Perception, Action & Dynamical Systems

Over Easter I visited the Center of Functionally Integrative Neuroscience at Aarhus University in Denmark, courtesy of the Interacting Minds group. I gave a talk, got the tour, and met some of the faculty and students - some interesting opportunities for future collaborations, I hope - thanks for the hospitality!

I wanted to lay out the basics of the talk I gave. I took the opportunity to present some ideas that have been developing as I work on this blog, reading Chemero and working on coordination experiments. There is a core of people in Aarhus interested in things ecological, as well as dynamical systems, so it was a good audience to try these ideas out and they seemed to go over well. This is also the sketch of a paper Sabrina and I are going to work on over the summer.

The take home message of the talk was simple - dynamical systems is the right kind of mindset for cognitive science, but it is not a theory of behaviour. Dynamics merely provides the right kind of modelling tools - the form of the model must be based on hypotheses about the specific kind of dynamical systems we are or else they are merely an exercise in data-fitting. Ecological psychology is the right theory, and the Bingham model of coordinated rhythmic movement is currently the only example of a genuinely perception-action dynamical systems model. My thoughts here are largely from my response to Chapter 4 of Chemero (on 'the dynamical stance') and Chapter 5, his initial attempt to use dynamics to serve as a guide to discovery which I think fails and which Chemero then replaces with ecological psychology. The description of Bingham's model comes from here.

I began the talk by pointing to some prominent examples of dynamical systems modelling in the literature. The examples I used were

1. Busemeyer & Townsend (1993) - this model of decision making tracks the evolution of a field over time, which takes information for and against the options as inputs and eventually allows one decision to produce sufficient activation to cross threshold;
   
2. Thelen et al (2001) - this model of the A-not-B perseveration error made by young infants pits a perceptual field (with one time scale) against a motor memory field (one a different time-scale) and sees the error emerge from this competition;
   
3. Schöner & Thelen (2006) - this field model of infant habituation modelled infant gaze behaviour in various preferential looking and habituation studies without positing sophisticated cognitive processes;
4. Fajen & Warren (2003) & Wilkie & Wann (2005) - these mass-spring models of locomotion are designed to predict trajectories as a function of the presence and absence of obstacles, goals and various optic flow features of locomotion. They are similar in design, although they make some different predictions about the specific flow variables used, etc.

Dynamical systems modelling mostly consists of taking various tools from the dynamics bag and whacking them together to capture various elements of performance. For instance, the Fajen & Warren (2003) model basically just has a goal pull on locomotion as a function of angular distance, and has obstacles push locomotion away but only when the person is close to the obstacle. Add these together, add some inertia because it takes a person time to accelerate (turn), and hey presto, locomotion trajectories!

The problem with this approach is simple: none of the components of these models actually reflect anything in particular about the specific perception-action system being modelled. They are therefore not different from any other model in psychology: just some bits of maths strung together to fit some data. The bits each do some work, but the way in which they do that work doesn't tell us anything about how the work is done by a perceiving-acting organism.

This wouldn't matter so much because models are generally quite handy, but dynamical systems models really are frequently touted as having explanatory power. Chemero tried to use the HKB model this way because that's how people like Pier Zanone and others talk about the model - the attractors supposedly represent an actual part of the system which exerts a pull on behaviour, rather than simply indexing stable states which emerge from the task dynamic. This is not only the wrong way to talk about this model, it leads to incorrect predictions. Dynamical systems provides us with a language to describe the composition of systems, and a syntax to describe the organisation of these systems, and I agree with Chemero that these are the right tools for the job. But dynamical systems is not a theory of behaviour and therefore cannot constrain the choice of elements to include in a model of the actual hypothesised mechanism. 

The Perception-Action Approach as Theory of Behaviour

As I will never get tired of telling people, psychology is in real trouble because it has no central organising theory (the way biology has evolution, for example). Without such a theory, we cannot guide discovery nor organise what we find into a coherent story. Worse, major journals will find themselves unable to reject a paper claiming to have found evidence for precognition because the author followed all the methodological rules. We need a theory: as Chemero lays out in RECS, I think that theory should use dynamical systems and be ecological psychology. 

