5.1 Introduction

Memory is a cognitive category that has a very illustrious history and an entrenched position in our cultural and linguistic practices. Episodic memory is a more recent classification, though it is arguably a category that many cultures possessed implicitly before it was explicitly identified by psychologists, even though they may not have distinguished it consistently from what are now widely considered other subtypes of memory.Footnote ¹ In recent taxonomic practice in the cognitive sciences, a distinction is usually made between episodic memory and semantic memory, where the former concerns the capacity to retain information from experiences pertaining to events that occurred in one’s own personal experience, while the latter involves a capacity to do the same with information of a more generic type that is retained independently of the circumstances in which it was acquired (cf. Foster Reference Foster2009, 39–40). First explicitly identified and labeled by Tulving (1972), episodic memory is often what we mean by the word “memory” in English: our capacity to retain and later recall autobiographical experiences, and the mental states produced by this capacity. The rest of this chapter will look at the arguments for and against considering episodic memory to be a valid category that corresponds to a real kind in the cognitive sciences. But before adjudicating that debate, it is worth considering the question as to whether the superordinate category memory corresponds to a real kind. Various theorists have argued that memory does not correspond to a real cognitive kind, so it is worth briefly exploring some of the arguments that have been put forward for this conclusion.

The standard scientific taxonomy of memory now includes a number of sub categories. I have already mentioned the categories of episodic and semantic memory. These two categories are subsumed under the superordinate category, declarative memory. Declarative memory is often distinguished in turn from non-declarative memory, which itself is subdivided into two or more subtypes, including procedural memory, classical conditioning, and others. This yields the taxonomic picture illustrated in Figure 5.1 (a), which shows that the over-arching category memory is now thought to be subdivided into at least two main types (declarative and non-declarative), and the latter is in turn subdivided into two main types (semantic and episodic). It may already seem unlikely that such an apparently heterogeneous category (memory) could correspond to a real kind. Given the diversity of the types of abilities and information associated with each cognitive capacities, there may seem to be insufficient commonality among them. Moreover, on some taxonomic divisions of memory, a further distinction is often made between long-term and short-term memory. Sometimes this distinction is made before the distinction between declarative and non-declarative memory, on the grounds that both of these types are sub categories of long-term memory, as shown in Figure 5.1. (b). But at other times, the distinction between long-term and short-term memory is considered orthogonal to the declarative/non-declarative distinction; in other words, there can be both long- and short-term declarative memory, as well long- and short-term non-declarative memory.

Figure 5.1 Two possible taxonomies of memory: The second incorporates the distinction between long-term and short-term memory and posits that both declarative and non-declarative memory are species of long-term memory.

A number of authors have argued that there is no superordinate category of memory that plays a useful role in cognitive science. Michaelian (Reference Michaelian2011b) points out that the different subtypes of memory are now widely considered to correspond to different cognitive systems. He proposes to individuate memory systems with reference to Marr’s three levels, the computational, algorithmic, and implementational levels, which are foundational for contemporary cognitive science (and which have already been encountered in Chapter 1, and deployed in individuating the kind concept in Chapter 2). That is, a cognitive system is distinguished from another cognitive system based on the properties that it has at each of these three levels. Accordingly, he defends the conditional claim that if the “multiple memory systems hypothesis” is correct (i.e. that memory is subdivided into distinct systems roughly along the lines indicated above), then memory is not a natural kind. His basic argument relies on saying that there are important differences between declarative and non-declarative memory at all three Marrian levels, which indicates that they are not subtypes of a single overarching kind, memory. By contrast with Michaelian, I think that it would be sufficient to show that memory is heterogeneous at the computational level to establish that it is not a unitary cognitive kind, so I will concentrate on that claim.Footnote ² In keeping with the general approach adopted in this book, the computational level identifies causal systems that are roughly coextensive with the cognitive domain, which is the domain of interest in this ontological investigation. At the computational level, Michaelian argues that while declarative memory is involved in information-processing and is cognitive in nature, non-declarative memory is not involved in information-processing and is noncognitive. Since it is not so involved, there is no computational-level description adequate to both declarative and non-declarative memory. One might object to this by saying that there is information-processing in non-declarative memory, though it is of a nonpropositional or implicit variety. This objection could be elaborated by saying that all cases of memory are “instances when information of the past is made available for present purposes” (Werning & Cheng Reference Werning, Cheng, Bernecker and Michaelian2017, 7).Footnote ³ However, as Werning and Cheng (Reference Werning, Cheng, Bernecker and Michaelian2017, 7) point out: “In this minimal sense the rings of trees are memories of the climatic conditions in the seasonal succession of years during certain periods of the past.” Thus, even if non-declarative memory is considered to involve information processing (albeit nonpropositional), there does not seem to be an overarching causal profile that can be attributed to all and only instances of memory at the computational level. This minimal characterization pertains to noncognitive phenomena, and would preclude memory being a real kind, at least a real cognitive kind.Footnote ⁴ We could try to avoid this problem by further narrowing down the characterization of memory in computational terms by referring to the processes of encoding, storage, and retrieval, which are often thought to be distinguishing characteristics of memory. But Klein (Reference Klein2015, 1) points out that if we take memory to include “any state or process that results from the sequential stages of encoding, storage, and retrieval,” that would include virtually all aspects of cognition. In other words, this characterization does not pick out a distinctive category at the computational level. This is not entirely convincing, since even though a preponderance of our cognitive states may be subject to encoding, storage, and retrieval, that does not imply that all our cognitive capacities perform this role. The reason that this may not be obvious is that our memory capacities interact with a broad range of other psychological capacities, such as perception, inference, and emotion, so their output states are generally the joint product of memory and other systems. That may be why this characterization of memory might seem to encompass all of cognition. Therefore, contrary to Klein’s contention, it is at least worth considering whether any cognitive system that implements a process of encoding, storage, and retrieval may be considered a memory system, and whether the class of all such systems corresponds to a cognitive kind.

I have presented some recent arguments against the claim that memory in general can be considered a cognitive kind comprising a number of subkinds (episodic memory, semantic memory, procedural memory, and so on). These arguments contend that there are no common computational or cognitive properties shared by all of the various alleged types of memory, at least not ones that are not also shared by phenomena that are not instances of memory. But these arguments do not appear to be decisive and there may be room for understanding memory as a general capacity for encoding, storing, and retrieving informational content, or perhaps more convincingly, as a capacity for specific types of encoding, storage, and retrieval. Thus, it may be that there is a cognitive kind that includes all the capacities included in Figure 5.1, or it may be that there is a less inclusive kind declarative memory that includes both episodic memory and semantic memory, or it may be that episodic memory is not a subkind of some more general type of memory. In what follows, in investigating whether episodic memory is a real kind, I will pursue the issue independently of the question of its superordinate and subordinate categories, if any. (For these purposes, we can think of “episodic memory” as a simple rather than a composite term indicating both a species and a genus.)

5.2 What Is Episodic Memory?

In the previous section, I mentioned episodic memory and other types of memory as cognitive capacities or systems, and that is how they are often theorized about in cognitive science. But there is another prominent use of the term “memory” in common parlance and in scientific discourse (at least in English), which corresponds roughly to the output of such a system, a type of mental state with informational or representational content. In what follows, when it comes to episodic memory, I will be discussing both of these cognitive categories, the capacity and the state, since I think that there is a close connection between the capacity for memory and its outputs, as I will argue in due course.Footnote ⁵ This also accords with common English usage, since we talk about episodic memory both as a capacity (e.g. “My memory for recent events is not what it used to be”) and a state (e.g. “I have a fond memory of eating pizza for breakfast last Sunday”). Moreover, when it comes to the cognitive state of episodic memory, there is a possible confusion between two different cognitive entities. One is the mental state that a cognitive agent is in at the time of recollection or retrieval, the memory produced on a particular occasion, and the other is the item that is presumably held in storage. On a naïve view, these two entities are virtually identical: The item that is retrieved by an agent in response to some external or internal stimulus is commonly thought to be the very same item that has been held in storage since the actual experience. But we will see later that this naïve view, which has been called the “archival view” of episodic memory (Robins Reference Robins2016a), cannot stand up to scrutiny. Accordingly, I will argue that the primary objects of investigation in memory research are the recollected memory state, what is retrieved by a particular agent on a particular occasion, and more centrally, the capacity that produces such states.

