Tuesday 22 February 2011

Chemero (2009), Chapter 2: Embodied Cognition

Chemero spent Chapter 1 creating space for himself and his book in the marketplace of ideas about how we should do our cognitive science. Chapter 2 is about situating his theory in amongst the competition, with the goal of establishing exactly what a radical embodied cognitive science (RECS) is, and that it is a perfectly respectable competitor. He reviews a little history and an example study using dynamical systems that explains a  'representationally hungry' cognitive task to set things up. This is still quite preliminary work, but time well spent I think.

A Taxonomy of Theories About Minds
Figure 1. Chemero's taxonomy
There are, broadly, two themes that emerge in theories about minds. Fodor & Pylyshyn (1988) call these 'representationalist' (those that claim cognition occurs using internal representations of the world, ourselves and our interactions with the world) and 'eliminativist' (theories that reject representations and focus on, for example, behaviour). [It's typical F&P style to define everything with reference to representations; it's not clear to me why I get described as trying to do away with something, rather than them getting described as trying to bring something (representations) in - other than historical accident, of course. But I digress.]

This cut is about right. Historically, Chemero notes, psychology has been done either by (European) structuralists (like Wundt and Titchener) or by (American) functionalists (like William James).  
  1. Structuralists focus their scientific attention on the structure of mental acts, their order and duration. This means you might compare two tasks that vary by a single stage, and measuring the time taken to do each task - the difference is the duration of the single additional stage. This is very much a description of modern cognitive psychology, and also modern cognitive neuroscience - fMRI studies are explicitly this type of contrast experiment. [This approach suffers if the stages overlap, etc, which is these days considered pretty likely.]  
  2. Functionalists insist that mental acts must be understood in context; you cannot decompose mental acts the way structuralists want because there are no abstract, task-independent components from which to build a mental act in the first place. One doesn't merely 'see a visual stimulus'; one 'sees an object in order to grasp it' or 'sees an object in order to throw it'. The process of 'seeing' is different each time, and it's just unfortunate we use the same word (instead of, for instance, the full phrases). Besides James, John Dewey's 'The Reflex Arc Concept in Psychology' (Dewey, 1896) is the best known exposition of this American brand of psychology.
Representational theories of mind therefore descend from the structuralist school. These theories have proliferated over the years, but are mostly (broadly speaking) computational - cognition is some form of rule based manipulation of some form of symbol. The key point is that organisms represent the world and act via those representations.

Eliminativist (anti-representational) theories of mind, including RECS, descend from functionalism. Dewey and James led to, respectively, behaviourism and ecological psychology (Chemero rests here on Heft's (2001) analysis showing Gibson's theories don't simply appear from nowhere but come to Gibson from James via Holt - see my previous posts on Heft's book). The challenges to representational theories from ecological psychology, combined with the 'situation semantics' of Barwise & Perry, then inspired the embodied cognition theories currently trendy in the field. [I don't know a lot about B&P other than their ideas are pretty well respected; they will come up again in this book in Chapter 6. Briefly, situation semantics is a way of getting meaning from an environment without representation or inference, and it depends on there being constraints between situations. Chemero will attempt to use this to expand the concept of ecological information.]

RECS and Dynamical Systems

There remains a problem, however; embodied cognition as currently practiced is still a representational theory of mind, specifically a computational theory. Embodied cognition attempts to 'offload' some of the cognitive work onto parts of the body and the environment, but still considers that work to be computational and the non-brain parts of the system to be representations. Chemero considers this to be a problem (and frankly, so do I). Specifically, he's identified the fact that this flavour of embodied cognition is trying to combine two incommensurable approaches to the mind and cognition by simply sexing up representations to be action-oriented by design rather than action-neutral controllers of action systems. This is Andy Clark's approach, for instance.

Enter RECS. Chemero defines radical embodied cognitive science as thoroughly functionalist, specifically
...the scientific study of perception, cognition and action as necessarily embodied phenomena, using explanatory tools that do not posit mental representations
Chemero, 2009, pp 29
Those tools are going to be dynamical systems. Chemero points out an initial concern, that dynamics is actually neutral with respect to representations (this is why many researchers are still able to talk representationally, even while trying to be embodied or more ecological). This of course cuts both ways and Chemero can rely on dynamics for his anti-representational accounts as much as Clark (or Barselou, 2008) can for their representational one. The next chapters will defend this interpretation of dynamical systems.