It drives me a little crazy that psychology doesn't give Gibson the due he is owed. Over the course of about 40 years he developed an elegant and sophisticated theory of visual perception that actually does everything a theory is supposed to do. It could be entirely wrong, and still be the most scientific thing in psychology so far. (I've just read EJ Gibson's autobiography and she notes his work was acknowledged in his lifetime through the numerous awards and positions he earned, but psychology has little sense of it's own history either so this doesn't help. A rant for another day, though.) The theory claims that perception is direct, and is for the control of actions: perception must therefore be of affordances and there must be information that specifies these available to an organism. Gibson laid out methods for uncovering the affordances guiding behaviour as well as the information for those affordances, and his empirical work demonstrated how these might be used fruitfully.

In my opinion, the best (and only) exemplar of a fully dynamical systems, perception action approach is Geoff Bingham's model of coordinated rhythmic movement. I've been working with Geoff and learning from him for nearly 10 years now and I'm still coming to grips with the full, deep meaning of this modelling strategy. This, and not a toy like the HKB, should be the reference point for the literature on how to build a model. 

I've blogged in detail before on the model and the empirical strategy it embodies, and the details from my talk all came from that post. In brief, the composition and organisation of the model spans both information and action elements, and uses dynamical systems as a common language to describe all these pieces. There is no perception, and action, there is only perception-action - no dualisms here. 

Critically, the form of the model is entirely constrained by empirical data from perturbation studies run on people engaged in rhythmic movement behaviour. Behaviour is task-specific, and there are limits on how far you can disassemble a task for study before it's not the same task anymore. Perturbation studies are one of the tools from dynamical systems, and rest on the assumption that a given system uses some elements but explicitly not others. Perturbing an element will therefore only affect behaviour if that element is part of the system, and the response of the system to the perturbation provides information about how that element is used. Perturbation studies are, therefore, diagnostic of the composition and organisation of a particular dynamical system. Perturbations of limbs tell us that the overall mass-spring dynamic must be autonomous and non-linear. Perturbations of information tells us that the information for relative phase is the relative direction of motion, with the relative speed acting as a non-specific noise term. 

The resulting model instantiates three specific hypotheses derived from experiments and the ecological approach: 1) movement stability is a function of perceptual ability (Wilson et al, 2010); 2) the information for relative phase is relative direction (Wilson & Bingham, 2008); and 3) relative speed is a noise term (Snapp-Childs et al, submitted). The theory the model embodies also guided us to make correct predictions about the perceptual consequences of learning 90°, and the overall modelling strategy laid out places serious constraints on the next part of the project, modelling the learning process.

One final note that I feel obliged to mention to people these days - the goal of this research is not a model of coordinated rhythmic movement. In a very real sense, I do not care about this task, in and of itself. The task is a simple model of the kinds of complex perception-action tasks we engage in every day. It entails perception, skilled movement, coordination, and all the other aspects of the more typical tasks, but it is simple enough that we can run experiments and interpret the data in a straight-forward manner. It is also simple enough to be captured by a simple model, and thus provides the perfect (and necessary) starting point for developing the experimental and analytical tools required for this particular approach to perception, action and dynamical systems.

References
Busemeyer, J. R. & Townsend, J. T. (1993). Decision field theory - A dynamic-cognitive approach to decision making in an uncertain environment. Psychological Review, 100(3), 432-459. DOI  Download

Schöner, G., & Thelen, E (2006). Using dynamic field theory to rethink infant habituation. Pyschological Review, 113(2), 273-299. DOI  Download

Thelen, E., Schöner, G., Scheier, C., & Smith, L. B. (2001). The dynamics of embodiment: A field theory of infant perseverative reaching. Behavioral and Brain Sciences 24 (1), 1-34. Download

Fajen, B. R., & Warren, W. H. (2003). Behavioral dynamics of steering, obstacle avoidance and route selection. Journal of Experimental Psychology: Human Perception & Performance, 29(2),, 343-362. DOI  Download

Wilkie, R. M., & Wann, J. P. (2005). The role of visual and nonvisual information in the control of locomotion. Journal of Experimental Psychology: Human Perception & Performance, 31(5), 901-911. DOI  Download

Wilson, A. D., & Bingham, G. P. (2008). Identifying the information for the visual perception of relative phase Perception & Psychophysics, 70 (3), 465-476 DOI Download

Wilson, A.D., Snapp-Childs, W., & Bingham, G. P. (2010). Perceptual learning immediately yields new stable motor coordination. Journal of Experimental Psychology: Human Perception and Performance, 36 (6), 1508-1514 DOI   Download