As indicated in the previous section, the first explicit identification of episodic memory as a distinct mental capacity is usually traced to Tulving (Reference Tulving, Tulving and Donaldson1972), though he himself states that the distinction was already implicit in both psychological research and in the philosophical literature. In that work, Tulving characterized episodic memory as a system that stores information with spatiotemporal and autobiographical content (e.g. “I remember that I met a retired sea captain on my summer vacation last year”). By contrast, Tulving described semantic memory as a “mental thesaurus” that stores information necessary for the use of language (e.g. “I remember that summers in Katmandu are usually quite hot”) – though as this brief characterization suggests, it might be better characterized as an encyclopedia.Footnote ⁶ As it stands, this way of making the distinction cannot be quite right, since I remember the date and place of my birth whenever I have to fill out certain official forms, which is information with autobiographical and spatiotemporal content, but that is more plausibly considered a product of semantic memory (similar to my memory of, say, my mother’s date and place of birth). Rather, the intended distinction seems to be between memories of an event that are based on a direct experience of the event itself and those that are not. Moreover, according to some researchers this experience is in some sense “relived” when the memory is retrieved. In later work, Tulving went on to characterize episodic memory in terms of “autonoetic” (“self-knowing”) consciousness. On this revised view, episodic memory involves a distinctive kind of mental experience, or a “special kind of consciousness that allows us to be aware of subjective time in which events happened” (Tulving Reference Tulving2002, 2). Many other researchers have agreed with Tulving that the phenomenal quality of reliving a past experience is what sets apart episodic memory from semantic memory.

There are several things to notice about this more recent characterization of episodic memory and its differentiation from semantic memory. First, it applies primarily to memory as a state rather than a capacity, since autonoetic consciousness evidently pertains to an occurrent mental state that is entertained by a cognitive agent, rather than a mental capacity of some kind. But it seems possible to use this characterization to identify the associated capacity (at least in part) as one that is disposed to produce states of that kind. Of course, even if states of episodic memory belong to a real kind, that does not guarantee that there will be a real kind of capacity dedicated to producing such states, but it is at least a live possibility. Second, by grounding episodic memory in autonoetic consciousness, researchers do not usually consider that episodic memory is necessarily ineffable or essentially subjective. Although states of episodic memory are characterized in phenomenal terms, episodic memory researchers like Tulving usually consider that this autonoetic aspect is a property that is amenable to scientific investigation from a third-person perspective, even though any investigation will rely heavily on self-reports by experimental participants.Footnote ⁷ They have developed various experimental techniques to try to assess this. As Tulving (Reference Tulving1985, 6; original emphasis) states: “one way of measuring autonoetic awareness could take the form of asking people, when they recall or recognize a previously encountered item, whether they remember the event or whether they know in some other way that it occurred.”Footnote ⁸ Other cognitive scientists have adopted this technique, notably in experimental paradigms in which participants are shown lists of words and then later asked whether certain words appeared on the original lists.Footnote ⁹ Admittedly, asking people whether they actually “remember” an item or whether they “know” it in some other way is not a foolproof technique for determining whether they have the requisite autonoetic consciousness. But it would seem to provide a fallible way of accessing the experience of the cognitive agents and the nature of the mental state in question. Third, despite the fact that Tulving and many other memory researchers distinguish episodic from semantic memory, many of them also see close links between the two capacities (and their respective states). In particular, some semantic memories may be strongly associated with episodic memories, specifically concerning the occasion on which one acquired the information. For example, I remember that Pretoria is the capital of South Africa (semantic memory), but I also remember when I first learned that piece of information, namely when I gave the wrong answer in a school general knowledge contest (episodic memory). Even when episodic and semantic memories are not associated in this way, it is plausible that some semantic memories originate in episodic memories, many of which are later forgotten (though this is not to say that semantic memory is just the accumulation of episodes; see Baddeley Reference Baddeley2001). For instance, I no longer remember when I first learned that Paris is the capital of France, but I do remember that fact. Similarly, some episodic memories can be strongly informed or supplemented by semantic memories. I might not remember all the details of a particular past experience but can “fill them in” using background information (semantic memories) in trying to recollect them to myself or recount them to others (as will be further elaborated in Section 5.4). Notwithstanding these important connections between episodic memory and semantic memory, there is also an impressive body of evidence that they are doubly dissociated. That is, there are instances in which human thinkers exhibit severe deficits in one without appreciable shortcomings in the other. This double dissociation is supported by the existence of patients with acutely impaired episodic memory and largely spared semantic memory (Tulving Reference Tulving2002, 13–16), as well as by semantic dementia patients who have severe deficits in semantic memory with relative preservation of episodic memory (Irish, Addis, Hodges, & Piguet, Reference Irish, Addis, Hodges and Piguet2012). Hence, Tulving’s original distinction has been corroborated by subsequent empirical work revealing a double dissociation between the two posited memory systems.

Thus far, we do not seem to have resounding evidence that we are dealing with a real cognitive kind. Despite some apparent confluence with first-person experience and vernacular categorization, episodic memory, when characterized in terms of its “autonoetic” phenomenal features, does not seem to have the causal profile found among other cognitive kinds discussed so far in this book. However, even though Tulving’s characterization of episodic memory in terms of its experiential properties has gained wide currency, there are other ways of individuating it in the psychological and philosophical literature. One way that is particularly prominent in philosophical discussions (and at least implicitly in psychological research) has to do with the claim that it is a type of mental state that carries information about one’s personal past. This way of identifying episodic memory relies not on its phenomenal properties but on its etiology and is often associated with the “causal theory of memory” proposed by Martin and Deutscher (Reference Martin and Deutscher1966). Their theory is based largely on conceptual analysis of the terms “memory” and “remember,” informed by various hypothetical scenarios.Footnote ¹⁰ By examining a number of cases and variations, they emerge with a definition of the vernacular term “remember,” whose main tenets are roughly as follows:

They also add two further clauses to (3). The first relates to cases in which the rememberer is prompted to remember by some clue or hint. In such cases, Martin and Deutscher hold that X’s past experience of the thing represented must be operative in producing the state (or the successive set of states) in X that is finally operative in producing the representation. The second concerns the relation of the representation to the experience: The state or set of states produced by the past experience must constitute a structural analogue of the thing remembered, to the extent to which X can accurately represent the thing. The upshot of this view is that episodic memories are states that are causally connected in an appropriate (i.e. non-deviant) way to a representation of an experience in a subject’s personal past. This connection is secured by means of a “trace,” which Martin and Deutscher (Reference Martin and Deutscher1966, 189) describe as “an indispensable part of our idea of memory,” describing it as a “structural analogue of what was experienced.” As Michaelian (2011, 332) explains in his exposition of their theory, “the memory trace has to exist all along; and it has to be doing causal work at the time of the remembering,” even if the memory is cued in some way. Thus, there is an uninterrupted causal chain between initial experience and subsequent representation that is mediated by means of a trace.

There is something intuitive about this notion of episodic memory and it seems to be in accord with at least many uses of the terms “remember” and “memory” in English. As Martin and Deutscher ingeniously show by using various hypothetical cases, we are loath to consider a current state to be a memory unless it has something like the causal history that they specify. This is also at least the implicit position of many researchers in psychology, who regard a causal link to the past as a crucial feature of episodic memory states, alongside their distinct autonoetic phenomenal properties. Klein (Reference Klein2015, 22) puts it as follows:

Even though Klein agrees with many other psychologists that autonoetic phenomenology is crucial to episodic memory, he also takes it as uncontroversial that episodic memories must be traceable back to past experiences. As we shall see in Section 5.3, there are some complications associated with this picture of episodic memory, especially in light of some empirical evidence that seems to undermine the existence of a distinct capacity of episodic memory. But before looking at these difficulties, it is worth asking why a science of the mind should be interested in identifying states of this type and categorizing them separately from other mental states, including states of semantic memory, as well as states of imagination, perception, learning, inference, and others.

