Chemero (2009), Chapter 5: Guides to Discovery

The dynamical stance laid out by Chemero in the previous chapter has a potential flaw (besides being a bit weak-ass) - it's not clear how it can serve as a guide to discovery. How do you do productive science taking this approach? Chemero is going to make two suggestions, only one of which I think works: first, he's going to suggest dynamical models such as the Haken-Kelso-Bunz (HKB) model can serve to stimulate empirical work even when they are entirely phenomenological. This approach is, I think, entirely incorrect, and this chapter is full of serious problems (only some of which are unique to Chemero). Second, he's going to suggest that Gibsonian ecological psychology can actually solve the problem much more robustly anyway, by serving as an underlying theory of behaviour. This will work better, and I would advocate Bingham's model of coordination as an exemplar of this, more promising route.

But first, the HKB model as guide to discovery (this chapter is largely the material from Chemero, 2000; I intend to turn this post into a paper to rebut that paper and point to the Bingham model as an alternative, so comments are especially welcome on this one). Time to get a little critical, I'm afraid.

I've described the HKB model in detail here; it's Kelso's potential function model which describes the basic phenomena of bimanual rhythmic movement coordination. The key feature of the approach is to treat limbs engaged in coordinated actions as nonlinearly coupled oscillators, where the coupling requires energy to maintain and so would dissipate without maintenance (Kugler, Kelso & Turvey, 1980). Chemero's suggestion is that if this model can lead to useful experiments and be adapted to cope with new findings without losing this essential character, then this would demonstrate the HKB model and approach can serve as a guide to discovery for a science, specifically RECS.

Problem 1
One initial problem with this analysis crops up while Chemero is describing the HKB model. He goes over the basic features (c.f. Kelso, 1995)  and then says

Importantly, the HKB model makes a series of specific predictions. First, it predicts that as rates increase, experimental subjects will be unable to maintain out-of-phase performance. Second, even at slow rates, only relative phase of 0 [0°] and .5 [180°] will be stable. Third, the behavior should exhibit critical fluctuations: as the rate approaches the critical value, attempts to maintain out-of-phase performance will result in erratic fluctuations of relative phase. Fourth, the behavior should exhibit critical slowing down: at rates near the critical value, disruptions from out-of-phase performance should take longer to correct than at slower rates.

Chemero, 2009, p. 88

The first two are not predictions of the HKB model: they are experimental phenomena which the HKB model was built to describe. This matters, because Chemero wants the HKB to serve as a guide to discovery and counts these as evidence that it can do this. In fact, they simply reflect the phenomenological nature of the model and the modelling approach.

The next section is a series of eight cases, seven about the HKB model, which supposedly show how it can act as a guide to discovery:

Case 1: Modifying the HKB model with a noise term to fit the data
Schöner, Haken & Kelso (1986) added a noise term to the HKB data. It's not at all clear how this is supposed to count in favour of the HKB as a guide to discovery, given that noise terms are standard in models. All that happens is the model fits the data better; this is still, therefore, a development driven by the data, and not the reverse.

Case 2: Reconceptualising learning as a phase transition
Chemero then (incorrectly, as far as I can tell from reading the paper) describes the first extension of the HKB to include learning (Schöner & Kelso, 1988) as allowing the B/A ratio to vary over trials; this control parameter dictates the shape of the attractor layout, and learning then becomes a change in this layout - a phase transition. The article (and all the later empirical work with Pier Zanone) actually models learning as the imposition of a required relative phase, via a third term in the model (p. 86 of Schöner & Kelso, 1988). Learning is indeed a phase transition for the HKB approach, but one brought on by competition between the existing and target landscapes. Note that this led to various predictions which fell through when tested, and although incorrect predictions would presumably be a cause for concern in any guide to discovery, Chemero does not discuss any of this learning work.

Chemero then makes another error, and cites a paper (Amazeen, Sternad & Turvey, 1996) which he claims demonstrates a phase transition due to learning. He describes the paper as training participants to wiggle their fingers in a difficult 5:4 poly-rhythm, which then led to a reorganisation of the attractor layout such that 180° became more stable than 0°. This paper has nothing to do with learning, however. There was no training involved, and the reversal in the stability layout was actually the result of detuning (if the two oscillators have different preferred frequencies, then (roughly) the HKB layout remains the same shape but is moved along the relative phase axis to a degree proportional to the frequency difference, hence the empirical result). I have no idea what happened here, but this case is a good lesson in the importance of reading the primary literature.

