Thursday, 18 April 2013

Embodied cognition in practice - some thoughts and an open invitation

Since our Frontiers paper on embodied cognition came out, Sabrina and I have been giving talks and fielding questions in emails about what we're trying to achieve. People first ask us 'why should we do this?' and the answer, from the paper, is because it works really really well. Research in this field has produced extremely powerful explanations of behaviours with extraordinary predictive power; no more small effect sizes plus a successful research programme for as long as you want to run it! What more reason could you need?

The question we then get from people who have become interested in using what we've done is 'How can I apply this approach to my research?'. The paper itself describes four key questions that we think are critical for guiding good experimental practice in psychology and cognitive science:
  1. What is the task to be solved?
  2. What are the resources available to solve the task?
  3. How might these resources be assembled so as to actually solve the task?
  4. Does the organism actually do what your described in 3?
We think this is really just good practice. The basic question in psychology is 'why did a person behave like this rather than like that?' and answering this means figuring out what the task is, from the point of view of the person doing the behaviour. We think these questions will help keep psychologists on the right track, and not get distracted by hypothetical mental entities like theory of mind and object concepts until the evidence demands them. 

We give two examples of this method in practice (the outfielder problem and the dynamical systems model of the A-not-B error), but people are still left wondering how to apply this to their niche. We want to help here; obviously we want people to start working this way and also we want people across a variety of topics to start working this way, so we can finally start accumulating some results outside of perception and action type tasks. 

I want to do two things with this post
  1. I want to describe the mindset you'll have to get into to start doing embodied cognition in your field, which I hope might make our steps and our examples make a little more sense.
  2. I want to open an invitation to people to post questions in the comments. We're busy and so I can't promise an immediate response, but if you have a task and you want to start thinking about it from our embodied perspective, post a comment with some details and we'll see what we can do to help.

Getting yourself in the right starting place
The main thing you will have to acknowledge is that you will almost certainly have to invest some time taking a few steps back to the beginning before setting off again with your data and making progress. Our approach is meant to guide discovery and interpretation from the get-go, and you, the already active researcher, are smack in the middle of a programme of research with established goals, questions, methods and interpretations. So switching to taking embodiment seriously means pausing and re-evaluating what it is that you are doing, and, while I think it's worth the time, it is a cost, and I want to make it explicit so you aren't surprised when you don't simply get immediate results.

Second, when you step back to the start, throw out as many of your assumptions as possible, and start only with what you know for a fact, or, to put it more succinctly, start out as a behaviourist. Most of the assumptions (e.g. that language simply must require mental representations that store word meanings) are derived from the standard cognitive approach and do not necessarily have any place in an embodied analysis (remember, we advocate replacement style embodied cognition (Shapiro, 2011). Focus on the facts at hand: you are trying to explain why you got behaviour X in task Y under conditions Z. That, at the start, is all you can count on; everything else is up for grabs. This is what Esther Thelen and Linda Smith did with the A-not-B task - it took time and a lot of data collection to map out the behaviour to be explained, but throughout they focused on how the task characteristics impacted on behaviour, without ever jumping the gun and assuming that there 'just had to be' something like an object concept driving everything. 

This is a problem in cognitive psychology that we're fighting against - the assumption that for a behaviour to unfold in a particular way, there has to be a description of how that unfolding should go living somewhere in the system (in the form of a mental representation). Cognitive science should therefore be in the business of finding these descriptions. But in biological systems it's rarely (if ever) true that behaviour is simply scripted by anything in this way. In genetics, for example, the field has rapidly realised that the idea of 'a gene for X' makes no sense. X is actually always the emergent result of the interaction over developmental time of multiple connected complex systems. In the same way, the form of a behaviour does not require a 'representation for' that behaviour. Instead, the form reflects the ongoing interactions of multiple complex systems, and so whatever the brain is up to, it isn't representing. 

In effect, we are trying to avoid the psychologist's fallacy: the error of assuming that your description of the thing is, in fact, the thing itself. Skinner's behaviourism was a method designed to avoid making this error and he was not wrong to do so. This kind of discipline is essential, so in effect we advocate that some careful behaviourist-style experimenting take place before the rampant speculation about mental processes, because (as we lay out in the paper) when you do, a) you end up not needing the rampant speculation because b) it works really really well and you end up with a very powerful explanation of the task at hand.

An open invitation
It's not always easy to identify what assumptions you need to throw out when you're in the middle of a field working on a task. One way we might be able to help is to have a dialogue where you tell us what you are doing and we ask you annoying questions until we identify the core description of the task at hand and strip away the things you don't need right away. We've been doing this via email with a couple of people, but I want to open up the comments on this post so we can help out some more.

A good comment will contain a description of the specifics of the task. We can't help you explain 'consciousness' or 'language comprehension' but we might be able to help you with your specific experiment. We'll ask for clarification if we need it, and so if you comment please keep an eye on the comments for replies so we can have a bit of a conversation. We also can't promise that we can solve your problem, but we will try to ask you questions that will hopefully help you identify where to begin - frankly, you will know way more about what you do than we do so your expertise will be critical here.