We often have occasion in science to type entities based on their etiology or causal history. This holds of real kinds in biology (e.g. species), geology (e.g. sedimentary and igneous rocks), astronomy (e.g. meteorite), and other sciences (cf. Khalidi Reference Khalidi2021). Moreover, we saw in Chapter 2 that cognitive scientists have good reason to individuate concepts in part according to their causes. In many cases, categorizing according to causal history in science is in the service of explanation. This would seem to apply to the case of episodic memory in particular. Suppose I see my next-door neighbor arrive at the door of her house, rummage around in her pockets, then lift up the flowerpot near the door and extract a key from under it, before letting herself in. In the typical case, the best explanation of why she looked for her spare key under the flowerpot is that she remembers having left it there. In other words, she has a memory that represents the key being in that location, and that mental state is caused ultimately by a past event in which she placed the key under the flowerpot. But one may object here that this explanation does not need to appeal to a state of memory that can be traced back to an episode in her past. What explains her behavior of looking under the flowerpot is simply a mental state that represents the key being there, whether or not that mental state has the right causal history. After all, she may well look for the key there even if she just imagined having put it there, or dreamt it the previous night, or had a vague hunch. But if we want to explain why she succeeds in finding it when she relies on her memory (in the event that she does succeed), then the explanation would need to rely on a causal connection to a past episode. Notice that she need not succeed each time she relies on her memory, she need not even succeed more often than not, but when she does succeed, the causal connection provides the best explanation of why she does. Otherwise, the success of the behavior would be a mysterious coincidence.Footnote ¹¹

This particular explanation is couched in a folk psychological context, but similar explanations occur in the context of scientific psychology. Consider, for example, the well-known Deese–Roediger–McDermott (DRM) paradigm, in which experimenters show participants lists of closely related words (e.g. table, sit, legs, seat), then determine how many words are correctly remembered, as well as which words are incorrectly recalled and how frequently. In this paradigm, it is found that participants often incorrectly recall words that are not on the list when those words are closely associated with words on the list (e.g. chair). In a seminal experiment, the mean probability of recall of studied words was 0.65, whereas the mean recall of closely associated words (“lures”) was 0.4 (Roediger & McDermott Reference Roediger and McDermott1995). While high, this latter probability was significantly lower than the probability of recall of studied words. This experiment shows that words that were not encountered, yet closely related to the encountered words, were incorrectly recalled at a relatively high rate. Here, what is in need of explanation is the production of lures, not the production of the studied words. It does not even occur to researchers to explain the latter fact, since the explanation is obvious: These words were produced precisely because they were previously encountered. This is not to say that all previously encountered words are correctly recalled in such experimental setups, but they are recalled often, and this is just what we would expect. In fact, as emphasized by Robins (Reference Robins2016a) the explanation of why the lures were produced depends on the explanation of why the studied words were produced: They are semantically related to them.Footnote ¹² There is no need to explain why most experimental participants succeed in recalling many of the studied words. That is the background against which much of the empirical work on memory proceeds: When items are successfully produced it is generally because they have been previously encountered. It is possible that some of these are simply lucky guesses, but in experimental setups of this type it is vanishingly unlikely that they all are.

Even the explanation of cases in which experimental participants claim to remember entire events that never occurred occurs against a background that presupposes that actual memories are causally connected, via some actual trace, to some episode in the past. Consider the influential work of Loftus and Pickrell (Reference Loftus and Pickrell1995) in which subjects are implanted with a “false memory” that they were lost in a mall during childhood. In attempting to explain why around one-quarter of their experimental participants were induced to falsely remember this incident (with the help of relatives who were in the know and conspired with the experimenters), the researchers presuppose that actual memories are created when an episode leaves a trace in the subject’s mind–brain. It is worth quoting them at length:

These psychologists are at pains to explain how such mental states can be generated, and their explanation is parasitic on an explanation of how actual memories are formed. Admittedly, the details of memory formation have not been provided, in particular of how events or episodes generate representations (“grains of experienced events”), which are then stored as traces, and later retrieved. Yet there is a pervasive assumption in much memory research that a process of this kind occurs in the case of genuine memories, even though the cognitive sciences have not yet uncovered the specifics of the hypothesized process, whether at the algorithmic or implementational levels.

The standard explanation of why actual episodic memories succeed (when they do succeed) in revealing details about the past is that they are causally linked in such a way to past events as to enable them to faithfully transmit information. Indeed, cognitive scientists explain memory errors (lure words in the DRM paradigm or “false memories” in Loftus’ experiments) against the background of successful memories, which they assume involve traces of past events. The details are still obscure, or at least there is no consensus on them, but something like a causal transmission process is bound to be in place for such an explanation to succeed. As already mentioned, this inference to the best explanation does not require episodic memories to be wholly or even mostly accurate or veridical when it comes to representing the past. We do not even need to assume that memories are veridical more often than not, or that the track record of memory is better than, say, imagination. Even if it turns out that memory is only occasionally accurate or veridical, those occasions would need an explanation. Compare: If dreams enabled us to predict the future, even sporadically, that fact would cry out for an explanation. The explanation of the fact that states of episodic memory succeed in accurately representing past events is that the mental states in question transmit information from that past event more or less faithfully by virtue of being causally connected to it via a trace or representation of the past event. That would seem to be sufficient scientific justification for associating episodic memory with a causal link to a past episode.Footnote ¹³

We have now encountered the two most prominent approaches to characterizing episodic memory: phenomenal and etiological. Both of these are attested in the philosophical and psychological literatures. Indeed, the lengthy passage from Loftus and Pickrell quoted above can be interpreted as linking the two characteristics, in attempting to explain why certain states of mind that do not have the requisite etiology can yet appear to subjects to be real memories, that is, have the right phenomenal features. There is a large body of empirical work that relies on these two ways of characterizing episodic memory. Given that this is the case, we can ask two questions. First, what reason do we have for thinking that these two purported features of episodic memory (usually) coincide? Second, do either or both of these features give us grounds for thinking that states of episodic memory or the cognitive capacity that generates such states corresponds to a real kind? Before tackling these questions, it is worth looking at some challenges to the notion that episodic memories (states) or episodic memory (capacity) are real kinds.

5.3 Empirical Challenges

A number of cognitive scientists have challenged the existence of a distinct category of episodic memory and have instead posited that humans have a more general capacity for “mental time travel” (Suddendorf & Corballis Reference Suddendorf and Corballis1997), “episodic hypothetical thinking” (De Brigard Reference De Brigard2014), or “constructive episodic simulation” (Addis Reference Addis2018), into which episodic memory is submerged. Although the connection between episodic memory and certain forms of imaginative, hypothetical, or prospective thinking had been drawn at least as early as Tulving (Reference Tulving1985), this recent work goes further by claiming that there is no distinct category of episodic memory at all and that what we call “episodic memory” is really just a manifestation of a cognitive system with a much broader function.Footnote ¹⁴ The evidence that this hypothesis is based on can be divided broadly into two main categories: neuroimaging evidence and evidence from neuropathology. In this section, I will survey this evidence briefly and try to understand its implications for the question of the status of episodic memory as a real kind.