Case 3: Modelling perception-action couplings
Chemero is clearly familiar with detuning, because this case is about how extending the model to include a detuning term (Kelso, DelColle & Schöner, 1990) bolsters his case for the model as guide; it can now cope with new phenomena (coordination between a person and a metronome beating with a variable frequency). Incorporating detuning was a perfectly interesting addition, but describing this as modelling 'perception-action couplings' is entirely incorrect (and note the modification is data driven again). First, even 1:1, within person coordination entails perception, so detuning cannot be the move that incorporates perception-action couplings. Second, detuning doesn't actually impact the coupling at all, except indirectly by altering the motion to be perceived; detuning does not change the basic HKB shape, which emerges from the coupling requirement. Of course, the HKB model can't tell you about this; only a fully perception-action model can. This problem doesn't stem from Chemero; it's typical in the literature to not notice that within-person coupling entails perception. In essence, this is still model fitting, not guided discovery.

Case 4: Social coupling
Schmidt, Carello & Turvey (1990) showed that the basic HKB phenomena persist when the coupling is between people; Chemero describes this as social coupling and is excited by the fact that the HKB model could be extended to not just perceptual (Case 3) but social phenomena. This is entirely the incorrect way to analyse these results; the Schmidt et al paper is, in fact, perceptual coupling. Just because there are two people involved does not change this fact. In fact, the kinematics from this experiment were used by Bingham to drive the coordination displays in his first visual perception studies, effectively using them to make a point-light display. (There is an extensive research literature on biological motion perception which uses these types of displays to explore the motion based information underpinning our knowledge of people's gender, age, psychological state, etc; Nikolaus Troje's BioMotion lab has numerous excellent references and displays). Again, only a fully perception-action model, with specific hypotheses about the information involved, could enable you to understand this fact; the HKB approach simply doesn't provide the tools.

Case 5: Asymmetries other than detuning
Treffner & Turvey (1995) expanded the model with some additional terms to account for symmetry-breaking, i.e. asymmetries between the oscillators caused by things other than detuning. They coped with handedness and attention by adding two additional sine terms to the model, with parameters which could be set to weight their contribution to fit data from handedness and attention experiments. Again, a perfectly sensible exercise in model fitting, but there is still no guide to discovery: the asymmetries were phenomena the model could not cope with, not phenomena revealed by the model, and the model fitting exercise was not constrained in the way I've suggested is critical (the added terms were simply additional sine components with their own parameters, one of which was just labelled 'attention' or 'handedness'). 

Case 6: Speech production
Port (2003), a linguist at IU, has suggested you could extend the HKB model to serve as a general model of meter in speech production. Chemero suggests that this work shows how the HKB model can be extended beyond it's original scope, a key feature of any guide to discovery. It's not clear how Port's work helps, however: his data from a simple speech task showed four attractors (at in-phase, anti-phase, and two other locations). You can indeed model this by adding an extra Fourier term to the basic HKB model, but again this is mere data fitting, and the HKB model has not served as guide to discovery other than loosely inspiring the experiments. Essentially, Port's model isn't a modified HKB, it's just an HKB-style model of a different dynamical regime.

Case 7: Cortical coordination dynamics
Kelso got quite interested in later years in describing patterns of cortical activity in terms of the HKB model. Coordination of activity between different regions seems to vary with task, most interestingly in that changing task seems to be associated with in-phase activity. Chemero is correct in noting that this work is quite useful to counter the claim that RE cognitive science ignores the brain. However, nothing in his review of this work suggests this work is anything more than an interesting way to describe brain activity. In addition, the work which has revealed the task dynamic from the HKB pattern emerges in the first place entail a specific mechanism which does not obviously generalise to cortical activity. The general principle, of modelling complex phenomena as non-linearly coupled oscillators, seems to be of preliminary use (the HKB model, for all it's limitations, has led to a lot of activity) but there's still no mechanism to constrain future work.

