Thursday 23 June 2016

Ecological Mechanisms and Models of Mechanisms (#MechanismWeek 4)

Mechanistic models are great, but so far cognitive science doesn't have any. We have functional models (of, for example, memory or categorisation) and dynamical models (of, for example, neural networks) but none of these can support the kind of explanations mechanistic models can. Is that it for psychology, or can we do better?

Here we propose that it's possible to do psychology in a way that allows for the development of explanatory, mechanistic models. The trick, as we have discussed, is to identify the correct level of analysis at which to ground those models. These models will definitely end up being multi-level (Craver, 2007), but the form of these final models will be dictated and constrained by the nature of the real parts and operations at the grounding level.

The correct level of analysis, we propose, is the ecological level. Specifically, ecological information is going to be the real component whose nature will place the necessary constraints on both our empirical investigations of psychological mechanisms as well as the mechanistic models we develop.

Let's see how this might work.
Our goal is to explain behaviour with explicit reference to the mechanisms that make our behaviour look the way that it does. Our hypothesis is that a mechanistic analysis of behaviour is possible if we ground the analysis in ecological information. A sketch of what a behaviour is reveals why information is the central player.
  1. Animals organize their behavior with respect to properties or events in the world with which they are not in mechanical contact. Tasks are mostly happening ‘over there’, and yet our behavior is shaped by those task dynamics. This is the context in which behavioural mechanisms have to work.
  2. Animals are only linked to those distant dynamics via structured energy arrays. Task dynamics causally interact with energy media and lawfully project themselves into those media as specifying kinematic patterns. These are all the organism has access to, and therefore behavior critically involves information; our models of these mechanisms should too. 
  3. These kinematic patterns can be used to shape behavior with respect to the distant dynamics because they are specific to those dynamics. However, information is a projection of those dynamics (possibly even a representation of them) and not a copy, and the difference has functional consequences. 
This is the point at which information is revealed as the right grounding level. It is the point at which at which the organism's contact with the dynamics of the world begins, and the form of information is not identical to that of the world. Therefore, the form of our response to the world (our behaviour) is constrained by the form of the real component information that we detect, and not the world per se. More on this below.
  1. Information is a real component that is created by the causal, lawful interactions of task dynamics with energy media. A given information variable therefore only exists when a given task dynamic is present. Information is task specific. 
  2. Functional behavior is produced when a behavioural system’s dynamical properties complement the task dynamics, and task dynamics are specified to the organism by task specific information. Behavioural systems are therefore task-specific; they are task-specific devices (Bingham, 1988). 
  3. Task specific devices are assembled, parameterized, maintained and controlled by information. Assembling and parameterizing a TSD is action selection, a process which can only be meaningfully constrained by the perceived task. Maintaining and controlling a TSD is action control, the real-time work of resisting perturbations and continuing to produce stable, functional behavior. All this activity must happen with reference to the task dynamics at hand, and therefore must be driven by information about those task dynamics,
We now end up with two categories of behaviour, both constrained and shaped by information;
  1. Action selection can be supported by law-based or convention-based information use (Golonka, 2015). Because action selection does not have to track the real time task dynamics, it can be shaped by both law-based and convention-based information use. 
  2. Action control can only be supported by law-based information use. Controlling an action means working to resist perturbations and noise so that the action remains functional. In other words, action control is about coordinating with the task dynamics in real time, and therefore action control is, by definition, a process of using information in a law-based manner, treating that information as meaning the relevant dynamics. 
An Ecological Mechanistic Model
Mechanisms are the reason why a given system behaves in a particular way. Mechanistic models should therefore account for that behaviour with reference to the mechanism. There is only one psychological model we know of that fits this bill, and that is Geoff Bingham's perception-action model of coordinated rhythmic movement (Bingham, 2001, 2004a, b; Snapp-Childs, Wilson & Bingham, 2011).

