Working Models, Predicting Behavior

[From Rick Marken (2003.11.28.0910)]

Marc Abrams (2003.11.28.0904)

David Goldstein:

> You say that we need a "working model of HPCT.

Marc Abrams:

_YES_ (Not shouting, just really emphasizing)

David:

> What do you have in mind?

Marc:

Nothing specific and concrete at this point, just _a lot_ of ideas. For
instance, I believe we need to bring emotion and memory into the model.

In my paper _Looking at behavior through control theory glasses_
(Review of General Psychology, 6, 260�270, 2002) I describe several
working models of behavior. Several are hierarchical control models.
None of them explicitly incorporates memory or emotion but all produce
behavior that is very similar to the behavior they were designed to
imitate (catching fly balls, balancing brooms, pointing at targets,
moving through crowds, etc.). Does this mean that they are not really
working HPCT models, from your point of view?

David:

> How will you recognize that we have a working model of HPCT?

Marc:

When the model can produce the predicted results of the theory

How about when the model can produce the actual results that are
observed?

My spreadsheet hierarchy model certainly produces the predicted results
of hierarchical PCT: successful simultaneous control of several
perceptions of the same _type_ by systems at one level of of the
hierarchy by systems that achieve these perceptions by varying lower
outputs that have direct effects on the environment or that determine
the references for lower level systems.

Other models, such as the simple integrated output with transport lag
model of tracking, predict actual behavioral results. A simple
one-level control model nearly perfectly predicts behavior in a
polarity reversal experiment. Try the "Levels of Control" demo at
http://www.mindreadings.com/ControlDemo/Levels.html. When you complete
a run, the filled squares plot the behavior of a one-level model, which
is a prediction of how you will behave when the polarity of the
connection between mouse and handle suddenly changes. The open squares
are your actual behavior. If the plot of your behavior matches the
behavior of the model the model has successfully predicted your
behavior. This is a true prediction because the model has no idea what
you are going to do when the polarity changes. The model can also
predict your "adaptation" to the polarity change that occurs about 1/2
second after the change but this requires adding another level of
control to the model.

I am frankly amazed that your complaints about PCT (or HPCT) are about
it being neither a "working model" nor "predictive". Maybe I've spent
too many years being a lapdog but it seems overwhelmingly clear to me
that my master has developed a theory of behavior that is not only
implemented as a working model but that predicts behavior with
exquisite accuracy.

Best regards

Rick

···

---
Richard S. Marken
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400

from [Marc Abrams (2003.11.28.1619)]

[From Rick Marken (2003.11.28.0910)]

In my paper _Looking at behavior through control theory glasses_
(Review of General Psychology, 6, 260�270, 2002) I describe several
working models of behavior. Several are hierarchical control models.
None of them explicitly incorporates memory or emotion but all produce
behavior that is very similar to the behavior they were designed to
imitate (catching fly balls, balancing brooms, pointing at targets,
moving through crowds, etc.). Does this mean that they are not really
working HPCT models, from your point of view?

A 'hierarchal control model' is _NOT_ necessarily a working model of HPCT.
It certainly could be, but having a 2 to 3 level hierarchy and claiming it
validates HPCT in my mind is quite a stretch. _That_ is my point, and my
_only_ point

I am frankly amazed that your complaints about PCT (or HPCT) are about
it being neither a "working model" nor "predictive". Maybe I've spent
too many years being a lapdog but it seems overwhelmingly clear to me
that my master has developed a theory of behavior that is not only
implemented as a working model but that predicts behavior with
exquisite accuracy.

I'm glad your happy and satisfied.

Marc

[From Bruce Gregory (2003.1128.1632)]

Marc Abrams (2003.11.28.1619)

A 'hierarchal control model' is _NOT_ necessarily a working model of
HPCT.
It certainly could be, but having a 2 to 3 level hierarchy and
claiming it
validates HPCT in my mind is quite a stretch. _That_ is my point, and
my
_only_ point

O.K. I think I understand where you are coming from. Perhaps we should
distinguish hPCT from HPCT. The former is a hierarchical model based on
PCT. The latter is the theory Bill described in B:CP and has since
modified somewhat. There are hPCT working models. There are no HPCT
working models.

Bruce Gregory

from [Marc Abrams 2003.11.28.1920)]

[From Bruce Gregory (2003.1128.1632)]

O.K. I think I understand where you are coming from. Perhaps we should
distinguish hPCT from HPCT. The former is a hierarchical model based on
PCT. The latter is the theory Bill described in B:CP and has since
modified somewhat. There are hPCT working models. There are no HPCT
working models.

