Radical Embodied Memory (Wilford & Anderson, 2025)

Of all the representation-hungry problems out there, memory seems to be the hungriest. It is clearly a fact that we can organise our present behaviour with respect to things that happened previously; we learn, I can shape my talking behaviour to be able what I did last summer, and so on. In order for this to be possible, something has to persist from the past into the present, and it seems intuitive that what persists is some sort of record of that past event. That seems like it has to be a representation, by the very definition of the term. 

Obviously, I don't think that ecological psychology can't find a way to talk about memory. I've done it a bit (Rob and I had a chat on his podcast, I've thought about it a little in papers) but never anything formal; all just thinking through the problem. So I was excited to see a preprint on the topic by Robyn Wilford and Michael Anderson. In this post I want to review the basic approach (overall I like the paper and I think the basic idea is exactly right) and list a few suggestions I would make if I was reviewing this paper. It's exciting to see this topic getting some attention, even if it is still very early days.

We can organise our present behaviour with respect to things that happened in the past and are not currently happening. Turvey (2019) gives this capacity the general name 'retrospectivity'. But how is this implemented? 

The first assumption is that something the organism has access to must remain stable over time, and that this thing must in some way and to some degree be about that past event. The first thing this paper does is pay the ecological tax and explain the computational account. In this approach, the past event must be encoded, that encoding must be stored, and then that encoding must be retrieved from storage and brought into the present. The stable thing an organism has access to is the stored encoding of the event; psychology therefore should study encoding and retrieving methods and formats (so you get things like Craik & Lockhart (1972) levels of processing frameworks). 

Obviously ecological psychology can't use any of this; it's all just the wrong kind of conceptual tool. So what could possibly remain stable and accessible, if not an encoding of the event in memory storage? Enter Anderson's Transiently Assembled Local Neural Subsystems (TALoNS; Anderson, 2014; see this blog post). TALoNS propose that different parts of the CNS get recruited, used and then reused by varieties of tasks. The neural contribution to a given task is not the activity of a certain location in the brain, but the stable relations formed between various parts. This pattern of interaction can be implemented by a variety of neural components (there is redundancy and flexibility) and so it's not the parts that persist, but the functional relations. Wilford and Anderson propose that the same principle is at work and could ecologically enable retrospectivity. 

In this analysis, 'memory' is what happens when the brain-body-environment system is brought back together to form the same relational form as was around when the past event happened in the first place. It doesn't have to made of exactly the same pieces (and in fact it can't be; at least some things will be missing in the past for anything that can be described as a memory task) and it doesn't have to be exactly the same relation; just the same kind. This is analogous to synergies and redundancy at the level of behaviours. There's many ways to implement 'hitting a target' (by punching, or kicking, or throwing); what remains stable is that various parts are being brought into a certain kind of relation that satisfies the description 'hitting a target'. 

But why does these patterns of interactions remain options? What kind of stability do they have that makes them capable of implementing retrospectivity? Are they stored somewhere, and if so, are we just back in encoding territory? Wilford & Anderson instead describe a kind of dynamical stability, in which systems have state spaces in which all the potential states of a system are described. Those states don't need to be encoded or stored; they are instead possible states of this particular dynamical system and they 'persist' in the dynamics, even if they aren't currently the state of the system.

Figure 1 from Wilford & Anderson

Wilford & Anderson use water as an example. Water doesn't become ice because it remembers how to become ice; it becomes ice when conditions (specifically the pressure and temperature) take water back into that region of the state space. Ice is always a possible state of water, and this fact is not stored or encoded; it's a possibility defined by patterns of relations between pressure and water. Wilford & Anderson propose we think of human memory in the same way; remembering is not retrieving an encoding from storage, but it is the reestablishment of a pattern of relations in the body-environment-brain system to one of the states that is an option for those system dynamics. Learning at the time is about altering those dynamics so that the desired pattern of relations is an option. I have previously phrased this idea like this (Charles et al, 2014):

I don't think it's meaningful to say I currently (sitting at my desk) have the ability to hit a softball; that ability isn't stored somewhere for me to access on demand. I do think it's closer to say I have the ability to become something that can hit a softball

This leads to the question of 'how to move the system through it's state space to that possible state'. Wilford & Anderson discuss the rule of the cue in memory, and the fact that how you prompt memory seems to be part of the memory that is prompted (cf Tulving, and also I think maybe Bartlett). I think this is probably pretty crucial, but there isn't enough detail here yet; more on this when I 'review' the paper below. I've phrased it this way previously on the blog:

When you're going out to bat, you pick up the bat, and stand 'on deck' waiting your turn. In between pitches, we all swung our bats, warming up. When it was my turn, I'd get my feet planted and stand by the plate; I'd touch the far corner of the plate with the bat to get a sense of where the strike zone was and swing the bat a few more times from that stance. I'd also often let the first pitch go by, just to watch it. All of this was on the advice of the better players in our team.

What I was doing was generating information that was guiding the formation of a softball-hitting device. Swinging the bat generates information about it's inertial characteristics, which is the basis of dynamic touch (discussed at the end of this post on Chemero's book). I now have information about how the bat is changing my upper limb functionality, as well as information about what's required to move the bat (how long it takes, etc). Repeating this at the plate continues this flow of information in a more specific setting, driving the continued formation of the right kind of device. Watching the first pitch go by was also a way of sampling some task relevant information without yet trying to interact with it. My claim is that this is all happening in the moment; until I pick up a bat, I am not currently capable of swinging a bat. The job of the nervous system is to enable the formation of that device once I begin the process, and the specific device I form is the result of the structure of the flow of information and the kind of nervous system and body it's flowing into.

Wilford & Anderson then proposes some reinterpretations of existing memory work within this radical reframing, and note some empirical opportunities. I won't cover these here in detail, as I think this all needs some work. 

Some Peer Review

If I was reviewing this paper, I would be generally positive about the idea but have some suggestions about the framing. I thought I'd take the opportunity here to note these, given they were kind enough to preprint the work.

I think the initial section about the encoding/storage/retrieval can be made much shorter; there's a bunch of discussion about the notion of the stimulus that seemed out of place. Just set up the vocabulary, note that there are issues with it, and get to the good stuff.

The next section lays out the dynamical state space approach. I think this is mostly good. I would suggest explicitly framing it as the necessary preliminary reframing required to get an ecological approach off the ground. I would also suggest fronting up to the fact that while it might be conceptually possible to define the state space of a brain-body-environment dynamic and look for memory there, actually doing so is a very non-trivial task. That's why this work is good conceptual reframing, and not yet the actual research programme it is kind of claiming to be. 

Because this idea is not yet an actual practical research programme, I found the framing of the next section very ambitious. It was pitched as 'how to reinterpret key memory findings' and the sections read as if the dynamical account was already established. What the authors have actually done is present some examples of how to go about engaging with these established findings from their dynamical perspective; how to implement the reframing needed before the ecological work can get going. Frame it explicitly as some example reformulations, because these are not yet viable alternative explanations. 

I think the role of the cue is actually pretty important, and that you need to spend more time on how you might get the system dynamics into that previous configuration. The dynamics story replaces encoding and storage, but you need to talk about how to replace retrieval. 

Same basic idea with the final section about practical consequences; it is, at this point, quite light on specifics. Perhaps spend some time working a specific example to more clearly demonstrate what you mean by changing 'where you should exert experimental control'. 

Overall, the proposal is a sensible first draft at reformulating what retrospectivity could be, if not encoding. It is not yet a viable alternative research programme, and the writing slips into treating it as if it is sometimes. It's ok that it's just the reformulation - this has to happen first! Just embrace where you are at in the programme.