetico-Deductive Account of the Justification of Beliefs about the Past

7.0 The Indirect Realist Account

Let us consider, then, indirect realism, where this is the view that memory beliefs are inferentially justified.

If one is to defend the claim that memory beliefs are inferentially justified, four things need to be done:

(1) First, one needs to specify the type of beliefs that it is claimed can provide an evidential foundation for memory beliefs;

(2) Secondly, one needs to show that one can be justified in accepting beliefs of the type in question even in the absence of any justified memory beliefs, so that the path of justification that one is proposing is not circular;

(3) Thirdly, one needs to specify the type of inference involved;

(4) Finally, one needs to show that that inference is sound.

As regards the first of these tasks, the idea is that the proposal is that one can justify beliefs about the past on the basis of justified beliefs to the effect that one does have beliefs about the past that exhibit certain characteristics. So the idea is that certain first-order beliefs - beliefs about the past - can be justified on the basis of certain second-order beliefs - beliefs about beliefs about the past.

Next, let us skip, for the moment, the second task listed above, and turn to the third task, that of specifying the type of inference involved. Here the basic idea is that the inference is an inference to the best explanation. Or, in more detail, the claim is that the theory that provides the best account of an individual's present sense experiences and memory beliefs is one involving the following hypotheses:

(1) There are more things that exist than merely an individual's present experiences and (ostensible) memory beliefs. There are also past experiences and beliefs, and enduring external objects.

(2) There are causal laws that relate states of physical objects at different times.

(3) Physical objects can act upon organisms to produce both sense experiences and beliefs about those sense experiences.

(4) Some of the beliefs that an organism acquires about his experiences can be retained by the organism for at least short intervals of time.

The fourth task is to show that the inference to this hypothesis is justified, What is immediately clear is that given a theory T with these features, one can, together with appropriate assumptions about initial conditions, explain the fact that a given individual has certain sense experiences and certain memory beliefs at a given time. So theory T is an explanation. But the critical question is this:

Does theory T provide the best explanation?

If the claim that theory T provides the best explanation is to be sustained, one must show that it is superior to alternative explanations. In the present case, it seems that there are two main competing theories that one must show are less satisfactory:

(1) Theories such as "Russell's hypothesis" that the world has only existed for five minutes;

(2) The more dramatic theory that nothing at all existed up until the present moment.

7.1 Against Russellian-type Theories

Russellian-type theories involve three of the hypotheses that are present in theory T - namely, (1), (3), and (4). The difference is that hypothesis (2), in contrast, is not accepted in its unrestricted form. For while, on a Russellian-type hypothesis, there are laws relating present and past physical events to earlier physical events, those laws are restricted in a certain respect. In particular, if the world began five minutes ago, then all physical events have at least partial, prior physical causes except for those physical events that occurred exactly five minutes ago.

But the problem with any hypothesis of this sort is that the five minute period during which the physical world has existed provides evidence for the unrestricted generalization that all physical events have prior physical causes. One would be justified in restricting that generalization to events that have occurred within the last five minutes only if one had evidence that the world had only existed for five minutes. But on Russell's hypothesis, one has no evidence that the world began to exist five minutes ago. Consequently, to restrict the generalization that physical events have prior physical causes to events occurring less than five minutes ago is arbitrary and unwarranted, and it results in a more complicated theory. The simplest assumption is that there is a law that physical events always have prior physical causes. Hence our ordinary theory of the world is preferable to Russellian-type theories.

The basic point is perhaps clearest if one considers a Russellian-style hypothesis according to which the age of the physical universe is considerably greater than five minutes. Consider, for example, the hypothesis that the physical universe popped into existence about 10 years before sentient life came into being, and popped into existence in just the state that we normally believe it was in before the appearance of sentient life. On that scenario, there will be an enormous number of events that confirm the unrestricted laws in question. But shortening this period down, as Russell did, to one of five minutes in length, does not fundamentally change things: there will still be an enormous number of events that confirm the unrestricted causal hypothesis, and this disconfirm the Russellian hypothesis

There is also a second way in which Russell's hypothesis is a less satisfactory one, and it is one that will emerge very clearly when we consider, in the next subsection, the second competing theory - namely, the hypothesis that the world has just now begun. For what will become clear there is that any theory that postulates a beginning of the universe at some time when there are going to be people with memories cannot explain the correlations that exist within the memories of people at that time. Russell's hypothesis that the world began five minutes ago cannot explain, then, the correlations that exist within the memories of people exactly five minutes ago.

