There is no such thing as a memory trace

Stephan Braude on trace theories of memory

I remember myself sitting on a balcony in my teens, one autumn night, peering through a telescope to find the North America Nebula in the constellation Cygnus, while “I like to be in America” from West Side Story was playing on the radio. It’s easy to see that the synchronism of looking for the North America Nebula and listening to “I like to be in America” makes this a particularly memorable episode. But how did I manage to form and retrieve this memory?

North America and Pelican Nebulae. Own work by Wesley Chang

According to C.B. Martin and Max Deutscher,¹ this problem “can be solved by recourse to the idea of a memory trace,” which is “an indispensable part of our idea of memory.” Since many theorists agree, the existence of memory traces is rarely debated. Discussion centers on their character, physiological embodiment, and the like. People have relied on an explanation as simple as that of a print of a coin in wax (Plato), or as simplistic as that of a particle track in a piece of mica (Sabine Hossenfelder²). They have used examples such as the structural analogy between music and the groove in a gramophone record (Wittgenstein). Or they have invoked the “principle of isomorphism" regarded as a “psychophysical axiom,” which posits that “our experiences and the processes which underlie them have the same structure” (Wolfgang Köhler³). Yet others have likened memory traces to tapes and films “on which are registered all of those things of which the individual was once aware” (Wilder Penfield⁴).

What all these views have in common is the notion that explanations of memory require the postulation of copies of whatever it is that is remembered, and such copies are typically thought to be stored structural analogues of earlier experiences. All versions of the memory trace theory moreover require that traces, in some sense or other, represent past states of affairs. What is conspicuous by its absence is any independent psychological or physiological evidence of the existence of memory traces. The latter are typically regarded as items which simply must exist.

The trace theory thus takes for granted that the world has some kind of intrinsic structure, and that whatever holds the memory in some sense mirrors this structure. But to attribute a structure to a state of affairs A is to say inter alia that A consists of parts, and that these parts are configured in a certain way. One challenge for the trace theory therefore is to make sense of the notion of states of affairs having definite, nonconventional, natural parts or elements. As any mother who has baked a pie for her family but then made it do for unexpected guests as well, any single whole whatsoever may be divided into any number of parts chosen at will.

What I have reported so far has been gleaned largely from an outstanding paper by John Heil.⁵ In what follows I shall quote liberally from a more recent paper by Stephen Braude,⁶ whom I had the pleasure to introduce in my last post. On the issue of intrinsic parts, Braude has this to say:

unfortunately for trace theory, objects and events can always be parsed in an indefinite number of ways, and whatever parsing we select can only be conditionally, and never categorically or intrinsically, appropriate. We always determine a thing’s components relative to a background against which certain features of the things (but not others) count as relevant. But then it’s only against shifting and nonprivileged background criteria of relevance that we take two things to have the same structure; they’re never isomorphic simpliciter— that is, intrinsically or inherently.

So the trace theorist’s inevitable appeal to privileged, inherent structures and intrinsic mappings is literally absurd. It’s on a par with claiming that a pie has a basic or privileged division into slices or elements—that is, a context-independent answer to the question “How many pieces are there to this (unsliced) pie?” as if the number of potential pie eaters were irrelevant to our answer. Similarly, it’s as absurd as claiming that there’s an absolutely context-independently correct and privileged answer to the questions “How many events were there in World War II?” and “How many things are in this room?” [11–12]

It is generally assumed that memories are stored in brains, but Braude’s arguments are of a more general nature:

the problems with the concept of a memory trace are hardware independent. It doesn’t matter whether traces are conceived as mental or physical, or more specifically as static, dynamic, neurological, biochemical, atomic, subatomic, holographic (à la Pribram), nonspatial mental images, or (as Plato suggested) impressions in wax.... Memory-trace theory requires them to perform functions that nothing can fulfill. [3]

All the same, Braude restricts his attention to humans, in which case

the capacity to remember is causally dependent not simply on having a functioning brain, but probably also on changes to specific areas of the brain. However, it’s one thing to say that the brain mediates the capacity to remember and another to say it stores memories. [4]

Braude begins his critique of trace theory with an analogy drawn from Heil’s paper:

Suppose I invite many guests to a party, and suppose I want to remember all the people who attended. Accordingly, I ask each guest to leave behind something (a trace) by which I can remember them. Let’s suppose each guest leaves behind a tennis ball. Clearly, I can’t use the balls to accomplish the task of remembering my party guests. For my strategy to work, the guests must deposit something reliably and specifically linked to them, and the balls obviously aren’t differentiated and unambiguous enough to establish a link only with the person who left it.