If anyone else wants to jump in with ideas about where someone can go with their question, feel free. We only vouch for our answers thought :)

Shapiro, L. (2011) Embodied Cognition. NY: Routledge Press  

Wilson, A. D., & Golonka, S. (2013). Embodied Cognition is Not What you Think it is Frontiers in Psychology, 4 DOI: 10.3389/fpsyg.2013.00058


  1. This is excellent. I just gave a paper on framing in linguistics where I argued (as an aside) that cognitive psychology is in a real crisis precisely because it reaches for mentalist explanations far too early. I've been calling this the 'isomorphism fallacy' - thinking that just because you see some behavior or pattern in language, it must also exist in the mind and therefore the brain:

    But it's even worse because psychologists often don't look at anything outside the world of reified concepts. I've been working on a blog post comparing it to the drunk man looking for his keys where the light is not where he lost them.

    I'm not a practicing psychologist - just a linguist - so my question is not about a specific experiment, but I think one big area that needs a whole new experimental paradigm is memory research. In particular the interaction of working memory and long term memory. There are so many concepts bandied about that are mere suppositions based on a very limited set of experiments. One concept that I think is due for a rethink is priming. It's thought to assist things like memory retrieval but in fact relies on the retrieval (or activation) of something much more complex than the memory it's supposed to assist.

    But there are also lots of accidental memory events like the ones I described here that don't seem to fit into any paradigm.

    Maybe the problem is the word 'memory', we have it so we keep looking for it. But what if there are just memory like things. But I have no idea how to describe it in terms of tasks and resources.

  2. Dominik,
    "Memory" is a great example of an area ripe for reinterpretation. One reason is that such a wide variety of tasks are treated as identical, when an embodied approach would likely identify quite different challenges that the organism faces, and quite different resources available to support the necessary behaviors. For example, "remembering" how to tie your shoes, "remembering" how to drive home from work, "remembering" what you had for breakfast, "remembering" how to do long division, and "remembering" a favorite song, are all quite different things for an organism to do. Add on top of that our habit of bandying about the term "memory" where it does not belong --- for example saying "false memory" when referring to something that is not-a-memory --- and the list of distinct tasks is enormous.

    A task-based study of "memory" will eventually reveal, as Gibson and Skinner anticipated, most so-called memory tasks are easily explained as the result of adjustments to the perceptual-motor systems, i.e., they are completely explainable without re-peresentation. There may be some situations left over that require additional explanatory mechanisms, but these will be distinct types of situations (and see Tonneau's work for a good embodied understanding of those).

    (And see also my recent post about what we know about the brain:

    1. I agree with Eric here; the problem is the fact that 'memory' does not refer to a single thing. This is where the task analysis helps.

      First ask yourself, what exactly am I asking people to do? If it's 'remembering what you had for breakfast', call it that, not just 'remembering'. This seems trivial but its step one in avoiding the psychologist's fallacy (I'm sticking with James' name for it because he beat you to the naming rights by 100 years or so :)

      Second, the focus is on process, not objects, so 'remembering' is a better word than 'memory'. The latter implies that there is a thing called memory (a problem Sabrina nails here) while the former keeps the focus more on the behaviour to be explained.

      Then it's a matter of looking at the specific experimental setup and saying 'People behaved in this way in the presence of this information and this task demand'. Assume their behaviour emerges from this set up, rather than reflecting some underlying core competence. Then tweak each element of the set up to see what matters. At this stage there's no discussion of underlying representational mechanism, no discussion of what 'simply must' be present to solve the task. Just some careful poking around to map out the behaviour to be explained.

      It's at this point that specific details of the study are required to fill in the blanks :)

  3. I have an idea that I've been kicking around. I will think about it and post something later.

  4. There are some people thinking about the "memory" or rather, "remembering" question in a related way...Maybe this will help guide some of the discussion?


Wagman, J. B., Thomas, B. J., McBride, D. M., & Day, B. M. (2013). Perception of Maximum Reaching Height When the Means of Reaching Are No Longer in View. Ecological Psychology, 25(1), 63-80.

    *My apologies for the deleted post...I posted it using a different (old) profile

    1. Thanks all. These are great examples. But they are all quite partial. I wonder if there is an overview of "embodied" approaches to memory.

      But I completely agree about the danger in experimental psychology where a concept like working memory is postulated and tasks are invented to "reveal" its nature. Whereas in fact these are just new tasks that the subject needs to solve. They may marshall some general capacity to do that but they may do it in such a way as to completely obscure that capacity.

      I'm often thinking that we need an ethnography of memory to go alongside lab experiments. Discursive psychology seems to have explored some interesting avenues (albeit not related to memory).

      Finally, a question I never quite understood the outright dismissal of behaviorism or the benefit of postulating maximally general cognitive realities but sometime it's hard to imagine solving some problems in language processing without some concept of mental representations. I think this needs to be very cautious and have a status of a heuristic rather than a "neural" reality but still quite important. Would you reject that approach in all cases? (BTW: I know "language processing" itself is a theory laden term but I'm using it without its computational baggage as a particular perspective on "language use".)