7.2 Against the Theory that the World Has Just Now Begun

The first objection urged against Russell's hypothesis cannot be used here. If only the present moment exists, there are no laws that one can argue are being restricted in an arbitrary way. The central objection to the present theory is instead that it cannot provide any explanation of certain striking correspondences of two sorts:

(1) Between one's present sensory experiences, if one is having any, and some of one's present, apparent memories;

(2) Between different apparent memories that one is now having.

Thus, as regards the first, consider the following:

Time: t1 t2

The idea, in short, is that if the world had just now begun, these very striking correlations would be simply fantastic accidents that could not be explained in any way, whereas if the world has not just begun, then they can be explained - in particular, by theory T.

This sort of situation is surely very common, since one is almost always aware of whether one’s experience is changing or remaining the same, and therefore one must have memories of what one’s experience were like at very slightly earlier times.

The drawback, however, of this type of case is that this particular inference to the best explanation is not available if one is having no sensory experiences, or very limited ones, at a given time. As we shall now see, however, there are other inferences to the best explanation that are available when one is not having any experiences, since they involve only correlations between different beliefs about the past.

Such correlations can arise in two different ways. First, a single experience can give rise both to direct memories and indirect memories - where a direct memory is a memory of an experience, and an indirect memory is either a memory of a memory of an experience, or memory of a memory of a memory of an experience, or . . . and so on. Secondly, two distinct, but very similar experiences that occur near one another may give rise to very similar memories at a given time. Here, for example, is a picture of the first sort of situation:

Time: t1 t2 t3 t4

The idea here is this. If the time between t1 and t3 is not very great, and if one’s memory is at least reasonably reliable, then the two different memories, M3 of E1, and M3 of M2 of E1, which one has at time t3 should be quite similar in content. Accordingly, if one finds two memories that are quite similar in content, one is justified in inferring a common cause — which, in the above case, will, as a matter of fact, be experience E1.

In this case, one is comparing a direct memory of an experience with what is, as a matter of fact, an indirect memory of one and the same experience. The two memories do not by themselves enable one to determine, of course, that they are memories of one and the same experience, but that does not matter: all that is crucial for the inference is that the contents of the two memories are very similar.

One can also have cases where what is involved are two indirect memories that are, as a matter of fact, indirect memories of one and the same experience. Consider, for example, the two memories, M4 of M2 of E1, and M4 of M3 of M2 of E1, which exist at time t4.

Here, now, is a picture of the second possibility - that is, where two distinct experiences are involved:

Time: t1 t2 t3 t4

In this case, the idea is this. If the time between t1 and t3 is not very great, and if one is not in a situation that is changing rapidly, then one’s experience at time t2 may be very similar to one’s experience at time t2. Then, if one’s memory is at least reasonably reliable, those two very similar experiences may well give rise to corresponding direct memories at a slightly later time t3 — here M3 of E1, and M3 of E2 — that will be very similar in content. Accordingly, if one finds two memories that are quite similar in content, one is justified in inferring a common cause — which, in the present case, in contrast to the first, will not be a single experience. Instead, it will be the relatively unchanging perceptual environment that has given rise to the two very similar experiences.

The preceding case involves a direct memory and an indirect memory. But one can equally well have cases where what is involved are two indirect memory beliefs. Consider, for example, the two indirect memories M4 of M2 of E1, and M4 of M3 of E2, that exist at time t4.

How common are these cases that do not involve any current experience? It would seem that they are also very common, since it is surely often the case that one has an apparent memory belief about having an apparent memory belief about an experience of which one also has, as a matter of fact, a direct memory.