There is an appreciable body of neuroimaging evidence from fMRI studies indicating that tasks requiring the use of episodic memory and those that involve simulation recruit the same neural network, or at least overlapping neural regions. Addis (Reference Addis2018, 73) claims that the neural system in question is the “default mode network” (DMN), a set of interconnected regions that include “medial temporal lobes, medial aspects of the frontal and parietal cortices, inferior lateral parietal cortex and lateral temporal cortex.” The DMN was originally identified as a result of imaging experiments in which participants lying in the scanner between tasks generally exhibited activity in a certain combination of neural regions.Footnote ¹⁵ Accordingly, it was posited that these regions jointly subserve “default” activity in the brain, that is, they are active when a subject is not carrying out any particular cognitive task. Later research maintained that the network was associated with daydreaming or “mind-wandering,” since this is presumably what subjects are doing when they are not engaged in specific cognitive tasks (e.g. reading a word, solving a puzzle). But Addis (Reference Addis2018) and other researchers claim that the DMN performs the function of imagination or simulation, and that this cognitive function subsumes what we have labeled “episodic memory.” Other theorists (including Addis herself in earlier work; see e.g. Schacter, Addis, Hassabis, et al. Reference Irish, Addis, Hodges and Piguet2012) take a more moderate view, claiming only that this network, or parts of it, are involved in both episodic memory and some types of imaginative thinking, perhaps because imagination draws on memory in some ways.

A number of neuroimaging studies are thought to support the view that there is no distinct capacity for episodic memory, but rather a single cognitive system for episodic recollection, future projection, episodic counterfactual thinking, and similar cognitive processes. These studies tend not to find an exact coincidence between the networks recruited in the tasks relevant to all (or some of) these cognitive functions, but considerable overlap. For instance, one study found “significant neural overlap between brain regions engaged during autobiographical recollection and those engaged during episodic counterfactual thinking” (De Brigard, Addis, Ford, et al. Reference De Brigard, Addis, Ford, Schacter and Giovanello2013, 2408). Needless to say, significant overlap when it comes to neural correlates cannot be taken as conclusive evidence for the equivalence of the corresponding cognitive capacities, or their subsumption into some overarching cognitive function. There are a number of alternative interpretations for findings of this type. One would be to maintain that episodic memory relies upon some of the same resources as these other types of thinking (i.e. prospective thinking, hypothetical thinking, counterfactual thinking, imagination), though it is a distinct cognitive capacity. This is similar to what was described as Addis’ former view above (see e.g. Schacter, Addis, Hassabis, et al. Reference Irish, Addis, Hodges and Piguet2012), namely that imagination draws on memory, or that they both depend on common cognitive systems that perform other functions. One candidate proposed in the psychology and neuroscience literature for this function is “scene construction” (see e.g. Suddendorf, Addis, & Corballis Reference Suddendorf, Addis and Corballis2009; Szpunar Reference Szpunar2010). Szpunar (Reference Szpunar2010, 157) describes this cognitive function as a capacity for constructing a “coherent representation of a specific scenario,” which may well be needed both for reconstructing memories and for constructing states of imagination or prospective thinking. In other words, activation in this network might correspond to a component cognitive process that is common to these different functions. It need not indicate identity among these cognitive functions. A completely different interpretation of these findings would have it that neural overlap, or even coincidence, as indicated by neuroimaging studies is inconclusive when it comes to the cognitive or computational level because of the phenomenon of neural reuse. As discussed in previous chapters (see especially Section 1.4), the activation of a neural region or indeed an entire network is not a decisive indication of cognitive function. Even without taking into account different levels of activation in various regions that emerge across these tasks, neural reuse holds that the very same regions or networks can be deployed for different cognitive tasks depending on phenomena such as neuromodulation of neural circuits by chemical and genetic means. Therefore, the neuroimaging evidence does not indicate decisively that episodic memory is not a distinct cognitive capacity. It can be interpreted as implying either that it taps into some of the same cognitive resources as other cognitive capacities, or that it recruits some of the same neural resources as they do.

Another source of evidence cited in support of the hypothesis that episodic memory is not distinct from other cognitive capacities, particularly those for prospective or future thinking, comes from neuropathology. This evidence dates back to some of Tulving’s seminal work on patient KC, who was involved in a motorcycle accident at age thirty that resulted in multiple brain lesions particularly in the medial temporal lobes, and suffered from severe amnesia as a result. He experienced both anterograde and retrograde amnesia for episodic memories, meaning that he could not recall personally experienced events that occurred before the accident, nor could he form new memories of events that he experienced after the accident (Tulving Reference Tulving1985, 4–5; Tulving Reference Tulving2002, 13–14). In addition to these deficits, KC was unable to engage in thinking about the future, and when given the opportunity to do so, would generally draw a blank (Tulving Reference Tulving1985, 4). Subsequent research on some other amnesic patients has revealed similar deficits when it comes to prospective thinking. In particular, some other patients with damage in the medial temporal lobe (which includes the hippocampus and is often considered to be an integral region in the DMN) also exhibit similar deficits. One of the most widely cited studies tested a group of five amnesic patients with bilateral hippocampal damage to determine how they compared to controls in imagining future experiences. The study found that such patients not only suffered from severe amnesia but had serious deficits when it came to imagining new experiences and fictitious scenarios (Hassabis, Kumaran, Vann, et al. Reference Hassabis, Kumaran, Vann and Maguire2007). Such findings are sometimes cited to support the contention that episodic memory is not a distinct cognitive capacity, but rather is just a manifestation of a broader cognitive capacity that enables us to engage in “mental time travel,” whether into the past or the future. However, this study posits not that episodic memory is one manifestation of a cognitive capacity that is also dedicated to imagination and prospective thinking, but rather that these capacities draw on some of the same cognitive resources. In particular, the researchers conclude that the “patients’ imagined experiences were strikingly deficient in spatial coherence, resulting in their constructions being fragmented and lacking in richness” (Hassabis, Kumaran, Vann, et al. Reference Hassabis, Kumaran, Vann and Maguire2007, 1729). Hence, they hypothesize that the hippocampus, which was the site of neural damage in all five participants, “may make a critical contribution to the creation of new experiences by providing the spatial context or environmental setting into which details are bound …” (Hassabis, Kumaran, Vann, et al. Reference Hassabis, Kumaran, Vann and Maguire2007, 1729). Therefore, the results from neuropathology do not seem decisive in rejecting the reality of a capacity of episodic memory. Like some of the neuroimaging studies mentioned, one interpretation of these findings would simply be that episodic memory and the ability to describe future scenarios draw on some of the same cognitive resources, and that the patients in question are impaired in these other abilities, which affect episodic memory as well as other cognitive capacities. Finally, it is worth mentioning that patients with severe deficits in episodic memory do not seem to be impaired when it comes to another aspect of future thinking, namely temporal discounting. When some amnesic participants are asked whether they would prefer to receive a monetary reward in the present (e.g. $50 now) or a somewhat more valuable monetary reward in the future (e.g. $100 in three years), they respond similarly to controls, namely by discounting future rewards to roughly the same degree (Kwan, Craver, Green, et al. Reference Kwan, Craver, Green, Myerson, Gao, Black and Rosenbaum2015). Hence, it is not the case that amnesic patients are thoroughly compromised in their ability to think about the future.

We have now encountered two prominent challenges to the reality or kindhood of episodic memory as a cognitive capacity, which argue for the elimination of episodic memory, but we have also seen that neither of these eliminativist proposals is conclusive. There are other, more limited, challenges both to episodic memory as a type of capacity and to episodic memories as types of states, which tend to go by the name of “constructivism.” It is difficult to summarize the variety of positions that are commonly tagged with this label, but perhaps the single common denominator is that episodic memory is not a preservative capacity that simply encodes, stores, and retrieves traces of past experiences. Rather, episodic memories are composed of representations that are constructed in response to endogenous or exogenous retrieval cues and combine resources from a variety of sources. These representations may harbor traces of past experiences, but they are cobbled together using information from semantic memory, perception, the retrieval cue, and elsewhere. As I have just characterized this family of positions, it does not obviously challenge the kindhood of episodic memory. But a more radical variation of this view, which denies the existence of traces altogether, or denies that they are a necessary component of memory states, does challenge the kind as I have characterized it so far. In the previous section, it was proposed that episodic memory states are distinguished at least partly on the basis of their etiology. That etiology is a matter of harboring a trace of a past experience, so if traces are denied altogether, that would undermine the proposed characterization of this representational kind. Moreover, it would erase the distinction between this representational kind and other such kinds, like states of imagination, which need not bear such traces. In practice, few cognitive scientists adopt such a radical position and those who do tend to subscribe to the eliminativist position just discussed, which would submerge episodic memory into a broader cognitive capacity. Most of those who consider themselves constructivists still incorporate a trace into their memory states. Hence, the etiology of memory is still important to most brands of constructivism, even though it is not emphasized to the exclusion of other aspects of the episodic memory state. For example, Schacter (Reference Schacter1996, 71) puts it as follows:

The reference to the “engram,” which is the hypothesized neural substrate of the memory trace, as well as the “fragmentary remains of experience,” indicate that Schacter is committed to the idea that states of episodic memory incorporate some distinct representational resources caused by a past experience. This is fairly typical of constructivist approaches to episodic memory, even though few if any cognitive scientists would claim to know how precisely the capacity of episodic memory concocts memory states out of a variety of representational resources, nor how these resources combine to issue in the memory state itself. But the existence of a trace, as at least one component of states of memory, is a presupposition of many if not most constructivist approaches to episodic memory.