Bullseye.

Thank you Bruce.

Marc

[David Goldstein (11,29.2003.0637)]
[Marc Abrams 2003.11.28.1920)]

Marc,

If all you meant was the distinction of hPCT versus HPCT, then you have
called a lot of attention to something that most of us would have agreed
with you from the start, if you said it that way.

Maybe the real question is the following. What are the barriers to
studying the hierarchy?

How does one determine that two experiences are at different levels?
This issue becomes very important when one tries to do the MOL. It is
not so easy to judge this, as anyone who has tried the MOL knows. Not
all of a person's comments about a topic A are "going up a level." Some
are at the same level. Some are at a lower level and some are at a
higher level.

How does one determine that one experience is necessary for another
experience to take place? Even if an experience B is at a higher level,
does experience A evoke experience B as a component evokes a whole? Or
are we talking about an arbitrary memory association thing?

Have to go Marc,
David
David M. Goldstein, Ph.D.

···

-----Original Message-----
From: Control Systems Group Network (CSGnet)
[mailto:CSGNET@listserv.uiuc.edu] On Behalf Of Marc Abrams
Sent: Friday, November 28, 2003 7:22 PM
To: CSGNET@listserv.uiuc.edu
Subject: Re: Working Models, Predicting Behavior

From [Marc Abrams 2003.11.28.1920)]

[From Bruce Gregory (2003.1128.1632)]

O.K. I think I understand where you are coming from. Perhaps we should

distinguish hPCT from HPCT. The former is a hierarchical model based
on PCT. The latter is the theory Bill described in B:CP and has since
modified somewhat. There are hPCT working models. There are no HPCT
working models.

Bullseye.

Thank you Bruce.

Marc

from [Marc Abrams (2003.11.29.1044)]

[David Goldstein (11,29.2003.0637)]

[Marc Abrams 2003.11.28.1920)]

Marc,

If all you meant was the distinction of hPCT versus HPCT, then you have
called a lot of attention to something that most of us would have agreed
with you from the start, if you said it that way.

David, I tried saying it any number of ways. When you saw Bruce Gregory's
post and then mentally reviewed what I had been trying to say it made all
the sense in the world to you. I think _why_ it was not perceived that way
is both interesting and instructive.

Things were fine until Bill perceived me as trying to mock and tear him
down. Everyone else picked up on that and instead of _reading_ my posts and
asking for clarifications, the focus moved off the topic and onto my
intentions, sincerity, and general character.

Our exchange was a nice example of that. Your questions were challenges to
me, not for the purposes of further clarification for you. If they were for
clarification our conversation would not have ended with my last post and
question of "How do you improve something without talking about it?" _Why_
and _how_ I was saying something became the focus instead of _what_ I was
saying.

hPCT is the "PCT illusion" or "HPCT illusion" if you will. It was never
intended to be a derogatory statement and I said so. I specifically said it
was a useful construct which included the MOL

Maybe the real question is the following. What are the barriers to
studying the hierarchy?

That very well might be the right question. On the other hand it might not.
Maybe the current hierarchy is not
the appropriate one or maybe there are others involved. I think the real
question now is how do we _TEST_ the hierarchy.

How does one determine that two experiences are at different levels?

Great question

This issue becomes very important when one tries to do the MOL. It is
not so easy to judge this, as anyone who has tried the MOL knows. Not
all of a person's comments about a topic A are "going up a level." Some
are at the same level. Some are at a lower level and some are at a
higher level.

I have no answers but a possiblity _might_ be that we are dealing with a
_network_ and not a _pure_ hierarchy here and what you are describing are
not neccesarily different levels but merely different 'states' or
'configured patterns' your CNS might be in. Please note my use of CNS
(Central Nervous System) There is a great deal more involved in our behavior
than our brain and it _does_ matter. More than even the CNS but I use that
to shorthand the need to incorporate a more global view.

How does one determine that one experience is necessary for another
experience to take place?

_THAT_ is one of my major questions with regard to the hierarchy. Can we
actually say that _all_ perceptions are 'built' in that fashion. It is not
simply a matter of whether all perceptions _have_ those properties. The
hierarchy _demands_ that they are, in fact, built in that very specific way.

Even if an experience B is at a higher level,
does experience A evoke experience B as a component evokes a whole? Or
are we talking about an arbitrary memory association thing?

All great questions.

Great post David, thanks.

Marc