One often has, in short, pairs of apparent memories, either direct or indirect, that exhibit a high degree of similarity with regard to the type of experiences involved in the two cases. How is one to account, then, for the high degree of correspondence of content between, for example, a memory, M, of a certain experience, and a memory, N, of a memory of a very similar experience? The point, once again, is that the correspondence cannot be explained if one assumes that the world has just begun. But it can be explained if one assumes that there has been was a past. For one can then account for the correspondence by saying that the reason that there is the relationship of close resemblance between M and N is that when one traces back the different causal chains that resulted in M and the N one is led back to a single collection of physical objects - though slightly different temporal slices of that single collection of physical objects.

The upshot, in short, is that the structural correspondences in question have to be treated as a colossal accident on the hypothesis that there is no past, but can be explained if one adopts the hypothesis that there is a past that causally affects the present.

Notice, finally, that this basic point can be made without reference to memories. For consider, instead, a number of photographs taken of some object, from roughly the same position. The content of those photographs will be quite similar, and if the object in question is still around, and hasn't changed too much, the content of the photographs will be similar in structure to the object. If one were to suppose that the world has just now popped into existence, those similarities would have to be treated as sheer accidents. But if one assumes, instead, that there was a past, etc., one can explain those relationships by the hypothesis that the object existed in the past, was photographed a number of times, and that the object and the resulting photographs have continued to exist without undergoing significant change.

In the case of memory, one does not have only what correspond to photographs of the object, since one also has what corresponds to photographs of photographs of the object: memories of memories. But while this alters things slightly, it does not change the basic logic of the argument.

Finally, notice that the photographic case establishes a claim that I advanced near the beginning of this chapter - namely, the claim that it might well be possible to have some justified beliefs about the past even if one had no memory beliefs at all.

8. The Choice between Direct Realism and Indirect Realism

Let us now turn to a consideration of the relative merits of direct realism and indirect realism. The choice between these two views rests, I shall suggest, upon certain general considerations concerning the sorts of states of affairs that can be the object of noninferentially justified, or prima facie credible, beliefs.

8.1 The Scope of Noninferentially Justified, or Prima Facie Credible, Beliefs

According to direct realism, memory beliefs possess prima facie credibility. One might begin by asking, then, what support, if any, one can offer for this claim.

Some philosophers would, I think, be content to respond by arguing that unless one is willing to accept this view, one will inevitably be driven to skepticism, on the grounds that any attempt to show that knowledge of the past is inferential is doomed to failure. We have just seen, however, that that claim is not true.

But one also wants to ask whether, even if no inferential account of the justification of memory beliefs were available, one would be justified in simply assuming that memory beliefs possess prima facie credibility. After all, mightn't skepticism be the right position? Mightn't it be true that we just cannot have any justified beliefs about the past, even including the belief that there is a past? For may it not be that the reason why we believe, in everyday life, that we do have knowledge of the past is that we think that it must be possible somehow to show that memory is generally reliable? And if it then turns out that there is no way of doing that, shouldn't our conclusion be that we were mistaken in thinking that we had knowledge of the past? To conclude, instead, that we do have knowledge of the past, but that it must therefore be noninferential, rather than inferential, would seem ad hoc, and unjustified.

In addition, the skeptic can push his or her case further by arguing that one should not be allowed, in general, simply to postulate that certain beliefs are noninferentially justified, or prima facie credible, whenever one has trouble providing a justification for some knowledge claims. One should try, rather, to delimit the sorts of things that one can have noninferentially justified beliefs about. Moreover, this delimiting should not be done simply by offering a list. One should, instead, try to find some property that is possessed by all and only things of which one can have noninferential knowledge.

Having made this general point, the skeptic can then go on to mention some characteristics that have been suggested in "past" philosophical discussions:

How does indirect realism fare in the face of this sort of challenge? According to indirect realism, the justification of memory beliefs rest upon noninferentially justified beliefs to the effect that one has certain apparent memory beliefs. So the objects of one's noninferentially justified beliefs are - in contrast to the view embraced by direct realism - present states of oneself. But being a present state of oneself is surely not sufficient to make something a possible object of noninferentially justified, or prima facie credible, belief. If so, then indirect realism also faces a problem, since, first, beliefs about memory beliefs are not incorrigible; secondly, memory beliefs are not self-intimating, since one can have a memory belief, yet not be are of it; and thirdly, beliefs are either never immediately given in experience, or, if it is contended that they are so given when one thinks the relevant thought, it is certainly true that they are not in general immediately given in experience..

