providing feedback; Peter Burke's models

[From Bill Powers (960610.1030 MDT)]

Jeff Vancouver 960610.10:30 EST --

Bill:

What you seem to be saying here is that some error exists, but in the
case of interest that the system has no way of knowing that the error
has been reduced until told by an external agent.

Jeff:
     Information that results in a reduction in error is what you (and
     me) would call negative feedback, but if that information comes
     from an external agent, it is called positive feedback in our
     literature (because the agent is able to say "yes, that's it" or
     something like that).

OK, I see. "Positive" feedback is good, affirming, supporting, etc.,
while negative feedback is the opposite. That's not what _you_ say, but
what the psychologists you're studying say.

     The point is that I set up a situation in which the focal system
     will detect a difference between the current state and the desired
     state, the system will be allowed to act, and the system will be
     allowed to get new information about the current state, following
     that act. That some external system was involved in providing that
     information should be of little interest to understanding the
     behavior of the focal system (as I have so far manipulated the
     situation).

So in this situation, the focal system can't perceive the current state
unless given information by the external agent? I think I need to see an
example of this; it seems like an unusual situation.

But is this learning? Here the only thing that prevents you from
producing the right act is a lack of information about the effects of
the two levers.

     I do not agree. An output function looks different after the
     learning you describe has taken place. It is your word, but is not
     changes to functions reorganization?

I see what you mean. Before the information is received, the focal
system doesn't know which output to produce; afterward, it does know. My
reservation about calling this reorganization, however, comes from the
fact that the system doesn't learn to produce any output it couldn't
produce before. The system could have pressed either lever -- that is,
it already knew how to press levers. What was missing was a basis for
choosing a lever to press -- a perception, not an action.

Clearly, it is not the subsystem that actually reaches toward a lever
and presses it that is changed. What is changed is the reference
position given to the reaching/pressing system. In fact, before the
information is given, there is no reference position that has been
selected -- the higher-level system that would provide it doesn't know
which reference signal to produce.

It seems to me that this process doesn't involve any new skills. If you
already have a general method for converting verbal inputs into their
equivalent perceptions, and of selecting a perception as a reference
state, and if the higher goal is to do what the words suggest, then that
enables the process we are talking about to happen without learning any
new control skill. The situation would be different if you didn't
already know the meanings of the words, or didn't recognize the
communication as advice about what to do, or were unable to match the
reference perception (press the left lever) with an actual perceptual
result. Then you would have to reorganize, to learn how to do something
new. Anyway, that's my take on this.

     I am still working on the simulation. It would facilitate
     discussion if you could see this. Once I do get it working, it
     requires Windows 3.1 or higher (and would be better Zipped). Do
     you have these programs?

Yes, I have Windows 3.1 and PKzip and unzip. I take it that you would
send both source code and executable code. I can read the source code
but don't have Visual Basic to compile it. Maybe if you just posted the
source code that would get us somewhere. Don't worry if it's crude or
unfinished -- I'm used to that in my own programs.

     I find I cannot follow the PASCAL programs you all post.

It will be interesting to see what the problem is. Maybe reading your
code will tell me something about it.

···

-----------------------------------------------------------------------
Peter Burke (xxx) --

Peter, I received your two papers Saturday, and have read them both with
delight (and thanks to Chuck Tucker for the preview he posted!). You're
actually doing something I have only fantasized about doing. This is
very good PCT and a great start toward PCT modeling in sociology. I'm
particularly impressed by your ability to relate your model to others
that have been proposed, and in the case of the operant conditioning
model to compare the results in terms of model predictions. That is the
Right Stuff. You will no doubt be accused of setting up a straw man, but
from my point of view the results are unequivocal.

Most particularly, your approach of matching a model's behavior to the
dynamic behavior of real people is exactly in the PCT tradition. I can
imagine the thrill you got when you first got your simulation to behave
right -- righter than you had any right to expect.

Somewhere down the line I expect to see a PCT economics come out of your
work. When you start varying the amount of the exchanges as well as
their occurrance, you will have a big first step. Then (as I expect you
are already thinking) you can start bringing in more goals beside just
the goal of participating in an exchange. That's more or less what I was
fantasizing about -- but I never got as far as producing an actual
simulation.

By the way, I think I sent you my source code for a bargaining program.
This program sets up two people bargaining for some unnamed commodity
that one of them wants to sell, with each one having a minimum price
received or maximum price paid in mind, and controlling a perceived
current value of the commodity that is a simple function of the previous
bid and asked prices. The outcome is either a done deal if the final bid
price is greater than or equal to the final asked price, or no deal if
this condition can't be reached. Of course as part of your system, this
bargaining algorithm would be supplied with reference bid and asked
prices that are adjusted on the basis of experience in past rounds of
bargaining. I don't know if you want to include this level of detail in
your program, but, as Gilbert and Sullivan said, it might lend an air of
verisimilutude to an otherwise bald and unconvincing tale.

I really do like what you're doing.
-----------------------------------------------------------------------
Best to all,

Bill P.