Task Dynamics And The Information They Create

Over the next weeks I want to turn my attention to a detailed account of the process by which you go about studying affordances (formalised as task dynamics) and the perception of affordances (via the kinematic consequences of those task dynamics) using throwing for maximum distances and for accuracy as the task. This post will introduce the basic research programme. Future posts will work through papers from my colleagues Qin Zhu & Geoff Bingham in order (I've done a couple already), as well as work from the animal literature because I want to find ways to use the analyses we're developing to answer questions about throwing and weight perception there.

These posts will do a few things. First, it's important to be as clear as possible about what affordances are, how we might possibly perceive them and how we can do the relevant science within the ecological approach to answer those two questions. Sabrina is developing ways to apply these methodological principles to the study of language, and we have both been working on the issue of information and how it comes to have meaning for us. Being clear about how this all unfolds in the perception-action literature is vital, because this is the foundation for what comes next. Second, I'm working on some throwing data right now and I need to work through the key papers in detail anyway. Third, I'm going to be developing an undergraduate perception-action class for 2014, and this will help me develop course material by laying out the form of the analysis and getting feedback on how well it's coming across. One of my goals is to look at all my collated and edited notes and realise I've accidentally written a text book :)

I'm going to talk about throwing because it's utterly fascinating. It's a complex task but it's one centred around a core dynamic (that of projectile motion) that physics has a pretty good handle on. This is letting us run detailed simulations of the task to identify the affordance structure of the task and see how throwers are operating with respect to those. Throwing entails perception of object and target affordances and the coordination of multiple body segments into precisely timed actions controlled by that perception. It also connects to all kinds of things in our evolutionary history (including, possibly, the origins of spoken language in the form we know) and our psychology (including the size-weight illusion and issues of the psychologist's fallacy). It's close to being that grail of psychology, something only humans do (other animals throw but rarely if ever for the kinds of distances and accuracy we can manage with ease). And most of all, it is endlessly interesting. The deeper I get into this, the cooler it gets. 

Task specific, smart solutions
The ecological approach to perception and action studies tasks one at a time. Why? Because it proposes that the perception-action system solves things one at a time as well. We produce functional behaviour by temporarily assembling ourselves into smart solutions to the problems we face (Runeson, 1977). Smart solutions differ from general purpose rote solutions the way heuristics differ from algorithms; they take advantage of things which are locally true and use those as reliable shortcuts to solve particular problems, instead of trying to come up with a longer, more complex solution that can be successfully applied to multiple situations. 

Why do we think perception-action systems are smart? Two reasons: first, for tasks you do often that contain reliable shortcuts, smart solutions are faster, more efficient and more stable than rote solutions. These are good things. Second, it is often the only way to solve a given problem. Human bodies have a large number of redundant degrees of freedom (Bernstein, 1967), which means that we can in principle solve a given task in many different ways. The only way to reliably pick one solution is to allow yourself to be constrained by the task dynamics, because otherwise there are too many possible ways to move and we would be frozen by indecision. So you identify the task and then pick a solution. This problem also applies to the analysis of human movement - in order for a scientist to understand why we moved the way we did, we have to constrain our possible explanations to those that fit in the task dynamic or else we will never find the answer (Bingham, 1988).

An example: I can, in principle, reach for my coffee in a straight line, or in a wide curve, or with my arm going behind my back. However, I typically don't, and the solution I adopt in typical cases (reaching in a straight line) is the stable one offered by the dynamics of the task at hand (things like where the objects are, etc). If those dynamics change (say an obstacle is placed in front of me) my redundant degrees of freedom provide the flexibility to produce a new solution - but this solution will again reflect the dynamics of the new task, the one which now includes an obstacle. Each solution is constrained by the dynamic properties of the part of the world I'm interacting with; these properties are the affordances of the task and so this is what it means to say actions are controlled by the perception of affordances. (By the by, Rosenbaum investigates this kind of thing in the context of his 'end-state-comfort' effect, which states that people typically plan their actions so that they end their movements in a comfortable posture. What he actually means is that their movements unfold so as to conform to the perceived underlying task dynamics, but his analysis has never taken that final, critical step into dynamics that we will here. I just realised this as I was writing and it's useful.)

Task Dynamics
These facts of the matter mean that we have to be able to distinguish one task from another, and by we I mean both me the perceiving-acting organism and me the scientist studying perceiving-acting organisms. You can only do this at the level of task dynamics (Bingham, 1995) and this is why dynamical systems theory got such a foothold in this kind of work.

Dynamical systems theory is a set of mathematical tools for describing how a system is put together and how that system changes state over time. The latter is dictated by the former, and therefore measuring those changes over time (the dynamics of the system) & how that dynamic responds to perturbations provides information about how the system is put together; both it's composition (what elements make up the system) and it's organisation (how those elements are arranged; I talked about this previously in the context of developing the perception-action model of coordinated rhythmic movement).

Dynamics provides all the elements required to describe both the form of the change over time (the kinematics) and the forces which caused that particular motion (the kinetics). Kinematic variables include time, position and all the temporal derivatives of position (velocity, acceleration, jerk, etc). Kinetic variables are all of these as well as mass. A given dynamical system description of an event in the world uses some combination of these variables and arranges them in a particular way, in an equation (via addition, multiplication, etc). These equations then fully describe how the given event type unfolds over time and why; by changing the parameters on the variables (say, making the movement faster or slower) creates a specific instance of this event type. Finally, two or more dynamical systems are coupled to each other if some of the terms in one equation reflect terms from another, and vice versa (see the model again for an example).