In this section, I have considered some of the most prominent challenges to the idea that episodic memory is a real cognitive kind, both as a type of cognitive capacity and as a type of mental state. In the case of the eliminativist position that contends that there is no capacity of episodic memory distinct from a broader capacity for simulation, I argued that the evidence from neuroimaging and neuropathology is inadequate to support the position. As for constructivist approaches to memory, many of them are still committed to the existence of a memory trace. Memory traces are seldom explicitly defined in the empirical literature, but one notable exception is due to Tulving (Reference Tulving, Roediger, Dudai and Fitzpatrick2007, 66): “A memory trace is the neural change that accompanies a mental experience at one time (time 1) whose retention, modified or otherwise, allows the individual later (at time 2) to have mental experiences of the kind that would not have been possible in the absence of the trace.” This definition would seem to accord broadly with the notion of a memory trace utilized by Martin and Deutscher (Reference Martin and Deutscher1966), which posits an uninterrupted and nondeviant causal chain between a past experience and a representation at the point of recall, though Tulving explicitly identifies the trace with a neural change, while Martin and Deutscher do not. Moreover, Martin and Deutscher describe it as a “structural analogue” of the original experience, but this additional condition seems unnecessary (cf. Robins Reference Robins2016c). What seems to be required is that it be caused by the past episode, represent it, and be recoverable as pertaining specifically to it (in a word, traceable back to that episode). In Section 5.5, I will argue for a distinction between the trace, which can be identified at the computational level, and the engram, its supposed neural substrate, which belongs to the implementational level. But before doing so, I will make the case in Section 5.4 that episodic memory (both the state and the capacity) is a real cognitive kind.

5.4 Episodic Memory as a Cognitive Kind

In previous sections, I introduced etiological and phenomenological approaches to identifying episodic memory, and I also outlined some challenges to episodic memory being a real kind, at least if it is understood to have a distinctive etiology and phenomenology. I argued that these challenges can be resisted on the basis of available empirical evidence, but that is not enough to establish episodic memory as a real cognitive kind, either as a kind of mental state or capacity. In order to do so, we need to determine whether it can be identified with a set of causal properties. In particular, we will need to specify these properties more precisely. Then, we will have to ascertain how these properties relate to one another, if at all. Finally, we will need to determine whether episodic memory states or the capacity of episodic memory has the causal profile to constitute a real cognitive kind. In this section, I will attempt to tackle these tasks, focusing primarily on episodic memory states.

When it comes to the posited phenomenal property of episodic memory states, which Tulving characterized as “autonoetic,” various researchers have tried to characterize it in qualitative terms. As we have seen, participants in experimental conditions are probed for their episodic memory using the contrast between the verbs “remember” and “know.” Moreover, many researchers have described episodic memory states as characterized by the quality of reliving an original experience or by an awareness that the event occurred in one’s past. Klein (Reference Klein2015, 2) says that an episodic memory state “includes the feeling that one is reliving a past experience – that is, it provides a directly-given, non-inferential sense that one’s current mental state reflects a happening from one’s past.” Similar proposals have been made by numerous others. However, on most psychological and philosophical accounts, the phenomenal properties of episodic memory are not causally inert or epiphenomenal, but have a distinct functional profile that causally influences both behavior and verbal reports (cf. Boyle Reference Boyle2020). Moreover, in healthy human subjects, episodic memories are thought to be readily recognizable as such and can be distinguished from semantic memories, as well as from mental states such as perceptions, dreams, and imaginations.

As for the purported etiological property of episodic memory states, there are two prominent difficulties pertaining to it, one of which is epistemic and the other ontological. On the epistemic front, there is a general problem when it comes to the reconstruction of etiology, since delineating the causal chain that links a unique past event to a current mental state is typically not amenable to direct investigation, and must be inferred indirectly, whether in everyday or experimental settings.Footnote ¹⁷ As for the ontological challenges, these have to do with the now widely acknowledged constructive nature of states of episodic memory. As we saw briefly in Section 5.3, constructivists about episodic memory hold, on the basis of a considerable body of empirical evidence amassed over several decades, that episodic memory states are generated from memory traces as well as other representational resources. Thus, such states are widely thought to be hybrids, wherein traces may be amalgamated with non-trace representational content or otherwise incorporated with representational resources that do not derive from the event that is the supposed subject of the memory. This means that the posited trace may not be capable of being isolated or distilled from the mental representation that is the state of episodic memory, since it is ontologically entangled with it in certain ways. Moreover, this will inevitably lead to questions about the extent to which the trace needs to figure in the resultant representation for it to qualify as a genuine state of episodic memory. I cannot pretend to have answers to these questions, but will try to say something about each of them in order to show that the etiological account of episodic memory is not hopeless.

The epistemic challenge of ascertaining whether a current representation incorporates an etiological trace is not insurmountable. Especially under experimental conditions, researchers can control aspects of an event or stimulus that participants are asked to remember and can compare it to their responses at a later time. As mentioned in Section 5.2, when subjects correctly recall previously shown words or objects, it is possible that what they say they remember are merely lucky guesses, but with enough trials and sufficient numbers of participants, experimenters can rule out guesswork and can determine when participants have actually remembered, and infer that they have indeed retained a trace of a past event. Numerous experimental paradigms have been designed to determine when subjects retain traces of past events.Footnote ¹⁸

The ontological question has to do with the extent to which “impurities” can be tolerated in an episodic memory state. Memory researchers have detailed numerous types of distortions that can affect episodic memory and it is now well known that a great many of our memories are subject to these systematic errors (for an overview, see Michaelian 2011). This means that memory states are not simply preserved mental representations of past experiences, but incorporate much else besides. Should we consider a mental state to be an episodic memory if it contains a negligible representational trace, despite the fact that it is largely composed of other types of representation?Footnote ¹⁹ This problem can be finessed by bearing in mind that the capacity of episodic memory is ontologically prior to the states of episodic memory, since the capacity is causally responsible for generating the states. That means that episodic memory states can be individuated primarily by their having been produced by the capacity of episodic memory. This dispenses with the need for trying to say how much of a trace is enough for a state to be a memory. If, as I will go on to argue in this section, there are grounds for positing a capacity of episodic memory whose function is in part to transmit traces of past experiences, then the states generated by such a capacity will generally incorporate traces. Depending on how exactly the capacity works and its propensity to malfunction, it may sometimes generate states that do not have such traces or have negligible traces, but those states can still be considered states of episodic memory so long as they are outputs of the episodic memory system.Footnote ²⁰

