a question

[from Jeff Vancouver (2002.02.01.0930 EST)]

First just FYI. I consider that I do basic and applied research in PCT.
Nearly all my work is published in applied psychology journals, though with
a decidedly basic flavor (relative to other articles in those journals).
Indeed, at the upcoming conference in Industrial/Organizational Psychology
my research group is presenting 7 things all based on PCT. To see what we
are presenting, click on the "full list" to presentations and invited
addresses on my home page
(http://www.psych.ohiou.edu/people/Faculty/Vancouver/vancouver.html).

Anyway, I have found CSGnet very helpful and hope to continue to find it
helpful. Toward that end I have a question which I am sure has been
addressed, but for which I have no answer currently.

Specifically, how does PCT explain the following type of learning. You have
just finished a satisfying meal with friends when one says "That was almost
as good as the meals and service at such-and-such restaurant." Never having
heard of this restaurant, you inquiry more and decide to try it in the
future.

A more mundane example is the rat in the Skinner box who is acting away for
whatever reason (i.e., because of errors in unspecified control units),
stumbles on an action (press lever) that results in a food pellet. From
this, a reference signal ("press level") now eminates from the output
function of "get food." That is, an output function has been reorganized,
though it was not in a reorganizing mode (no essential or intrinsic variable
discrepancy - see esp. example #1).

Yes, this is operant conditioning, so I expect Bruce Abbott to chim in.
What am I missing?

Jeff

[From Bjorn Simonsen (2002.02.02.12:00 EMT)]

from Jeff Vancouver (2002.02.01.0930 EST)

Pardon me for my contribution.

Specifically, how does PCT explain the following type of learning. You

have

just finished a satisfying meal with friends when one says "That was almost
as good as the meals and service at such-and-such restaurant." Never

having

heard of this restaurant, you inquiry more and decide to try it in the
future.

A more mundane example is the rat in the Skinner box who is acting away for
whatever reason (i.e., because of errors in unspecified control units),
stumbles on an action (press lever) that results in a food pellet. From
this, a reference signal ("press level") now eminates from the output
function of "get food." That is, an output function has been reorganized,
though it was not in a reorganizing mode (no essential or intrinsic

variable

discrepancy - see esp. example #1).

Yes, this is operant conditioning, so I expect Bruce Abbott to chim in.
What am I missing?

I don't think you are missing anything in those two examples. I guess I
could have experienced as you describe in your first section.
But I also guess that I have had the experience that a friend of me told me
about an interesting picture (book) he had seen (read). And I think there
are examples where I decided not to see the picture (book) myself.

And we all know about the rat which behaved as in your description in your
second section. But we also know about the rat which the next day didn't
notice the press lever at all.

You may call your examples operant condition, I explain them with PCT.
PCT explains that we don't behave or learn to produce a particular action as
a reaction to events in the world outside, we behave according to (or
learn) control systems. Some times the actions are as you describe them and
sometimes the actions are different. It depends on the error and .... .

As I look it PCT include your examples (and all examples) and Operant
condition don't explain all actions.

This is so simple for me and your contribution make me unsure. Maybe I
misunderstand you?
I feel I have been fooled to write what I have.

Specifically, how does PCT explain the following type of learning.

You don't fool me to explain that.

bjorn

[From Bill Powers (2002.02.02) -- another magic number --

Vancouver (2002.02.01.0930 EST)--

Specifically, how does PCT explain the following type of learning. You have
just finished a satisfying meal with friends when one says "That was almost
as good as the meals and service at such-and-such restaurant." Never having
heard of this restaurant, you inquiry more and decide to try it in the
future.

No definitive answer from here, but I'd guess that there would have to be a
desire to eat better or at least different food before you'd take the
trouble to write down or memorize the location of a recommended restaurant.

A more mundane example is the rat in the Skinner box who is acting away for
whatever reason (i.e., because of errors in unspecified control units),
stumbles on an action (press lever) that results in a food pellet. From
this, a reference signal ("press lever") now eminates from the output
function of "get food." That is, an output function has been reorganized,
though it was not in a reorganizing mode (no essential or intrinsic variable
discrepancy - see esp. example #1).

The standard conditions for getting rats to do this sort of thing include
withholding food until they are at 80% of free-feeding body weight. What
would you weigh at 80% of your present weight, and how hungry would you be?
I think there is plenty of intrinsic error in a Skinner box. I think the
"get food" motive is very definitely present from the start.

Second point: when a rat is in an experimental cage with a lever, it noses
around continuously, apparently looking for food or water or female rats or
anything else rats find interesting. The first lever press results from
this search pattern, which pretty much covers the cage. The result of food
delivery (as I have seen in a few videotapes made by Bruce Abbott) is not
an immediate switch to regular bar-pressing, but a concentration of the
search in and around the food dish as well as around the position where the
rat was when the food was delivered. It's pretty clear that the rat does
not immediately identify which action it was that produced the food. The
lever pressing emerges as a distinct action only after a great many trial
and error actions around the food cup and lever (Bruce tells me that the
lever-pressing eventually becomes a distinct sort of action -- in the few
tapes I've seen, it never got much past a side-effect of swarming all over
the food-cup and lever areas). What I have seen looks to me very much like
a reorganizing process, but not purely -- there's also an initial search
pattern that is quite similar across the rats I saw, six or seven of them.
After the rat learns to feed itself, the search pattern ceases, at least
until the rat is full. After 20 or 30 minutes of eating, the rat is likely
to take a nap or search some more (note that no water is usually provided
in the experimental cage, and the food is dry).

