[From Bill Powers (2003.12.29.1457 MST)]
Bruce Gregory (2003.12.29.1528)--
This is what I am trying to understand. How do we model "when the
occasion arises" which sounds to me a lot like S-R.
"The occasion" that I had in mind was an action by a higher-order system. I
suppose you were thinking of "discriminative stimuli", which would be the
S-R version of an occasion.
"I set positive
reference levels" means, "the hierarchy sets positive reference levels"
does it not?
That's the theory.
Why would this happen if not because some high level
perception is being disturbed?
Actions are not produced ONLY because of disturbances. Suppose you decide
to have cashews for breakfast. What happens next? Since you're not yet
having them, this reference signal (set by a higher system which is
maintaining its own perception of some sort) produces an error signal. The
error signal has to be translated into those actions that will end up
finding the cashews, getting them out of their can, and with some degree of
ritual, picking them up and putting them in your mouth. None of those
actions is caused by a disturbance in the environment.
The nature of some perceptions is that in order to keep them near steady
reference levels, it is necessary to produce extended, repeated, or
continuous actions for some length of time. Just think of "jogging," which,
to be kept in a constant state, requires half an hour or so of running
actions. Or even more extreme, think of the perception of "getting an
education."
I can imagine a set of "scripts" of the
kind Roger Schank describes (perceptions of sequences) that are always
testing for appropriateness ("Are their cashews in sight? If so, take a
few and eat them. If not keep looking) but I doubt that this is what
you have in mind.
I know what you mean -- why does simply seeing food make us hungry, when a
moment before we hadn't been thinking about food at all? The obvious
explanation is that the food is a stimulus to food-obtaining behavior, with
the feeling of hunger being a side-effect, an epiphenomenon. This could
even be merged with PCT by saying that the stimulus causes some system to
react by setting reference levels for food intake. Bruce Abbott has
suggested such a possibility for higher levels of control.
Such an arrangement is, of course, possible. So how do we decide whether it
actually exists? One way is to follow out the logic of the model and see if
it predicts what we observe.
Suppose certain foods are in fact stimuli that cause certain systems to
start producing actions that obtain food, position it, and ingest it. If
that's the correct explanation, we should be able to cause food-obtaining
behavior at any time, just by applying the correct stimulus. I'm sure that
as you think of examples, problems come to mind -- what if you've just
eaten a pound of cashews? One would predict from experience that presenting
another bowl of cashews would NOT cause cashew-obtaining behavior, but
something more akin to throwing up. However, the theory says that the
stimulus should cause the response, so you should eat the bowl of cashews.
Since we know this prediction is unlikely to be supported, the theorist
would be required to explain why the new cashews are rejected. This is
where auxiliary theories come in, as required to patch up the incorrect
predictions of the starting theory. For example, it could be proposed that
intake of more than a certain amount of cashews in a day produces
"satiation," which shows up as failure to ingest more than some maximum
amount of cashews per day.
I don't like such patchwork theories, since every new phenomenon requires a
modification of or addition to the basic theory. And I also don't like
theories that require switching back and forth between one fundamental
brain architecture and a totally different one -- between SR theory and
control theory. There are many phenomena that can be explained only by
control theory, so we know that the organism must be organized as a control
system in many situations. We also know that a control systen can _appear_
to behave like an S-R system if the "S" is a disturbance of a controlled
variable and the R is an action that cancels the effect of the disturbance.
Given such well-established principles, what are we to do when we see an
example that seems reminiscent of the S-R model? One possibility is to view
these examples as refutations of control theory (disregarding all the cases
where only control theory works), or to propose that the brain stops acting
like a set of control systems under certain circumstances and switches to
an open-loop type of organization. The other possibility, which I greatly
prefer, is to search for an explanation that is consistent with the
control-system organization that works in most other examples. I do not
want the brain to have two different and incompatible organizations.
So -- do you want me to conjecture about possible PCT explanations for the
way the sight of some foods seems to make us hungry, or do you want to try
your hand at it? It doesn't really matter which of us does it, since
without experiments these will remain untested possibilities. But perhaps
you need the practice somewhat more than I do.
As to Schank's proposal (basically the TOTE unit, it would seem), can you
think of any way to test it?
Best,
Bill P.