So perhaps it would help if each guest signed his or her own tennis ball or perhaps left a photo of himself or herself stuck to the ball. Unfortunately, this threatens an endless regress of strategies for remembering who attended my party. Nothing reliably (much less uniquely and unambiguously) links the signature or photo to the guest who attended. A guest could mischievously have signed someone else’s name or left behind a photo of another person. Or maybe the signature was illegible (most are), or perhaps the only photo available was of the person twenty-five years earlier....

But now it looks like I need to remember in order to remember.... What the situation requires is an unambiguous representational calling card, and the tennis ball clearly doesn’t do the job. So we supposed that something else might make the tennis ball a more specific link—a signature or a photo. That is, we tried to employ a secondary memory mechanism (trace) so that I could remember what the original trace (the tennis ball) was a trace of. But the signature and photo are equally inadequate. They, too, can’t be linked unambiguously to a specific individual.... So ... it looks like yet another memory mechanism will be required for me to remember who left behind (say) the illegible or phony signature or the fuzzy photo. And off we go on a regress of memory processes. It seems that no matter what my party guests leave behind, nothing can be linked only to the guest who left it. We’ll always need something else, some other mechanism, for making the connection between the thing left behind and the individual who left it.

In fact, it seems that the only way to stop the regress is for a guest to leave behind something that is intrinsically and exclusively linked to only one individual. That’s why Wolfgang Köhler, for example, proposed that traces must be isomorphic with (i.e., inherently and structurally similar to) the things of which they’re traces—that is, the things they represent. But what Köhler and others have failed to grasp is that this kind of intrinsic connection is impossible, because nothing ... can represent unambiguously (or represent one and only one thing); representing is not something objects can do all by themselves; and representation can’t be an intrinsic or inherent relation between the thing represented and the thing that represents it. [4–5]

If we still go along with assuming the existence of a trace, we are faced not only with the problem of memory storage (the creation of a trace) but also with that of memory retrieval (the activation of a trace): how can something present trigger the memory of something past?

[I]f we follow Köhler’s lead, then we have to assert some kind of intrinsic similarity or resemblance, some kind of psychophysical structural isomorphism, between three—potentially quite different—things: the original experience or event, the trace produced on that occasion, and the subsequent triggering events. [7]

However, what triggers a memory (or activates a trace) can be quite different from what established it in the first place. What made me remember that episode on the balcony wasn’t anything like that episode. Trace theory remains completely mute on such questions as: how did reading Braude’s piece on memory traces or racking my brain for some interesting memory elicit that particular memory?

[I]n trace theories, the appeal to structural isomorphism is really just the appeal to an inherent similarity between two things, determined solely by their respective structures. It’s merely a kind of copying, and perhaps if trace theorists spoke only of copying rather than isomorphism, their theories would appear as silly as they really are. That’s why nobody takes seriously the theoretically identical position that Plato proposed—namely, that memories are like impressions in wax. It sounds much more impressive to speak instead of isomorphism, and it’s also much more effective than speaking of wax impressions, because it drives the confusions and theoretical silliness underground. [8]

Inherent similarity is a static (i.e., unchanging) relation obtaining between similar things. And it must hold between such things no matter what. But, as we have already seen in the second part of my previous post, context matters. Not only that; it’s actually indispensable. The reason intrinsic similarity is nonsense is that things are never similar solely in virtue of static relations between them or in virtue of properties inherent in them.

Things must count or be taken as similar or dissimilar relative to some context of inquiry and criteria of relevance. For example, the movements of an elephant aren’t inherently similar or dissimilar to those of a flea. They might count as similar in a situation where the size of an organism isn’t relevant but dissimilar in a context where size is a major concern.... Although cognitive scientists and memory theorists seem to ignore it all the time, the point is painfully clear: similarity exists only with respect to variable and shifting criteria of relevance. It can only be a dynamic relation holding between things at a time and within a context of needs and interests. [8-9]

[U]nfortunately for trace theory, objects and events can always be parsed in an indefinite number of ways, and whatever parsing we select can only be conditionally, and never categorically or intrinsically, appropriate. We always determine a thing’s components relative to a background against which certain features of the things (but not others) count as relevant. But then it’s only against shifting and nonprivileged background criteria of relevance that we take two things to have the same structure; they’re never isomorphic simpliciter—that is, intrinsically or inherently.

