Monday, 13 April 2015

Specification and Some of Its Consequences (A Purple Peril)

Perception is how we maintain psychological contact with functionally relevant objects and events in our environments. Explaining how we do this means describing that environment in appropriate terms and investigating what information might possibly exist for that environment, given that description. The ecological hypothesis is that the correct level is dynamics, and that describing the environment this way allows there to be information that can specify those dynamics. This information can support the kind of behaviour we need to exhibit. 

This Purple Peril describes what is meant by specification, and what that implies for how information comes to mean something to an organism. There is more detail in the various links, so check those for information too.

Dynamics is the description of the behaviour of objects and events that includes an account of how they change state over time with reference to the forces that create those changes. Events and objects in the world can only be uniquely identified at the level of dynamics (Bingham 1995); this is therefore the level at which tasks must be described.

Organisms need to coordinate their activity with respect to the dynamics of a task, but they have no access to the dynamics per se. This is because the vast majority of what happens in the environment happens 'over there' and it makes no mechanical contact with the organism. However, (at least some of) the dynamics of an object or event can be projected into patterns in ambient energy arrays as those dynamics interact with the arrays. These patterns are described kinematically, that is, as patterns of motion or change without reference to the forces that created the change. Patterns in these arrays are available 'at a distance' from the dynamic that created them. These patterns are therefore all the organism has to work with in order to maintain psychological contact with the local task dynamics (the perceptual bottleneck; Bingham, 1988). Organisms are (extremely) sensitive to these patterns.

These patterns are not identical to the task dynamic, but they can be specific to them (Kinematic Specification of Dynamics, KSD: Runeson and Frykholm, 1983). This means that there can be a 1:1 mapping between the aspect of the dynamic being projected and the kinematic projection; one aspect, one projection. Because of this specification relation, detecting the projected pattern is equivalent to perceiving the dynamic that created it. Organisms can therefore detect the patterns and use it to respond to a local task demand. When organisms learn to use the patterns to organise their behaviour with respect to the dynamic property that created the pattern, that pattern becomes information about the property. This perception-action system of detection and use is the mechanism that supports direct perception of the environment; no internal processing of the information to turn it into something other than itself is required.

Specification and meaning
This analysis has the following corollary. Because the organism is using the information as if it were the dynamical property, the information comes to mean the property to the organism. This works because of specification: organising your behaviour with respect to the information as if it were the dynamical property succeeds because of the 1:1 mapping between them. You really can let one stand in for the other. This is why specification is so central to the ecological approach (Turvey, Shaw, Reed and Mace, 1981) and why messing with specification is such a risky business. We will get to messing with it later, because Sabrina's analysis has already shown we need to get into it. But later. 

Meaning via specification means no inference

The organism does not store this meaning anywhere. It demonstrates what the information means by virtue of it's behaviour in the presence of the information. It does not need to have this meaning available somewhere other than it's activity, because it doesn't need to have access to the meaning to form part of any inference. Alternatively, because it doesn't need to store the meaning anywhere as separate information, this is evidence that the organism doesn't engage in any inferences involving that meaning. 

Key References
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. Download

Turvey, M. T., Shaw, R. E., Reed, E. S., Mace W. M. (1981). Ecological laws of perceiving and acting: In reply to Fodor and Pylyshyn (1981) Cognition, 9 (3), 237-304. Download 

Relevant, More Detailed Blog Posts
Task Dynamics And The Information They Create


  1. I haven't read much of the literature on this, but I have a question on the issue of "vagueness". My intuition is that it is possible to directly perceive vagueness.

    That we cannot yet act, or that we need more information (in a non-technical sense) to decide what to do is something that is often apparent. Sometimes it comes with a strong sense of how to relieve the ignorance (see Erik Reitveld's philosophical stuff on unreflective action and the Wittgenstinian notion of directed discontent). Sometimes it is simply a matter of confusion and inability to organise action effectively.

    Saying that there's a 1:1 relationship between a particular vagueness and a particular set of ecological information seems like an odd claim though. It seems to me that the more tolerant forms of the notion might be necessary to allow us as actors to move from situations of uncertainty to situations where there are a clear set of actions with will have concomitant dynamics with which to attune.

    I may need a whole post of my own to unpack some more of this question, but I'll wrap up with just this:

    My intuition is that standard ecological approaches deal well with the clearly bounded domains where the dynamics can be specified 1:1, but that much cognition happens as a means to get us from more nebulous situations into those more cleanly defined ones. I know Gibson talks about exploratory behaviour but I haven't read him properly on it. Is there some notion of nested information structures (complex-but-vague task domains within which more specific tasks occur) extant in the theoretical literature?

    1. My intuition is that standard ecological approaches deal well with the clearly bounded domains where the dynamics can be specified 1:1, but that much cognition happens as a means to get us from more nebulous situations into those more cleanly defined ones.
      This is certainly true. The specification story is not the complete story of information, which is complicated because of all the important things specification makes possible.

      Can you be more specific about what you mean about vagueness? What's an example?

      At a first pass, it's unlikely the 'vagueness' will create information. However, 'vagueness' might be the experience of inadequate/insufficient information, with that inadequacy revealed when you try to achieve some behaviour using that information and fail because it's not up to the job.

  2. "Can you be more specific about what you mean about vagueness? What's an example?"


    Irony aside though, I suspect that the ideas I've had in mind are mainly situations as you describe - where there is insufficient information, though also situations in which the information is present but I am unable to attune to it due to lack of ability.

    There is a metacognitive aspect to the experience that's interesting and important. I know what kind of thing to do, but not what specific thing.

    For instance, in a conversation I might know that I'm supposed to comfort someone, but have no idea how to do so. Or in a game of soccer I might know I should pass the ball to a teammate making a run, but also know that I probably can't. The inadequacy of the information is not simply revealed by my failure - my experience of the situation is one in which failure is the most likely outcome even before the attempt is made.