Perceptions as Variables

[From Rick Marken (930907.0900 PDT)]

These are some meditations that occurred during a beautiful hike
yesterday in the Santa Monica mountains. They started as an
attempt to formulate a reply to Gary Cziko (930906.0315 UTC).

I think the essential insight about perception that comes from PCT
is that perceptions are VARIABLES. Unfortunately the people who did
the first single unit recording work (like Hubel and Wiesel) and found
relationships between unit firing rates and variables in the environment
did not really get the idea that the variations they observed could be
viewed as variable (and, thus, potentially controllable) perceptions.
I think the version of "encodingism" to which Gibson provides an
alternative is based on a failure to see perceptions as variables.
For example, an "encodingist" apparently looks at the Hubel and Weisel
work and concludes that neurons "code" for environmental events like
vertical lines; a vertical line is "perceived" when the firing rate
of the "vertical line detector" is maximum. This way of looking at the
single unit results does imply that there must be another perceiver
around, looking to see when the output of the "vertical line" unit
is maximal -- so it can say "now there's a vertical line out there".
This "encodingist" view of the single unit results suggest that all
values of the neural signal except one (the maximal value that codes
"vertical line") are, at best, indications of the degree to which
the perception is "vertical line" or, at worst, noise.

PCT sees the results of the single unit studies completely differently;
the rate of firing of the unit DOES depend on an external variable, but
it doesn't "encode" anything in the sense described above. The variations
in the firing rate are seen as an analog of variations in an environ-
mental variable (in this case, the orientation of a line). Accord-
ing to PCT, ALL values of unit firing rate are significant; its the
variations that matter, not many particular value of the firing rate.
The variation in the firing rate is what does the "encoding" --
it is encoding variation in an environmental variable.

In PCT, the system doesn't need to know what any particular value of a
perceptual variable (neural firing rate) corresponds to in terms of
the external physical variable. If the perceptual variable is controllable
(because the system can influence the environmental correlate of that
variable) then the system can control that variable. In order to model
this process, the modeller must make some assumptions about what variations
in the environment (as represented in other models -- like physics)
are represented by variations in the perceptual variable; the PCT
modeller must make guesses about the nature of the perceptual function,
f(), that transforms the sensory representation of environmental variables,
s, into perceptual variables, p. So modelling f() is an important part
of PCT. That's why saying things like Gibson's "the system detects
environmental invariants" is not very helpful (one of several reasons,
actually; saying this is just a rejection of modelling -- as Bill said).

So PCT really incorporates sensible versions of encodingism AND
interactionism. PCT encodingism sees the sensory consequences of
environmental VARIATIONS,s, encoded by perceptual VARIATIONS; it
assumes that p = f(s) where both p and s are VARIABLES. This
version of encodingism is combined with a sensible version of
interactionism -- one that recognizes that the sensory consequences
of environmental variations, s, can be, at least in part, a result of
VARIATIONS in actions of the organism, o, so s = g(o).

Failure to recognize perceptions as VARIABLE is also the reason for
pseudo-issues like the idea of a "grandmother cell". The idea that
there might be a "grandmother" cell is again a result of misreading
(from a PCT perspective) the single cell studies. A "grandmother
cell" implies that you see "grandma" when a particular cell fires
maximally. Actually, the perception of your grandma (from a PCT
perspective) results when many different perceptual VARIABLES happen
to be a particular levels. There may be a configuration cell that
represents variations in overall facial configuration, for example.
When this cell fires at a particular rate you see a configuration
that is your grandma; when it fires at different rates you see
anything from Helen Hayes or Golda Meir.

The single unit recording work should have placed conventional
perceptual psychology on a track compatible with PCT. The fact
that it didn't (and that the debate is at the level of what
we see in Bickhard, et al) is, I believe, a result to the
abandonment of the analog model of the behavior of the nervous
system. The version of "encodingism" that Gary Cziko rightly
castigates is a product of an inappropriate comparison of neural
activity to the behavior of the digital (rather than the analog)