If the epistemic and ontological challenges to the proposed features of episodic memory can be met, it remains to be seen what we can say about the causal profile of episodic memory and specifically how its etiological and phenomenal features might fit together. There is no shortage of introspective and experimental evidence to suggest that mental states that have the phenomenal properties typical of episodic memory states do not always have a nondeviant causal link to specific episodes in the lives of subjects,Footnote ²¹ and conversely (although less prominently), that some mental states that do have such a causal link do not have the right phenomenology. But despite the fact that these two features do not coincide invariably, it seems frankly surprising that they coincide so frequently and reliably. As we have seen, even work that emphasizes the susceptibility of subjects to “false memories” acknowledges this fact. What is in need of explanation is the fact that the mental states that have the requisite phenomenal quality are reliably traceable to the actual past events that they purport to represent, and vice versa. Without denying our susceptibility to the implantation of “false memories” and the various memory distortions that we may fall victim to, the coincidence of these two seemingly unrelated features is in need of explanation. One way of accounting for this coincidence is by positing the existence of a cognitive system or capacity that both transmits traces of particular past events and produces representations with certain phenomenal properties. Hence, the existence of a kind of mental state that possesses these properties can be tied to the existence of a mental capacity that tends to generate states with both types of properties, etiological and phenomenological. States of episodic memory can be hypothesized to be generated by a type of capacity whose function is to transmit information about the past, though its function may not be to do so perfectly or even with high fidelity. The resultant states’ being traceable to past experiences allows them to carry information about those experiences. To be sure, it does not guarantee that they will carry information about past episodes, but to the extent that they do, this is a result of their etiology, and is explainable by it (as argued in Section 5.3). This capacity of episodic memory states to carry information about the past, albeit modified in the various ways stressed by constructivists about memory, may have a variety of different functions. There are numerous theories currently in play about the function of the capacity of episodic memory, many of which tie its function to the ability of cognizers to deal with future eventualities and contingencies.Footnote ²² If that is the case, then mere preservation or archiving is not the point, which would go some way toward explaining some of the results stressed by constructivists about memory.

Consider a theory about the function of episodic memory that is due to Klein, Cosmides, Tooby, et al. (Reference Klein, Cosmides, Tooby and Chance2002), who propose an evolutionary account of this cognitive capacity. On their view, episodic memory functions, in part, to restrict the scope of generalizations provided by semantic memory. In other words, part of the point of episodic memory is to provide exceptions to the generalized information supplied by semantic memory. They argue that in addition to a capacity whose function it is to deliver generic information gleaned from the past (semantic memory), we need a capacity that supplies exceptions in the form of unique episodes (episodic memory), since both are valuable when it comes to planning and decision-making, particularly in our social interactions with others. As Klein, Cosmides, Tooby, et al. (Reference Klein, Cosmides, Tooby and Chance2002, 318) elaborate: “A generalization is most useful when its scope is delimited: when it is accompanied by information specifying those situations in which it does not apply.” This quick summary oversimplifies what these researchers say about the function of episodic memory. But my aim is not to argue for the view as much as to consider it as a candidate theory that would attribute a plausible adaptive function to the capacity of episodic memory. If it can be shown that episodic memory has one or more adaptive functions that set it apart as a distinct capacity, that would help establish it as a cognitive kind. That is not the only way to establish episodic memory as a cognitive kind, since there are surely such kinds that are not adaptive.Footnote ²³ But it would put this capacity on a par with other cognitive systems, whose causal-functional profile establish them as real kinds. Moreover, the capacity itself would be an etiological kind that has evolved for a specific purpose, namely to generate cognitive states with certain characteristics. In this case, the chief characteristic of such states would be to bear information about past events that register exceptions to generic facts, notably about our social partners. If this hypothesis is correct, these information-bearing states are causally efficacious in equipping thinkers to plan and make decisions in the social domain, armed not just with generic information but more specific episodic information as well. This account also goes some way toward accounting for a moderate constructivism about episodic memory, since representations that enable us to plan and make decisions are likely to be ones that combine specific information about past episodes with generic information in such a way as to enable us to navigate our environment, particularly social aspects of the environment. Finally, this type of account forges a link between the capacity being a real kind and the states generated by the capacity being members of a real kind (see Figure 5.2).

Figure 5.2 Schematic diagram of episodic memory (capacity) and episodic memory (state) showing their relationship and their partial causal and etiological profiles.

Although this account is somewhat speculative, it is illustrative of an account that would make episodic memory a real cognitive kind. Not only does the capacity of episodic memory constitute an evolved adaptive capacity on this account, the states that it generates also belong to a real kind. That is because they not only share an etiology but they also play a causal role in the cognitive life of thinkers who are able to have such states, namely in facilitating social decision-making. In addition, this account can also be used to shed light on the phenomenal properties of episodic memory states. Why is it that such states share (by and large) a certain phenomenology, and what is the relationship between their etiology and their phenomenology?Footnote ²⁴ One prevalent theory in the psychological literature is that the phenomenal properties of episodic memories enable the thinker to distinguish episodic memories, which bear information about the past, from other mental states, such as states of imagination. On this “source-monitoring” or “reality-monitoring” framework, episodic memories have a distinctive phenomenology to set them apart from mental states that have a different source (e.g. imagination, inference) and are not rooted in a specific past event. For example, Szpunar (Reference Szpunar2010, 148) writes: “According to reality-monitoring theory …, these differences in phenomenological characteristics [between representations of imagined and actual events] are essential because they play a pivotal role in helping people to discriminate imagined events from actual memories.” Indeed, even Addis (Reference Addis2018, 68) acknowledges that there is a wealth of empirical research indicating phenomenal differences between episodic memories and states of imagination: “Relative to imagined representations, memories typically have more sensory detail …, contextual clarity …, and consistency with pre-existing knowledge that makes them more plausible …” This shows how the etiology and phenomenology of episodic memory may be linked and why these causal features go hand in hand. To summarize, on this account, the capacity of episodic memory is an evolved system whose function is (at least in part) to transmit representations of past episodes that constitute exceptions to generic information gleaned from the past (information that is stored in semantic memory). The representational states generated by the capacity of episodic memory are characterized primarily by their etiology (the link to past episodes), but they also play a synchronic causal role in cognition, in that they enable the thinker to plan and make decisions about the future, equipped with specific information about the past. Moreover, these states are also typically characterized by a distinctive phenomenology to enable the thinker to distinguish such states from those of imagination, inference, and indeed states of semantic memory, which do not have such a phenomenology.

This may seem like an adaptive “just-so story,” yet any complex cognitive capacity that delivers distinctive states and is doubly dissociated from other systems, is likely to be an evolved adaptive capacity. Other functions have been proposed for episodic memory, such as motivating prosocial behaviors based on past experiences (Boyer Reference Boyer, Boyer and Wertsch2009), anticipating future regret in order to plan accordingly (Hoerl & McCormack Reference Hoerl, McCormack, Michaelian, Klein and Szpunar2016), or demarcating epistemic authority for the purpose of communicating with others (Mahr & Csibra Reference Mahr and Csibra2018), and indeed, episodic memory may have evolved to serve multiple functions. But whatever function or functions it has been selected for, the capacity of episodic memory can be argued to be a real kind both due to its etiology (hypothesized history of natural selection) as well as for the causal role that it plays in the lives of the thinkers who have this capacity, namely to fulfill the function for which it was selected, in addition to possibly fulfilling other causal roles. Like many other adaptive capacities, episodic memory may have come to play other functional roles in the lives of thinkers, many of which have been investigated by social psychologists, such as: to enhance self-understanding, reduce boredom, harbor grudges, reminisce about dead loved ones, cope with thoughts of mortality, teach lessons to young people, and forge ties with friends, to name some (see e.g. Harris, Rasmussen, & Berntsen Reference Harris, Rasmussen and Berntsen2014). Moreover, the states that are the outputs of this capacity are also members of a real kind by virtue of sharing an etiology (being caused by past experiences and generated by the capacity of episodic memory) and having certain adaptive causal features, including phenomenal properties. The claim that episodic memory is a real cognitive kind does not depend on the specific adaptive hypothesis described above, but is compatible with any number of hypotheses about its adaptive function and its current causal role in the mental lives of thinkers. The same goes for the states that are the outputs of this system. This particular hypothesis not only posits a link between the evolved function of the capacity of episodic memory and the causal role it plays for cognizers, it also bridges the gap between this role and the phenomenal properties of the states produced by the capacity. On this account, the capacity is not assumed to be foolproof. As we have seen, there is experimental (and anecdotal) evidence to suggest that states with the requisite phenomenal properties may not bear information about episodes in one’s personal past. Such missteps may arise along the lines that Loftus and Pickrell (Reference Loftus and Pickrell1995) suggest, as mentioned in Section 5.3, or for some other reason. Cognitive systems are not flawless and may often malfunction, so this finding should not be surprising. But this account at least explains the convergence between etiology and phenomenology, whereby those states that are causally linked to past episodes in the life of the thinker and represent them are also phenomenally distinctive.