Notice, however, that the problem here is not a problem specifically for indirect realism, since regardless of what view one adopts with regard to the justification of memory beliefs, one surely wants to accept the following theses:

(1) One can have justified beliefs about one's own present beliefs;

(2) One can have justified beliefs, about one's own present beliefs, that are not justified on the basis of evidence.

But if this is right, then the question becomes what account one can give of the scope of noninferentially justified beliefs that will allow one to have noninferentially justified beliefs about one's own present beliefs.

The most appealing candidates for objects of noninferentially justified beliefs are objects that are immediately given. Experiences are such objects. But is there anything else which is? Initially, one might think that thoughts are also immediately given in experience; however I think that the fact that thoughts are characterized by intentionality provides grounds for holding that thoughts cannot be completely, given in immediate experience. And the situation is even worse in the case of beliefs, since while thoughts are states of consciousness, beliefs are not.

Let us consider, however, the case of thoughts. Can one be directly acquainted with the fact that one is presently thinking that p? To do so, it would seem that one must be able to tell two things. First, one must be able to tell that one is asserting that p is the case, rather than merely contemplating the proposition that p. Secondly, one must understand the content of the proposition in question.

How does one distinguish between merely entertaining a proposition, and asserting it? The answer may emerge if compares what happens when one attempts to mentally assert, on the one hand, that grass is green, and, on the other, that grass is red. Is the only phenomenal difference one that reflects the different propositions involved? When one merely contemplates the propositions that grass is green, and that grass is red, this may be so. But it seems to me that when one tries to assert the two propositions, there is a further difference. For isn't it the case that if one tries to assert that grass is red, the thought in question is immediately accompanied by a further thought to the effect that that proposition is not true?

If this is right, then what one believes does not reveal itself only when one acts. Beliefs, in addition to being maps by which we steer, involve a disposition to mentally assert the proposition in question whenever one attempts to mentally assert its denial.

What this suggests is that although beliefs are not given in immediate experience - indeed, one can have a belief at a time when there is nothing in one's state of consciousness at that time that is connected with the belief - one has potential, conscious access to one's beliefs, since any attempt to assert the denial of one of one's beliefs will be immediately be followed by a mental rejection of that denial.

My proposal, then, as regards the types of beliefs that can be noninferentially justified, or prima facie credible, is this:

8.2 The Advantages of Indirect Realism over Direct Realism

There are, I now want to suggest, four reasons for preferring indirect realism to direct realism:

(1) Indirect realism can offer a more plausible account of what beliefs can be noninferentially justified;

(2) Indirect realism requires fewer basis rules;

(3) Indirect realism can specify, in a non-arbitrary way, the level of confidence that is warranted in the case of beliefs about the past;

(4) The question of the relation between first-person beliefs and third-person beliefs poses a problem for direct realism, but not for indirect realism.

8.2.1 What Types of Beliefs Can Be Noninferentially Justified?

This is the issue that we have just been considering. The most appealing view, it would seem, is that the only beliefs that can be noninferentially justified are beliefs about what is given in immediate experience. But if this view is correct, then beliefs about one's own present beliefs cannot be noninferentially justified. The question therefore arises as to whether a less stringent principle - such as the Potential Access Principle - might not be defensible. If so, then it may be possible to show that beliefs about one's own present beliefs can be noninferentially justified, and this will provide the indirect realist with the necessary starting point for a defense of the view that beliefs about the past can be inferentially justified. The Potential Access Principle, on the other hand, does not enable the direct realist to show that beliefs about the past are noninferentially justified.