So the trace theorist’s inevitable appeal to privileged, inherent structures and intrinsic mappings is literally absurd. It’s on a par with claiming that a pie has a basic or privileged division into slices or elements—that is, a context-independent answer to the question “How many pieces are there to this (unsliced) pie?” as if the number of potential pie eaters were irrelevant to our answer. [11–12]

Besides, we do not only remember past experiences; we also remember birthdays, the time and date of a forthcoming appointment, that the whale is a mammal, the meaning of a word or phrase, and so forth. So even if trace theory were intelligible, it wouldn’t be a theory about memory generally.

Furthermore, if (like me) you believe that the meanings of sentences or words aren’t things that have a structure, something whose parts and relations between them can correspond to another structure in the brain (or somewhere else), a trace theory of memory can’t appeal to a system of representations and structural isomorphism or similarity to explain how one remembers the meanings of words. [7]

Trace theory is a specimen of the general view that particular mental states are caused by (or are identical to) certain distinct brain states, and that what those different brain states are is explainable wholly in terms of their distinctive structural features. Furthermore, since some if not most mental states are intentional in that they concern external states of affairs, on this view some of those brain states represent external states of affairs. This, according to Braude, signals a “pervasive problem in the so-called cognitive sciences—namely, confusions about and equivocations on the term ‘representation’.” [12] These confusions and equivocations stem from “the utterly false assumption that a thing’s representational properties can be determined solely by its structural or topological features.” [13]

Not only is it the case that “anything can represent anything” — for which Braude offers a wide array of examples, both in this paper and in the two books from which I quoted in my previous post — but also

what a thing represents depends ultimately on the way we place it in a situation, against an enormously rich background of needs and interests and both local and global traditions and assumptions about the way the world is and which things matter. There are no purely structural or context-independent forms of representation or meaning. [13–14]

So much for that. But let’s once more play along with the trace theorists and grant that what links together traces both with their causes and their activators is a common underlying structure. Braude next addresses the question of what sort of thing this structure is. Being also a professional pianist and composer, he remembers Beethoven’s Fifth Symphony (abbreviated as B5).

Modern versions of trace theory require that my memory be explained in terms of a representation of B5, stored in some concrete physical form in my brain and (let’s say) produced in me by an experience of hearing B5 in the past. This trace must have certain structural or topological properties that link it both to the thing(s) that caused it and also to those which later activate or trigger it. These must also be properties that distinguish the trace of B5 from traces of other pieces of music. So presumably, this B5 trace was produced by and captures specific features of a performance I heard of B5 which will enable activation of the B5 trace in the presence of subsequent items sharing those specific features.

But which features might these be?... I (like many others) can remember B5 by recognizing a wide variety of musical performances as instances (or as philosophers would put it, tokens) of B5. And these tokens can differ from one another and from the original trace-producing instance of B5, with respect to any features of that original event. Even wild parodies of B5 are instances or tokens of B5. That’s why I can tell what they’re parodies of—that is, that they’re B5 parodies....

So what is it that the B5 trace has in common with the concrete events that can cause and trigger it?... Whereas the remembered and triggering events are (typically) concrete instances of B5 (e.g., performances of one kind or another), and whereas the trace itself is also a specific, concrete thing—that is, some kind of persisting modification of the person (e.g., a brain state), this common unifying element must be a relentlessly abstract object—what philosophers call a type. Moreover, since memory-triggering instances of B5 can differ from the original trace-producing event with respect to any of the original event’s features (e.g., rhythm, dynamics, timbre, tempo, pitch, absence or presence of embellishments, etc.), and since the B5 trace is presumably an even more radically different kind of version of B5 (say, a neurological version that itself has no pitch, tempo, timbre, etc.), the structure that they allegedly share must be so abstract that it contains none of the concrete musical features found in the events that produced it (e.g., precise rhythm, pitch, etc.). In fact, it can’t have specific features found in any possible version or embodiment of B5. [15–16]

Trace theorists thus have no choice but to posit a common abstract type (a B5 type) linking the indefinitely many and different possible B5 event tokens.