This post on the model details the research programme that led to the development of the model. When you read that now, it should look an awful lot like the kind of research programme described by Bechtel & Abrahamsen (2010). Specifically, before the model was built, Bingham and others went though an extended period of empirical investigation of the actual mechanism, with the specific intent of figuring out the composition and organisation of the mechanism. Geoff's unique and critical contribution was to focus on identifying the real informational component (Andrew Note: work I was privileged to contribute to for my PhD).

The model takes the form
These are two phase driven oscillators (models of the kind of dynamical system rhythmically moving limbs are) coupled via perceived relative phase, the information for which is relative direction of motion (Ρ). This describes the state of the local optic flow field being structured by the moving limbs and at any given moment is +1 (all common motion) or -1 (all relative motion).

The model contains only terms that refer specifically to empirically identified and characterised real components. It accounts for all the key characteristics of coordinated rhythmic movement and makes a variety of novel predictions, all of which have been confirmed; specifically, that movement stability is a function of perceptual stability (Wilson, Snapp-Childs & Bingham, 2010), that relative direction is the information (Wilson & Bingham, 2008) and that relative speed acts as a noise term (Snapp-Childs et al, 2011). Because it describes a mechanism, it has also motivated successful studies on learning without actually yet modelling that process (e.g. Wilson, Snapp-Childs, Coats & Bingham, 2010; Snapp-Childs, Wilson & Bingham, 2015); recall that this is where the HKB model fell down.

The success of the model is due to the fact that it is grounded at the ecological, informational level. Geoff was inspired to pursue this topic by results showing that the characteristics of coordinated rhythmic movement persist between people and you don't even need to be moving yourself, just judging displays, for the HKB pattern to show up. Changing the information with various feedback manipulations alters movement stability to track the information, and transfer of learning is determined entirely by informational overlap between tasks (Snapp-Childs et al, 2015). This means that the critical work of shaping the behaviour was being done by the information supporting the perception of relative phase (not relative phase itself, i.e. not the dynamical property).

Bingham's model is a successful, explanatory mechanistic model of a behaviour, and it stands as an example that such things are possible in psychology if you begin the analysis at the correct level - that of ecological information.

Bechtel, W., & Abrahamsen, A. (2010). Dynamic mechanistic explanation: Computational modeling of circadian rhythms as an exemplar for cognitive science. Studies in History and Philosophy of Science Part A, 41(3), 321-333.

Bingham, G.P. (1988). Task specific devices and the perceptual bottleneck. Human Movement Science, 7, 5-264.

Bingham, G.P. (2001). A perceptually driven dynamical model of rhythmic limb movement and bimanual coordination. Proceedings of the 23rd Annual Conference of the Cognitive Science Society, (pp. 75-79). Hillsdale, N.J., LEA Publishers. 

Bingham, G.P. (2004). A perceptually driven dynamical model of bimanual rhythmic movement (and phase perception). Ecological Psychology, 16(1),45-53.

Bingham, G.P. (2004). Another timing variable composed of state variables: Phase perception and phase driven oscillators. In H. Hecht & G.J.P. Savelsbergh (Eds.) Theories of Time-to-Contact. Boston: MIT Press.

Craver, C. F. (2007). Explaining the brain. Oxford University Press. 

Golonka, S. (2015). Laws and conventions in language related-behaviours. Ecological Psychology, 27(3), 236-250.

Snapp-Childs, W., Wilson, A. D., & Bingham, G. P. (2011). The stability of rhythmic movement coordination depends on relative speed: The Bingham model supported. Experimental Brain Research, 215, 89-100.

Snapp-Childs, W., Wilson, A. D. & Bingham, G. P. (2015). Transfer of learning between unimanual and bimanual rhythmic movement coordination: Transfer is a function of the task dynamic. Experimental Brain Research 233(7), 2225-38.

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

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.

Wilson, A. D., Snapp-Childs, W., Coates, R., & Bingham, G. P. (2010). Learning a coordinated rhythmic movement with task-appropriate coordination feedback. Experimental Brain Research, 205(4), 513-520.

No comments:

Post a Comment