Tuesday 16 March 2010

There is no poverty of stimulus

In my last post I outlined the basic state of affairs in the study of perception up to the modern day. Cognitive science assumes a poverty of stimulus that must be overcome with internal mental representations. The content of these representations is an empirical question, and the drive in psychology ever since the ‘cognitive revolution’ of the 1960s has been to uncover the contents and format of these representations such that they can do the job apparently required.

The thing I like most about Gibson is that he noticed the basic assumption was flawed, and went all the way back to basics to try again. His initial concern was that the research programme about how representations enriched a poor stimulus raised too many unanswerable questions: if the stimulus is so poor, how do you ever figure out what’s required to enrich it? In other words, on what basis is the representation formed? Gibson decided that this incoherence was pointing to a fundamental flaw, so he re-asked the only question that ever matters in perception: what is the information for a perceiving organism? The difference was that instead of looking at the physiology of the eye to decide what it could possibly detect, he looked at the physics of the world to see what might possibly be around and informative; only then did he ask whether the organism could use it.

This is an important move, and it was motivated by an attempt to ‘naturalise’ perception (make it continuous with biology, evolution, etc). Gibson reasoned that you couldn’t understand how the eye worked until you understood the kinds of things it might be interacting with; in other words, what is the niche that the eye evolved to fill?

So, unlike any other book on perception, Gibson’s 1979 book ‘The Ecological Approach to Visual Perception’ begins with a consideration of the environment; what there is to be perceived. He notes that at any given point of observation there is light passing through that either comes directly from a light source or that has been reflected off a surface. The latter light has been structured by its interaction with the surface: this structure might then be informative about the surface, if that structure is reliable.

Gibson noted two other things: first, while specific details in the structure vary wildly over time and space (e.g. the exact intensity of light from a surface), there are relational properties that remain invariant (e.g. the ratio of intensities). Second, perceiving organisms are constantly on the move (the point of observation travels continuously through space, over time). This motion actually causes a lot of the wild variation (which has always meant that motion is a real problem for cognitive theories of vision; Gibson realised any theory for which motion (which is the rule, not the exception) is a problem is a theory in trouble). The motion, however, also reveals the things that don’t change.

The net result is that motion reveals the invariants over the transformation, and anything that remains invariant over time is the kind of thing you might expect an organism to be able to tune into. Think about evolution: organisms quietly specialise into niches over extended periods of time, and are able to do so because the evolutionary pressures of the niche remain essentially constant over time (until the niche alters for some reason). Evolution into a stable form is only possible because niches don’t change fast (hence the current rapid climate changes are so potentially catastrophic). The only things that remain invariant in this way are higher-order relations, and they have a potentially vital feature, namely they can specify something.

Higher order relations: By higher order, I’m referring to the mathematical sense, specifically that the relation is the result of an operation on simpler elements and reflects their combination. The result of the operation can remain the same while the simpler elements change (e.g. 2*3 and 1*6 are both equal to 6); mathematically this means that the operation requires multiplication or division in it somewhere. Perceptually, we tend to talk a lot about ratios.

A simple example: Pythagoras’s theorem defines what it means to be a right angled triangle in the form of a relation. You are only a right angled triangle if you obey this relation, but if that is true then the specific details can alter widely.

Both of these are right angled, but only the higher order Pythagorean relation specifies this fact
Both of these are right angled triangles (they both obey Pythagoras’s theorem, which defines the specific relation (c2 = a2 + b2) required to be a right angled triangle) but one is twice the size of the other. You could readily effect this transformation by having the small triangle drawn on the wall and approaching it; the triangle would grow in apparent size but the relation between the sides would remain fixed. There is a many-to-one mapping between stimuli and identification of the object as a right angled triangle using the individual lengths, but a one-to-one mapping using the relation.

Gibson’s intuition (expressed in terms of my example) is this: any organism that could detect the relation would therefore always be able to identify a right angled triangle, regardless of changes in the details caused by motion, current perspective on the object, etc. The relation specifies a right angled triangle, and specification means that if you detect the relation, there must be a right angled triangle present. For our organism, perception is no longer ambiguous.

Of course, right angled-ness in triangles is not the kind of thing Gibson claimed we can do; this is merely an illustration of the point. But Gibson had realised that these kinds of invariant relations do exist, and that any organism that could detect them could therefore be said to have detected the thing in the world that expressed the relation. For the first time, there was a theory of perceptual information that allowed for that information to be unambiguously related to the world it was information about – the stimulus was anything but poor.

The immediate question then arises – what kinds of relations exist that might be functionally relevant to organisms, and can these organisms detect and use them? This, in a nutshell, is the ecological psychology programme of research.

No comments:

Post a Comment