Tom Bourbon [941012.0901]
[From Bruce Abbott (941011.2050 EST)]
Tom Bourbon [941011.1416] --
. . .
. . . I do not agree,
however, that the use of statistical averaging necessarily invalidates a given
finding. As an example, Ebbinghaus's careful studies of his own capacities
for learning and retaining nonsense syllables have been replicated time and
again in group-based studies. In some cases averaging is the only practical
route available. Sternberg's (1966) investigations of search in short-term
memory come to mind, which used average reaction times while manipulating the
length of the list to be searched as evidence that subjects search the items
in STM in serial order.
These seem to be examples of apples and oranges, Bruce. Ebbinghaus's data
are highly reliable; the data for a given person are very similar across
replications and the data for different individuals are similar, in much the
same way as some of the data from studies of psychophysics or conditioning,
or from studies of perceptual control. In those instances, one can
calculate averages, but the data from one participant reveal the phenomenon
clearly.
The Sternberg case is different. Whether there is a behavioral phenomenon at
all is a matter for debate. If all people have STM and if all of them
always search STM in serial order (as the summaries and generalizations
based on his work would have us believe), that fact will be obvious in the
data for every participant. If the phenomenon cannot be observed for all
persons, but can be found only by calculating descriptive statistics from
data gathered across many subjects and many trials, then by definition it
is not a feature of the behavior of each person. In such a case we run the
risk of creating a "phenomenon." Behavioral scientists are good at that,
even if they are mostly unaware of the fact.
. . . I'm also arguing that
some facts established in non-PCT-oriented research should help to illuminate
such things as the processes by which perceptual signals are created and
transformed,
Agreed. We need much more work along that line. As a theory of perceptual
control, PCT must come to include some pretty detailed models for many
diverse perceptual input functions. We must rely on other people to do most
of that work. I believe their investigations into the nature of perception,
and perceptual functions, might prosper, were they to recognize that
perception is the purpose of behavior, in place of the traditional model
that says perception is a step in a lineal chain of cause and effect that
runs in from the environment (through perception, to cognitive-neurological
processes that develop plans for action), then out to the execution of
actions.
or what determines WHICH perceptual control systems get access to
the final common behavioral pathway at any given moment.
I'm not quite sure what you mean by this.
. . . I do not assume
that the results of _all_ perceptual studies will be relevant to PCT. In
fact, I believe many of them will be irrelevant.I agree. I never made the claim that all would be relevant.
We agree.
. . .
Tom Bourbon [941011.1658] -- Re: Rescorla-Wagner and PCT
I have read your reply but please bear with me--it will be a little while
before I can respond (I'm teaching a 12 hour load).
_Only_ 12 hours? 
I'll wait.
. . . For the time being, let's
assume that V and L represent perceptual quantities, that dV = aB(L - V) and
V = V + dV compute the output, that the environment is represented as Vin =
Vout, and that the input function is Vp = Vin (i.e., input gain = 1). These
assumptions make V and the external variable presumably being controlled
(quantity of salivation per CS presentation) identical for the simulation, but
this could easily be corrected.
As I thought. Good, but not so good. The R-W model is a model for control
of output, not for control of variables affected by output. There have been
a number of serious attempts in learning theory to develop working models
for behavior. The ones that I know (Staddon; Timberlake; and others) have
produced models in which the organism is seen as controlling its output
(bar presses, running behavior, eating behavior, and so on). So far as I
can tell, their models have failed. Within the community of learning
theorists, I have seen many people then criticize the modelers for having
used a procedure (modeling) that is not a legitimate procedure for learning
theorists -- they missed the possibility that the models failed because they
represented organisms as controllers of their own outputs, rather than as
controllers of perceptions. (I always thought that, as bad as it can be for
PCT modelers, things might be even worse for learning theorists who dare to
work with generative models.)
Oh, one more thing. I have a computer and I know how to program simulations.
Great! Will we perhaps see some PCT models of learning phenomena?
Later,
Tom