A second concern about using dynamical systems is that many people feel it can't account for 'real cognition'. Even Andy Clark thinks dynamics tops out eventually, and while it will get us a long way, there will still be 'representation-hungry' problems such as 'thinking about things in the absence of that thing' that will require mediation by representation. As a brief counter-example, Chemero reviews the results of a paper (Van Rooij, Bongers & Haselager, 2002) that uses a simple dynamical systems model to predict behaviour in a task where participants had to imagine using a rod to reach for a target. The data supported the (qualitative) predictions of the model, suggesting that such models can handle this type of 'representation-hungry' problem. Chemero will use this line of attack (tackling 'representationally hungry tasks' with dynamical systems) again later.

Some comments

Chemero is hitching a ride with Heft (2001) at this point. Heft established that Gibson had a respectable intellectual pedigree to James via Holt and thus ecological psychology wasn't just arriving fully formed out of left field. Chemero is proposing that RECS is the next step in this academic family tree; this seems fair to me, and a point worth repeating; I appreciated Heft's analysis and I think it helps ground ecological psychology as continuous with some respectable themes in psychology. As much as I respect Turvey and what he's done for the field, his writing is filled with esoteric ideas and terms and I know a lot of people who simply rule ecological psychology out based on how 'out there' the whole thing sounds. Turvey does this for a reason: he's being disciplined and technically precise, and it's good to have this rigour, but it's had consequences for how people see the idea of direct perception.

The family tree approach is actually a nice contribution all by itself. It exposes a serious contradiction in one of the main branches of embodied cognitive science, and it would be interesting to see if anyone comes up with a meaningful defence. My hunch is that they will simply continue to run bad studies. It is a fact, though, that the term 'embodied cognition' has been checked out of the library by numerous camps, and this has just lead to Hegelian arguments among proponents that achieve exactly nothing. This analysis clarified a few things for me in a nice straight-forward manner.

Extended cognition
Chemero also believes that RECS can be a successful theory of extended cognition. Adams & Aizawa's coupling-constitution fallacy doesn't apply, he claims, because he isn't trying to couple anything to anything else. Chemero suggests that Andy Clark, etc, leave themselves open to the CC critique because they are still representationalists. When you have an environment and a representation of an environment, 'coupling' is still a sensible way to talk about how these separable element are related. But if you instead have a non-linearly coupled animal-environment dynamical system, the elements are not separable; the system is non-decomposable into parts, and thus there can only be constitution. Unsurprisingly, Ken doesn't really buy this decomposability idea; he and Adams also think you can have an extended cognitive system composed of cognitive and non-cognitive processes, but that doesn't make the non-cognitive processes cognitive.

Andy Clark tries to bypass this problem by accepting that only certain types of couplings will work. Other authors (including a paper I just finished blind reviewing) go further, and claim that genuine systems (where the behaviour in question simply cannot occur without both internal and external resources acting as an irreducible complex system) avoids the CC fallacy. Chemero is in this latter camp; I am, frankly, entirely sympathetic and I think this is a legitimate answer to the fallacy problem; it takes a step back and reconsiders what a cognitive act actually looks like, and comes to the conclusion that it spans animal and environment. This, I think, is the right path.

That said, it is complicated, and it's not quite solved by Chemero's appeal to non-decomposability. For instance: Chemero is still open to the CC critique because even though a given dynamical system might be non-decomposable into parts while functioning, the system does need to be disassembled and reassembled over time; I'll need my hand for something else later, for instance. This process of 'soft' assembly and reassembly entails coupling, uncoupling and recoupling of the components of dynamical systems that will include animal and environment elements. Adams & Aizawa can then simply point out that some of these components are non-cognitive (because they don't meet their 'mark of the cognitive' criteria).

I think this type of objection is beatable, but it's going to require more care than Chemero gives it here. The missing piece is, I think, the notion of the task-specific device (Bingham, 1988) which I intend to blog about in some detail when I get some time.

So Chapter 2 ends with RECS identified and placed firmly in historical context. The next two chapter defend the radical part of the story (anti-representationalism) by first laying out what a representation is, and then proposing a way to avoid them being inevitable.


Barsalou, L.W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645. Download

Bingham, G.P. (1988). Task specific devices and the perceptual bottleneck. Human Movement Science, 7, 225-264. Download

Dewey, J. (1896). The reflex arc concept in psychology. Psychological Review, 3, 357-370. Download

Fodor, J., and Z. Pylyshyn (1988). Connectionism and the cognitive architecture: A critical analysis. Cognition, 28, 3-71. Download

Heft, H. (2001). Ecological Psychology in Context: James Gibson, Roger Barker and the Legacy of William James's Radical Empiricism. Mahwah, NJ: Lawrence Erlbaum. (Link to posts on this book)

van Rooij, I., R. Bongers, and W. Haselager (2002). A non representational approach to imagined action. Cognitive Science, 26, 345 375. Download

No comments:

Post a Comment