Two different events are different tasks if they are unfolding according to different dynamics, i.e. if they are described using different equations (different combinations and arrangements of dynamical variables). The physics of something in free fall and something being propelled are described by different equations and are thus different tasks (this difference is the origin of our ability to distinguish actively generated biological motion from the passive motion of a non-biological object in motion). Two different events are the same task if the dynamics are the same, even if the parameters of those dynamics are different.

Information
Perceptual systems face a bottleneck. In order to be able to identify events and tasks, we need access to the dynamics. Perceptual systems, however, are only sensitive to kinematic patterns caused by those dynamics; no mass! The central question for ecological psychology is therefore, how can a kinematic variable come to mean the underlying dynamical event to a perceiving-acting organism?

The current answers are kinematic specification of dynamics or 'something else'. Gibson's big contribution to psychology was to identify that specification was possible; Turvey et al (1981) laid out how it might work (and insisted on it for direct perception) and Runeson & Frykholm (1983) then showed how it happened and expressed that in the language of dynamics and kinematics. The 'something else' options include standard cognitive indirect mechanisms of perception (e.g. Rock, Marr), or, more recently, Chemero and Withagen's attempts to establish that 'kinematic correlation with dynamics' is enough for direct perception. 

I still think specification is typical in perception-action cases because it can be; we will obviously need to widen things out a bit to handle language. Regardless, the set-up is the same: dynamical events produce kinematic consequences which are detected by perceptual systems and must somehow come to mean that event. The scientific analysis must therefore produce an exhaustive list of the kinematic consequences, identify their relationship to task relevant dynamics and then empirically establish which kinematic variable people are using to perceive which dynamical property. The complete story will also include how we learn what a given kinematic pattern means, dynamically; babies start out sensitive to kinematics but initially unaware of how those kinematics relate to the underlying dynamics.

The list can be exhaustive if and only if you have a firm grasp of the underlying dynamics. This, historically, is why all of this is coming out of the perception-action literature rather than, say, language. The dynamics that produce a language event are formidable in their complexity and identifying those dynamics is unlikely to be within the range of the average psychologist's grasp of dynamical systems theory. The dynamics of simple perception-action tasks are much more constrained. Coordinated rhythmic movement is very tractable, which is why the task got such a foothold in the literature. Interception, catching, bouncing balls on a racquet, prehension, aiming - these tasks are all managable and serve as model systems where we can develop the relevant methods in a task we can keep control of. If we get these basic systems right, we can then start to move back up, and that's what we're doing. 

Throwing is a great task for scaling up our analyses, because it is fundamentally just an example of projectile motion and these dynamics are straight forward enough that my laptop can run meaningful simulations in just a few hours. It is a complex task that contains a lot of elements, though, and so it's a perfect test bed for scaling our ambitions up. 

Summary
Tasks in the world can only be uniquely described (and therefore identified) in terms of their dynamics. When we say you need a task analysis, what we really mean is that you need a description of the composition and organisation of the underlying task dynamics in as complete detail as possible. This is why we tackle tasks where we can handle the dynamics, such as throwing.


Dynamical events create kinematic variables. It is an empirical question which of these acts as perceptual information for a given dynamical property, and the answer will only be good for a given task. It is also an empirical question as to whether the information is informative via specification; regardless, specification is an option and is not as high a bar as has been suggested.

We perceive dynamics via kinematics; there is the world, and there is information for the world, and our behaviour is constrained by the manner in which we perceive those dynamics. For example, in throwing, the relevant dynamical properties of a projectile are size and weight, but we perceive these as the perceptual quantity 'heaviness', and our behaviour reflects heaviness, and not just size and weight.

The research process is therefore 1) identify the task dynamics and therefore the affordances in a given task, 2) identify the kinematic consequences of that dynamic and therefore the potential sources of perceptual information for that dynamic, and 3) empirically establish which variables are used as information for what properties and how. This is also the form of the process by which we think you should study behaviour in general. Post on throwing will clarify how this process unfolds in perception-action research; posts on language, etc will then apply and adapt this process to work with more complex dynamics.

References
Bernstein, N.A. (1967). The co-ordination and regulation of movements. Oxford: Pergamon Press.

Bingham, G. (1988). Task-specific devices and the perceptual bottleneck Human Movement Science, 7 (2-4), 225-264 DOI: 10.1016/0167-9457(88)90013-9

Bingham, G.P. (1995). Dynamics and the problem of visual event recognition. In Port, R. & T. van Gelder (eds.), Mind as Motion: Dynamics, Behavior and Cognition, (pp403-448). Cambridge, MA: MIT Press.  Download

Runeson, S. (1977). On the possibility of "smart" perceptual mechanisms Scandinavian Journal of Psychology, 18 (1), 172-179 DOI: 10.1111/j.1467-9450.1977.tb00274.x

Runeson, S., & Frykholm, G. (1983). Kinematic specification of dynamics as an informational basis for person and action perception: Expectation, gender recognition, and deceptive intention. Journal of Experimental Psychology: General, 112, 617-632.