Moreover, it is not clear what plausible alternative the direct realist can put forward. One possible suggestion is this:

8.2.2 Indirect Realism Requires Fewer Basis Rules

This second point follows on immediately from the preceding point. For unless something like the Principle of the Prima Facie Credibility of All Beliefs can be sustained, the direct realist will need to accept basis rules that entail that beliefs about one's own present beliefs can be noninferentially justified, along with other basis rules that entail that at lest some beliefs about the past are noninferentially justified, or at least prima facie credible. The indirect realist, by contrast, needs only the former basis rules.

8.2.3 What Level of Confidence Is Justified for Beliefs about the Past?

The idea that one can be absolutely certain concerning propositions about the past does not seem at all plausible. On the contrary, C. I. Lewis's idea that beliefs about the past are, on their own, and initially, only prima facie credible, rather than being noninferentially justified, seems quite appealing. But if it is not appropriate to treat propositions about the past as certain, then what level of confidence is appropriate?

The direct realist cannot, I think, provide any principled way of answering this question. A level of confidence simply has to be arbitrarily selected. The indirect realist, by contrast, is in a very different position. For while it is true that the indirect realist faces the same problem with regard to beliefs about one's own present beliefs, this is a problem that the direct realist also faces, and if this shared problem can be resolved, then the indirect realist can, by using inductive logic, calculate the probability that one should assign to beliefs about the past.

This objection could be blunted, however, if one could argue that, regardless of the proposition in question, there is no principled way in which one can determine what level of belief is justified if the proposition is not one that is justified on the basis of evidence. But if this view is right, then it would seem to follow that there is no principled way in which one can determine what level of belief is justified in the case of any proposition, regardless of whether it is or is not one that is justified on the basis of evidence.  A kind of universal skepticism would seem, in short, to be the result.

8.2.4 First-Person Versus Third-Person Beliefs

Suppose that A has a noninferentially justified belief that event E took place in the past. What makes it the case that A is thus noninferentially justified? What is the ground, or the basis, of A's being thus justified?

It is hard to see what answer can be given, other than that the ground of A's being noninferentially justified in believing that event E took place is simply the fact that A does have a belief that event E took place.

Suppose this is right. Consider, now, another person, B, who has no beliefs about whether event E took place, but who comes to know that A believes that event E took place. Is B now inferentially justified in believing that event E took place?

It seems to me doubtful that B is justified, given only the evidence that A believes that event E took place. For while I think that this evidence certainly makes it more likely, other things being equal, that event E took place, it seems to me far from clear that it makes it more likely than not.

But now there seems to be a problem for the direct realist, since I think that the following principle is very plausible:

This objection could, of course, be avoided by adopting C. I. Lewis's view that beliefs about the past are, initially, merely somewhat more credible than they would otherwise be. But one will still be left with a residual objection. For now the question will be why it happens that the level of confidence that the relevant basis principle says is appropriate for A's belief that event E took place turns out to coincide with the probability that the proposition that event E took place has for B upon his evidence that A believes that event E took place. The direct realist, it seems, may have a problem supplying any answer to this question. By contrast, since the indirect realist holds that A's belief that event E took place is not noninferentially justified, the indirect realist is not confronted with this puzzle.

9. Summing Up

(1) Provided that one can be noninferentially justified in believing that one has memory beliefs, the use of hypothetico-deductive method can serve to justify the claim that specific memory beliefs are very likely to be accurate.

(2) One advantage of this approach is that what memory beliefs are justified at a given time does not depend either upon one's having any sensory experiences at a given time, or upon what memory experiences one is having at a given time.

(3) The skeptic can challenge this justification by arguing that noninferentially justified beliefs should be restricted to beliefs about one's present states of consciousness, and that one does not have noninferentially justified beliefs about any of one's present mental states that are not states of consciousness - such as beliefs. The indirect realist's account is, accordingly, incomplete, and needs to be supplemented by a defense of the claim that one can have noninferential knowledge of, or noninferentially justified beliefs about, one's own present beliefs. One way in which one might try to deal with this latter issue was briefly set out above.

(4) As was argued in the last section, there appear to be a number of ways in which direct realism is exposed to difficulties that do not apply to the indirect realist's account of the justification of beliefs about the past.