[W]e’ve arrived at the point where we see the ultimately nonscientific nature of trace theory. It’s committed to the view that a memory trace of B5 and all concrete instances of B5 have a structure that is essential to all things that are instances of B5 but none of the specific features that real, concrete versions of B5, including the trace itself and nightmare versions, can lack. This position is commonly called Platonic essentialism—the view that things are of the same kind in virtue of sharing a common underlying but abstract structure. And that’s not a scientific view at all. It’s a philosophical view, and a bad one at that. [18]

Memory research has progressed considerably since the days of Köhler, who identified both the essential characteristics of trace theory and (unwittingly) its fatal flaws.

[T]here’s been undeniable progress in thinking about the domain of memory—for example, taxonomic advances in identifying the varieties or types of memory. Likewise, increasingly advanced technology has enabled researchers to probe our neurophysiological systems in unprecedented detail. Nevertheless, in a crucial respect, recent memory research shows no progress at all and remains defiantly superficial. It takes for granted that some form of storage and retrieval takes place in the brain, whether the putative physiological mechanism is a unitary engram or something distributed or diffuse—say, across a cell assembly, and whether what’s stored in the brain is static or dynamic. But it never addresses the fundamental issues of how any physical modification can represent or stand for what is remembered and indeed how it can represent or stand for one thing rather than another. And trace talk is alive and well....

In recent work it’s simply taken for granted that information is stored somehow in the brain, as if the matter were settled a long time ago and all that we need to do now is to figure out what the correct hardware description of the process is. So that fundamental assumption is never defended or scrutinized, and the problems with it and with its associated reliance on the notion of representation apparently go unrecognized. [18–19]

In the introductory essay to a “comprehensive” four-volume set on learning and memory,⁷ R. Menzel writes: “The notion of a physical memory trace, independent of its use ... is a central presumption in neuroscience.” And again: “A neuroscientist cannot help but assume that the knowledge stored in memory continues to exist during time periods when it is not retrieved.” Q.E.D.

Thus, memory trace theory is “deep nonsense” as defined by Braude:

although the world isn’t suffering a shortage of stupidity, not all nonsense is stupid. In fact, the most interesting nonsense is deep nonsense, and it’s something which can all too easily deceive even very smart people. That’s because the problematic assumptions are buried well below the surface and require major excavation. [4]

In another chapter of Crimes of Reason⁸ Braude states:

Over the centuries, philosophers and psychologists have explained behavior in terms of causes of one sort or another.... But behind this apparent diversity of opinion is a significant point of agreement— namely, that the proper way to explain behavior (or render it intelligible) is to trace it back to an initiating cause (a state or set of states) inside the agent. Let’s call this class of theories inner-cause theories (ICTs). [49]

The relevant defect with ICTs is not that they’re physicalistic; it’s that they’re mechanistic. That is, they try to explain a system or structure’s function (in this case, a person’s behavior) entirely in terms of the operations, interactions, and organization of its component parts. The mechanist mistake, then, is to think we can explain behavior in terms of a causal chain leading from any kind of internal initiating state to the final emitting of the behavior. But in that case, …the triggering or initiating states could just as well be mental as physical. [50]

So, it’s not just physicalistic explanations of behavior that don’t work; neither do mechanistic ones. I’ll have more to say on ICTs in another post.

1. C.B. Martin and M. Deutscher, Remembering, The Philosophical Review 75 (2), 161–96 (1966).

2. S. Hossenfelder, Existential physics: a scientist’s guide to life’s biggest questions (Viking, 2022).

3. W. Köhler, Gestalt Psychology, p. 344 (Liveright, 1947).

4. W. Penfield, The Permanent Record of the Stream of Consciousness, Acta Psychologica 11, 47–69 (1955).

5. J. Heil, Traces of Things Past, Philosophy of Science 45 (1), 60–72 (1978).

6. S.E. Braude, Memory Without a Trace, European Journal of Parapsychology 21 (2) Special Issue, 182–202 (2006). Quotations and page numbers are from the revised version published in S.E. Braude, Crimes of reason: on mind, nature, and the paranormal (Rowman & Littlefield, 2014).

7. J.H. Byrne (Ed.), Learning and memory: A comprehensive reference (Academic, 2008).

8. S.E. Braude, Defense of Folk Psychology: Inner Causes versus Action Spaces, in Crimes of Reason, pp. 49–79.