Best,

Bill P.

[from Jeff Vancouver (2002.02.06.1010 EST)]

[From Bill Powers (2002.02.02)

No definitive answer from here, but I'd guess that there
would have to be a
desire to eat better or at least different food before you'd take the
trouble to write down or memorize the location of a
recommended restaurant.

So is this reflecting a reorganization process or a much more mundane
encoding into memory (but then what function "knows" to address that memory
link?).

The fundamental issue is that association models seem to handle some types
of learning (e.g., incidental learning) more readily than PCT models. What
you should (or maybe already) know is that association models no longer
refer to S-R assocations, but rather to S-S and R-C associations. This
change is highly compatible with PCT (S-S links are easily conceived of as
aspects of input functions and R-C associations are links between an output
function [Okay, R for response is really error signal or lower-order
reference signals from the output function] and perception [where C is
really perception of consequence, not consequence per se, although I wonder
is there is more here]). The question is, are the creation of these
associations via controlled processes? It is here that I am having trouble.
In particular is the case where some consequence results in a pleasurable
experience (i.e., the releasing of endorphins) where the pleasurable
experience cannot be claimed to be the reduction of a discrepany in a
reorganizing control unit.

Of course, a model of this would be great (note, my model of the adaptive
control unit cannot handle this type of learning).

Jeff

[From Bill Powers (2002.02.06 MST)]

Jeff Vancouver (2002.02.06.1010 EST)--

[From Bill Powers (2002.02.02)

No definitive answer from here, but I'd guess that there
would have to be a
desire to eat better or at least different food before you'd take the
trouble to write down or memorize the location of a
recommended restaurant.

So is this reflecting a reorganization process or a much more mundane
encoding into memory (but then what function "knows" to address that memory
link?).

Which part do you refer to as "this"? The proposed "Desire to eat better"
would be a reference signal, and the means of doing so might include
writing down the names of good restaurants, or "encoding into memory" the
same information.

The fundamental issue is that association models seem to handle some types
of learning (e.g., incidental learning) more readily than PCT models.

I've also thought that association-like phenomena might explain some kinds
of learning, though I never got around to figuring out how they would fit
into the memory model in B:CP. There's a discussion of associative memory
starting on page 211 or thereabouts, in the section on addressing memories.
Why doesn't everything get associated with everything else, and what would
be the effect of that? But there's probably a good idea in there somewhere.

What
you should (or maybe already) know is that association models no longer
refer to S-R assocations, but rather to S-S and R-C associations.

I never thought they did. Isn't an association the phenomenon of a
perception reminding you of another (past) perception? From the assumptions
underlying the PCT viewpoint, associations don't occur in the environment,
but in the brain. It's hard for me to see how a stimulus (a physical event
outside the organism) could associate with another stimulus, or how a
response, also a physical event in the outside world, could associate with
a consequence in the outside world. Doesn't perception really underlie each
of these pairings? I'm just guessing, since I don't follow that literature.

This change is highly compatible with PCT

Are you sure? Have you considered every aspect of PCT that might be
affected by introducing arbitrary connections of an unspecified kind at
various points in the model? Remember that "association" is the least
specific possible reference to a connection between two variables. It
doesn't say anything about direction or form, or the conditions under which
it is seen, or what the mechanism is. It just says that one variable
somehow has something to do with another variable. That's not saying very
much -- I don't see how you could even describe it sufficiently to put it
into a model.

It is here that I am having trouble.
In particular is the case where some consequence results in a pleasurable
experience (i.e., the releasing of endorphins) where the pleasurable
experience cannot be claimed to be the reduction of a discrepany in a
reorganizing control unit.

I don't see why the _releasing_ of endorphics should be pleasurable.
Pleasure is something we experience, isn't it? In other words, a
perception. If the reference level for the perception(s) affected by
endorphins is set to a high level, then when they are brought to that level
we experience that situation as pleasurable. What else is pleasure but
bringing or having brought a perception to the level we prefer? Any other
definition gets into circularity: why do we seek pleasure? Because we like
it. And why do we like it? Because it gives us pleasure. This is no better
than "What is pleasure? The release of endorphins. And why is the release
of endorphins pleasant? Because it gives us pleasure." As Mary says, slyly,
"Pleasure is Nature's way of telling us we're having a good time." Our
experience of pleasure is _defined_ by matching experience to reference
level. If we had a low reference level for the concentration of endorphins,
we would experience a high level as unpleasant. Endorphins, I would guess,
probably are part of some control loop that nobody has identified yet.
Whatever is being controlled, when control succeeds we feel pleasure. But
we also experience pleasure when we succeed in controlling many other
perceptions, endorphins or no endorphins.

It's not clear to me just what part of the control loop goes with the
concept of pleasure. It might seem to be the error signal, but so far I
have never found a way to allow perception of error signals that didn't
make a working model nonfunctional. The nearest to a solution I have come
is to suppose that we sense the effort to correct the error, in some
circumstances, as unpleasant (and the absence as pleasant). When you're
apprehensive, you feel tense -- literally, you feel your muscles tensing as
if to leap into action. And when the apprehension goes away, you feel
yourself relaxing, which is much more pleasant than the tension was. If the
source of the uncomfortable feeling doesn't happen to be among the
perceptions you're currently conscious of controlling, then the preparation
for action may be the first thing you consciously know about any problem.
And its's subsidence may give you a sense of pleasure without your having
been _aware_ that there was any problem in the first place.

That's a lot of conjecture, of course.

Best,

Bill P.