    2. First, partial is all you're likely to see so far. And remember, these accounts are figuring out solutions to specific tasks, not whole classes of tasks. So by definition they will only be partial accounts of 'memory'. This is ok; focus on the approach taken.

      sometime it's hard to imagine solving some problems in language processing without some concept of mental representations.
      I think it's only hard to imagine because of all the years of indoctrination we put ourselves through. It took me years or working with this stuff in detail to lose the sense of 'what else could it be?' (This is why van Gelder's paper is so important, because it gives you something to answer that question).

      We don't reject the idea of internal mediation and work by the brain. We do insist that you lay out the job description facing the brain correctly in order to stand a chance of getting the brain bit right, and we think the brain stuff is often best left alone for a long time because getting the task analysis right is hard work. Does that make sense?

  5. Thanks, it makes perfect sense although does not completely satisfy my need for a broader programmatic statement which I think might be necessary to bring more people across.

    I agree on the indoctrination. I have been long advocating the "death of the system" in linguistics. And I think the constructional approaches are going in the right direction. Which is why I find the ecological approach completely natural and don't take a lot of convincing - but I think a traditionally raised cognitivist (of any discipline) will need more. But perhaps I'm trying to generalize your program too far.

    I like the term "internal mediation" - there obviously is a lot of that going on in language use. Which is why mental representation is such a useful shortcut (leading to seduction into reification). But what do you do with something as complex as language where the "job description" is probably far too multifaceted to be susceptible to gentle manipulation of conditions? Otherwise, you're in the kind of danger that was the undoing of Verbal Behavior (as well as - in my view - Government and Binding). Do you have a theory of complexity to go with embodiment?

  6. But what do you do with something as complex as language where the "job description" is probably far too multifaceted to be susceptible to gentle manipulation of conditions?
    You start carefully and take your time! :) We took a cautious first step in our Frontiers paper.

    Do you have a theory of complexity to go with embodiment?
    Non-linear dynamical systems. We have a half written paper in which we argue that this is the right toolbox but that it's important not to treat it as a theory of behaviour (see here for the basic idea).

  7. I am working on research for design educators - trying to bring embodied cognition into the space of design thinking - particularly in relation to the movement between individual and collective creativity. The enactivist framework seems really useful along with Franscesco Varela's work on ethical know-how and co-emergence. It seems to suggest the cognition is taking place in bodies that are "larger than our own". Cognition finds its place in the interactions of - rather than within - the autonomous self. What are your views here?

    1. I've always been interested in design, and I think that embodied cognition and design overlap a lot. Norman nicked the concept of affordances off Gibson, after all, although he did get them wrong (he thinks they are cognitive, mental things).

      Design is literally about how our environments shape our behaviour. I think if design people took affordances seriously and also got into perception, we could make some real progress.

      We've interacted a lot with Andrew Hinton and design people he knows on Twitter, and he's right into applying our kind of analysis to design.

  8. Keep this going please, great job!

  9. So, I started pulling together something about emotion. I'm afraid much of it is really a background about how to think about them, Kinda.

    But they do need to be analyzed more ecologically (there has always been a notion of that they have evolved, if not explicitly so, at least implicitly).

    An issue is that there is so much to emotions, that there are loads of tasks that can be analyzed, and I'm not quite... there yet. I put it on my blog, as it became quite lengthy.

  10. You and Sabrina wanna be beta readers of my book, which I expect will be under contract by the end of this week? In exchange for helping me not say stupid things about the brain, you'll learn really an awful lot about sex, and possibly also some things about how to publish and write a science book for a general audience.

    1. Sure, why not! You know where to find us :)

  11. Willemijn Schot26 July 2013 at 15:15

    I've been going back and forth between this blog and the cited papers for a while now and I'm going to try to formulate something coherent now. I'm fairly new to the embodied cognition approach so feel free to set me straight wherever.

    I'm interested in how children in primary school learn. Specifically in the context of science and technology education. The literature on this topic mostly seems concerned with identifying content of mental models (what rules/strategies do children use when they have to decide whether something will float or not, or whether a balance scale will tip to the right or to the left?) and which type of guidance (if any) should be offered to endow children with mental models that are stable over situations so that they can apply what they have learned in one context to another (transfer: for instance, balls bounce of a wall in the same way as light is reflected by a mirror). This all sounds rather not-embodied to me and I feel the questions need reformulation but I’m not sure how. Hypothesizing that children who are allowed to physically explore the domain will, when tested or interviewed on the domain, outperform children who did not get this experience seems related but not at the core of the issue.

    Any ideas?

  12. Is there not an isomorphism between representationalism and a philosophy of science based upon explanation? Should the primary task of psychology be to offer an explanation (a representational picture) of human behaviour/nature, or to facilitate the attunement of people to their ecology? What might an embodied psychology look like after it has processed philosophy of science? I think these are the types of questions addressed by Wittgenstein and Foucault to some extent.