Case 8: Solving (representational hungry) gear problems
Chemero describes some work (e.g. Stephen et al, 2009) in which a classic problem solving task has been studying in terms of dynamics. Specifically, this work goes looking for various dynamical signatures (critical slowing down, fluctuations) in measures of problem solving behaviour. This seems to be quite interesting work, but to say it's inspired by the HKB model is misleading. While the HKB model is the most common example of this type of dynamical systems psychology, it is not the source of predictions about slowing down or fluctuations - dynamical systems theory is, and Stephan et al draw all their predictions directly from there, only citing Kelso (1995) once! Even if you allow for a more charitable reading, of research using an 'HKB-like' model, it's still not the model that's done any guiding to discovery; it's the principles of dynamical systems theory.

I was really disappointed with this chapter, and as I said I'm inclined to turn these concerns into a paper. The section on the HKB model is, as far as I'm concerned, an unnecessary tactical error. The HKB just doesn't do what Chemero needs it to (serve as guide to discovery) and this is made all the clearer by the presence of an alternative (a theory, ecological psychology). The history of physics he refers to in the previous chapter (Mach's phenomenalism vs. Boltzmann's atom theory) was decided convincingly in favour of the theory; RECS has a theory, and so it just doesn't need to try and lever anything out of the (phenomenological) HKB approach. This is especially relevant given a) the errors in Chemero's treatment of the model, which are not unique to him, and that b) it took a theory based approach to actually start to address the mechanisms underlying coordination.

To his credit, Chemero notes at the end that this type of approach won't satisfy many people, and that the better solution is, indeed, the theory based approach to science which he develops in the next two chapters. But he still saw fit to include it in the book, and so I'm assuming he still thinks this is a sensible strength of the approach he is developing. It isn't, and I think it represents a flavour of science that is causing nothing but trouble for psychology. A lack of a guiding theory in psychology has left the field chasing phenomena and unable to begin developing much in the way of a coherent account of human behaviour. Social psychologists barely speak to cognitive psychologists who know nothing about perception or motor control, and we rattle along in our separate little domains. Toy models designed to merely describe a limited set of phenomena, as the HKB is, cannot serve as guides to discovery. 

Amazeen, E. L., D. Sternad, and M. T. Turvey (1996). Predicting the nonlinear shift of stable equilibria in interlimb rhythmic coordination. Human Movement Science, 15, 521 542. DOI

Chemero, A. (2000). Anti-Representationalism and the Dynamical Stance Philosophy of Science, 67 (4) DOI: 10.1086/392858 Download

Kelso, J. A. S. (1995). Dynamic Patterns. Cambridge, Mass.: MIT Press.

Kelso, J. A. S., DelColle, J., & Schoner, G (1990). Action perception as a pattern formation process. In Attention and Performance XIII, ed. M. Jeannerod, 139-169. Hillsdale, N.J.: Erlbaum

Kugler, P. N., Kelso, J. A. S., & Turvey, M. T. (1980). Coordinative structures as dissipative structures I. Theoretical lines of convergence. In Tutorials in Motor Behavior, ed. G. E. Stelmach and J. Requin. Amsterdam: North Holland. Download

Port, R. (2003). Meter and speech. Journal of Phonetics, 31, 599-611. DOI

Schöner, G., & Kelso, J. A. S. (1988). A synergetic theory of environmentally specified and learned patterns of movement coordination. II. Component oscillator dynamics. Biological Cybernetics, 58, 81 89. DOI

Schoner, G., H. Haken, H, & Kelso, J. A. S. (1986). A stochastic theory of phase transitions in human hand movement. Biological Cybernetics, 53, 247 257. DOI Download

Schmidt, R. C., Carello, C., & Turvey, M. T. (1990). Phase transitions and critical fluctuations in the visual coordination of rhythmic movements between people. Journal of Experimental Psychology: Human Perception and Performance, 16(2),  227-247.  Download

Stephen, D.G., Dixon, J. A., & Isenhower, R. W. (2009) Dynamics of representational change: entropy, action, and cognition. Journal of Experimental Psychology: Human Perception and Performance, 36(6), 1811-1832. DOI

Treffner, P., & Turvey, M. (1995). Symmetry, broken symmetry, and handedness in bimanual coordination dynamics. Experimental Brain Research, 107, 163 178. Download

Tweet