5.5 Is Episodic Memory a Neural Kind?

As argued in previous sections, there is a widespread view in the cognitive sciences, as well as in philosophy, that the presence and persistence of a trace is a necessary element in a state of episodic memory (cf. Robins Reference Robins, Bernecker and Michaelian2017, 76). Some researchers in the cognitive sciences prefer to speak simply in terms of “persistence” rather than a trace. One researcher puts it starkly as follows: “Persistence is an essential property of memory. Indeed if memories did not persist at all, there would be no memories” (Thompson Reference Thompson, Roediger, Dudai and Fitzpatrick2007, 199; cf. Eichenbaum Reference Eichenbaum, Roediger, Dudai and Fitzpatrick2007; Lisman Reference Lisman, Roediger, Dudai and Fitzpatrick2007). A commonly mentioned obstacle to the scientific study of traces or persistence in memory research is that behavioral measures are incapable of distinguishing cases in which a memory trace is absent from those in which it persists but there is a failure in retrieval. Partly for this reason, some researchers propose that traces should be investigated directly by examining their neural underpinnings. Lisman (Reference Lisman, Roediger, Dudai and Fitzpatrick2007, 206) puts it this way: “My view is that once memory molecules responsible for persistence are identified, the investigation of persistence can be put on a solid footing. It will be possible to separate persistence from retrieval processes by direct biochemical tests.”Footnote ²⁵ It is tempting to stake the claim that episodic memory is a real kind on certain neuroscientific discoveries, making it conditional on discovering a neural correlate of memory traces. As we have seen, in many scientific discussions of memory, the engram is invoked as the neural correlate of the trace that is posited to make an appearance in all episodic memories. Although I would not rule out the discovery of the elusive engram, in this section I will argue that the claim that episodic memory is a real kind should not be conditional on discovering a neural correlate for it.

Despite the fact that we may occasionally be unable to distinguish failure of persistence from failure of retrieval, it is clear that there are many instances in which we can, and that we have robust methods of doing so. Take, for instance, a case from classical conditioning rather than episodic memory. After a conditional response has been rendered extinct, rapid relearning of the response generally demonstrates persistence (cf. Thompson Reference Thompson, Roediger, Dudai and Fitzpatrick2007, 199–200). This is often used to show that the conditional response persisted but that it ceased to be retrieved after a certain point. Similarly, in episodic memory, when certain retrieval cues succeed in generating the memory after other cues have failed, we can infer that their failure was not due to lack of persistence or the absence of a trace. At most, it can be said that in some cases of failure of memory we may not be able to distinguish failure of persistence from failure of retrieval. But that would seem to be a feature of any attempt at reconstructing etiology. (Compare: In some instances, we cannot distinguish cases of phenotypic homology from ones of convergent evolution.)

Still, it may be said that the proof is in the pudding, and in this case, that means locating the actual trace that has persisted, or indeed its neural correlate, the engram. In response, it is worth making a distinction between the engram and the trace. As I have tried to explicate it, a trace is a theoretical posit that explains the success of episodic memory.Footnote ²⁶ It is a cognitive entity at what Marr calls the computational level that is causally linked to a past event and carries information about that event. A trace is posited because of the need to explain how it is that memory states succeed (when they do succeed) in capturing aspects of past events and enabling cognizers to meet present challenges. Indeed, traces are also invoked to explain some cases of misremembering, as Robins (Reference Robins2016a) has argued (see Section 5.2). Traces feature in computational-level explanations that relate cognizers to their environments and their causal history, both ontogenetic and phylogenetic. They also enable us to explain why cognizers behave as they do, why they are sometimes successful in doing so, and why such behavior may be the result of certain cognitive adaptations. By positing a trace, one makes no commitment to the nature of the neural structures that might implement such traces, though in humans and similar creatures something at the neural level must ultimately contribute to the encoding, persistence, and retrieval of traces. But traces may be multiply realized relative to their neural correlates, when it comes to different episodic memories in the same individual, across individuals, and across species. They may also be highly distributed rather than localized, and indeed, there is considerable evidence to suggest that episodic memory traces cannot be located in a specific brain region (see e.g. Thompson Reference Thompson1991; Nadel, Samsonovich, Ryan, et al. Reference Nadel, Samsonovich, Ryan and Moscovitch2000). Moreover, although the trace can be distilled in theory and in certain experimental settings from other ingredients of an episodic memory (e.g. information from semantic memory, perceptual cues, and other sources), it is not clear that the neural correlate of the trace should be thought of as a distinct or separable neural component of an occurrent neural state. Finally, it should also be emphasized that the trace may be dispositional in nature, which means that its neural correlate need not be manifested when the memory is not being retrieved. If so, the neural correlate of the trace as manifested in an occurrent mental state is not likely to be a neural structure or process that can be identified in the brain when the memory is not actually being retrieved.

If any of these features hold of episodic memory traces (multiple realization, distribution, dispositionality), a trace cannot be straightforwardly identified with an engram, at least if an engram is thought of as a standing neural structure, set of synaptic strengths, or pattern of neural activation. In addition, as I have characterized them, memory traces at the computational level are individuated with reference to their etiology. Two intrinsically identical representations may not both be traces, since they may differ in their causal histories. This means that any attempt to identify traces with neural correlates will have to contend with the fact that traces are individuated extrinsically rather than intrinsically. This poses certain obstacles to investigation of the purported neural correlates of memory traces, at least using current methods. Experimental participants can undergo neuroimaging scans when they are in the process of retrieving a memory, but determining whether the neural state that they are in bears a trace of a past event will be a challenging task given current experimental techniques. These obstacles may not just be practical ones, since as I have argued in previous chapters, etiological individuation is not widely attested in neuroscience and when it is, the etiological factors that are tracked are not always those of psychological or cognitive interest. Various philosophers of neuroscience have emphasized the importance of mechanisms and mechanistic explanation to the neuroscientific enterprise (e.g. Craver Reference Craver2007; Craver Reference Craver and Huneman2013), and to the extent that they are right, this does not comport well with the etiological or functional inquiries associated with episodic memory. Mechanistic explanations generally focus on organized systems and their parts, and explain their activities with reference to spatially arranged components that are contiguous to each other and interact by direct physical contact. Functional explanations, on the other hand, typically involve systems that are spatially dispersed and interact indirectly through intermediaries. As mentioned in Chapter 1, as understood in this book, computational explanations are a species of functional explanation. Computational explanations invoke information-bearing states and processes, which typically transmit information over relatively large spatiotemporal scales. When we say that a trace explains why episodic memory succeeds in representing past events, or that the success of memory explains why a capacity of episodic memory was selected for in our evolutionary history, we are abstracting away from details of implementation. Of course, there is room also for explaining how memory is able to fulfill its functions, and this is often a demand for a mechanistic explanation. But the two types of explanation are somewhat independent of one another, and the scientific research programs that are primarily engaged in one type of explanation have different methods and taxonomies than the ones that engage in the other. Moreover, they pertain to autonomous domains or causal systems, since computational-functional systems are multiply realizable by mechanistic systems.

An analogy from biology might help here. Homologous phenotypic features are generally individuated etiologically and they are identified on the basis of their relations to ancestral characters. Like the attempt to identify genuine memories, our ability to trace homologous features in different lineages to their ancestral antecedents are fallible, but the fossil record sometimes enables us to make such identifications with a high degree of reliability. It might be thought that the ultimate test of a hypothesis concerning homology would be uncovering the genetic basis of the phenotypic feature. But there is ample evidence to show that some homologies persist despite a change in their genetic bases; they come to be sustained by different genetic mechanisms because they perform an adaptive function that responds to certain abiding selection pressures (Brigandt & Griffiths Reference Brigandt and Griffiths2007, 634; Kendig Reference Kendig and Kendig2015). Though investigation of the genotype may inform research into homologous phenotypic characters, it is not decisive and cannot be taken to settle the question of homology. The comparison can be extended further: It is usually impracticable to investigate ancestral genetic material and one can rarely track genotypes historically. As in the neural case, longitudinal investigation of the underlying mechanisms is generally not available to scientists.

Previous sections point to the conclusion that the primary object of investigation in the cognitive sciences is the capacity of episodic memory and its outputs, that is the states of mind that coincide with recall or retrieval, as opposed to the posited stored trace, which seems much less amenable to investigation, let alone the hypothesized neural engram, which I have argued need not coincide with the trace. This makes memory primarily a “retrieval phenomenon,” as some cognitive scientists have proposed (Mahr & Csibra Reference Mahr and Csibra2018, 51; cf. Klein Reference Klein2013). Episodic memory states are causally efficacious in generating the behavior of cognitive agents and the success of such states in producing appropriate actions is explained by the information-bearing trace. Moreover, the emphasis on retrieval agrees with the moderate constructivist position and the source-monitoring framework, which were discussed in previous sections (see e.g. Koriat & Goldsmith Reference Koriat and Goldsmith1996). Episodic memory states need not represent the past with total accuracy as long as they represent specific content that qualifies more generic information, to assist in planning and decision-making.

I have also argued in previous sections that episodic memory states are likely the outputs of a dedicated cognitive capacity or system, the episodic memory system, whose function it is to transmit certain types of information about the past with a distinctive phenomenology. This raises the question as to whether the system as a whole may have a neural correlate, or may indeed be identified with a specific neural network. We already saw in Section 5.3 that some researchers have suggested a great deal of overlap between the activity of episodic memory and neural activation in the default mode network (DMN). It may well turn out that the episodic memory system is subserved entirely by a particular brain network, either the DMN or some other (presumably, partially overlapping) set of interconnected regions, which is responsible for the encoding, storage, and retrieval of representations about past events for use in the present. But before jumping to the conclusion that the capacity of episodic memory can be identified with this or any other neural network, a couple of caveats need to be kept in mind. One is that the DMN has also been identified with other functions, as eliminativists about episodic memory have noted. This means that there may be a high degree of neural reuse when it comes to the network as a whole or its component regions, so an identification of the cognitive system with a neural network is unwarranted. Moreover, if neural reuse is a basic principle of brain organization, we should not expect cognitive systems to be identical with neural regions or networks, but perhaps with those regions or networks in particular configurations, modulated in certain ways, or as they undergo certain levels or patterns of activation. Another, related, point is that some researchers hypothesize that the capacity for episodic memory is a “uniquely human property of the human mind” (Tulving Reference Tulving1985, 1), and if they are right about that, then evolutionary processes might not have had time to effect structural changes in the brain to accommodate this cognitive system. Moreover, if the episodic memory system is an evolved capacity, then it is individuated in part by the adaptive role it plays for creatures endowed with that capacity. That means that if we were to find similar capacities in other creatures, we might not consider them to be instances of episodic memory unless they were the outcome of the same or similar selection pressures. Hence, the capacity of episodic memory may itself be individuated etiologically and any attempt to identify it with its neural substratum will run into the problem that neural mechanisms are usually not so individuated, as already elaborated for the memory trace.

5.6 Conclusion

In this chapter, I have made the case for considering episodic memory (both the capacity and the states it produces) as a cognitive kind. In doing so, I have tried to synthesize philosophical work on memory with work in cognitive psychology. States of episodic memory are widely regarded by memory researchers as being connected via a trace of some kind to a past event or episode in the thinker’s life. I argued that this claim is not just a feature of the “causal theory of memory” favored by some philosophers but is in fact implicit in much of the empirical work in cognitive science, and is indeed required for the explanatory work that memory does in both ordinary situations and experimental settings. This means that states of episodic memory are individuated primarily with reference to their etiology or causal history: They can be traced back to past episodes. They are also characterized by a distinctive “autonoetic” phenomenology, which differentiates them from other mental states. I also argued that there is evidence to suggest that humans have a dedicated faculty of episodic memory whose function it is to produce such states. There are a number of theories that purport to explain why such a capacity would be adaptive and I selected one such theory as being illustrative of a range of similar theories. Since these theories of episodic memory consider it to be an adaptive capacity, it is also etiologically individuated with reference to its evolutionary history, as are other adaptive capacities. Moreover, this means that the states produced by such a capacity are doubly etiological, since they are individuated by virtue of being traceable to past events, but they are also individuated in virtue of being outputs of the capacity of episodic memory. Meanwhile, the capacity of episodic memory has the function of producing states that enable thinkers to plan and make decisions based in part on information from the past. Thus, this capacity has certain synchronic causal powers. There are also grounds for thinking that this capacity endows states of episodic memory with a distinctive phenomenology, which explains why states of episodic memory are also (typically) characterized by this phenomenal quality.

This account of episodic memory regards the capacity or system as primary since it is responsible for generating states of episodic memory, though both can be considered real kinds. It should be clear from this characterization that the cognitive kind episodic memory is externalistically individuated in multiple ways, both phylogenetically (in the case of the capacity) and ontogenetically (in the case of the state).Footnote ²⁷ This is yet another instance of a cognitive kind whose identity conditions are tied up with etiology and environmental context, and I would argue that this is reflected in the taxonomic practices of many of the research programs that study memory. This mode of individuation is evident in the context of computational inquiries, which attempt to understand the adaptive function of episodic memory and how it enables thinkers to plan and make decisions in their social and physical environments. It is not prevalent in neuroscientific inquiries and taxonomic categories, which are primarily interested in mechanistic causal processes that are spatially circumscribed and take place on a more limited temporal scale.

Although this account of memory is supposed to be responsive both to a range of scientific research programs as well as the vernacular concept of memory, one aspect of the vernacular concept of memory and the verb “remember” that I have not tried to address in this account is its allegedly factive nature. In English, “remember” often takes a that-clause (e.g. “I remember that I ate pizza for breakfast last Sunday”), and assertions involving this verb presuppose that the content of the that-clause is true. At least on one view of the truth-conditions of such assertions, if the proposition expressed by the that-clause is false, the assertion is neither true nor false and it cannot be asserted. Like “know,” “recognize,” “realize,” “learn,” and “report,” “remember” is supposedly only correctly applied when it is said of something true. On the theory of episodic memory presented here, factivity cannot be upheld simply because nothing about the account guarantees that episodic memories will always be true. Traces represent the past, but (as constructivists emphasize) they may be so entangled with other, non-veridical information, that the resulting content of the episodic memory may be false. Some of the results of the experiments reported in Loftus and Pickrell (Reference Loftus and Pickrell1995) may be good examples of this: The participants’ mental states may incorporate traces of past experiences as well as traces of suggestions planted by the experimenters, but the ensuing memories are false. If we are interested in giving a naturalist account of the capacity of episodic memory and the states that it generates, we cannot also insist that episodic memory states are true by definition (as factivity would have it). Nothing guarantees that a cognitive capacity that forms part of our psychological endowment will only generate truths, and a definitional insistence on this is bound to collide with empirical results from cognitive science. In addition, it is worth noting that, like the verbs “report” and “learn,” some uses of “remember” in English are not factive. The sentence, “The police reported that the culprit had fled,” has a factive reading, on which what the police said is true, but it also has a non-factive reading, according to which it need not be. Similarly, it seems that I can correctly assert (at least according to some idiolects of English), “I remember that I ate pizza for breakfast last Sunday,” even if I believe that I didn’t. Admittedly, it might be more felicitous in such cases to say something like, “I seem to remember that I ate pizza for breakfast last Sunday,” but the rephrasing is not mandatory for at least some speakers. Thus, factivity is not an obligatory feature of our commonsense concept, and I would argue, that it ought not to be a desideratum for the scientific category of episodic memory.Footnote ²⁸