Domjan et al.

[From Bruce Abbott (2000.09.15.1300 EST)]

Remember when CSGnet was a "scientific forum"? Here's a chance to get back
into the scientific fray.

There's an article now available on the web that should be of interest to us
on CSGnet, as it proposes an application of "control systems theory" for the
purpose of explaining classical (Pavlovian) conditioning, particularly as
such conditioning is involved in "social" behavior. It's not exactly PCT,
so I thought it would be educational to read the target article and critique
it. You won't have to get any further than the title to find something to
criticize; it's entitled "Pavlovian Feed-Forward Mechanisms in the Control
of Social Behavior," by Domjan, Cusato, and Villarreal, and is scheduled to
be published in Behavioral and Brain Sciences, if this hasn't happened
already. The "unedited penultimate draft" can be found at

http://www.cogsci.soton.ac.uk/bbs/Archive/bbs.domjan.html

I'm particular interested in your comments on Section 3, Control Systems
Theory and Pavlovian Feed-Forward Mechanisms, although comments on any part
of the paper would be welcomed. It ought to be fun as Domjan et al. commit
some really glaring errors. Can you spot them?

Bruce A.

[From Bruce Gregory (2000.0915.1745)]

Bruce Abbott (2000.09.15.1300 EST)

I'm particularly interested in your comments on Section 3,
Control Systems
Theory and Pavlovian Feed-Forward Mechanisms, although
comments on any part
of the paper would be welcomed. It ought to be fun as Domjan
et al. commit
some really glaring errors. Can you spot them?

Good God! They understand even less about control than I do! (I'm not
sure whether to be pleased or depressed. Maybe I just haven't reached
bottom yet...)

BG

[From Bruce Abbott (2000.09.15.2005 EST)]

Bruce Gregory (2000.0915.1745)]

Bruce Abbott (2000.09.15.1300 EST)

Good God! They understand even less about control than I do! (I'm not
sure whether to be pleased or depressed. Maybe I just haven't reached
bottom yet...)

Just be careful not to land on top of me when you get there . . .

Remember when CSGnet was a "scientific forum"?

Sure. I remember. It was right up until that article by Domjan,
Cusato, and Villarreal showed up. I made the mistake of printing
it out and ruined thirty perfectly good blank pieces of paper.

I don't sympathize; you could have browsed it on-screen first and saved the
paper. I take it you have some qualms with the Domjan et al. paper? (:->
Care to be more specific? It looks just peachy to me. (;->

Bruce A.

[From Rick Marken (2000.09.15.1740)]

Bruce Abbott (2000.09.15.1300 EST)--

Remember when CSGnet was a "scientific forum"?

Sure. I remember. It was right up until that article by Domjan,
Cusato, and Villarreal showed up. I made the mistake of printing
it out and ruined thirty perfectly good blank pieces of paper.

Best

Rick

···

---

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: marken@mindreadings.com
mindreadings.com

[From Rick Marken (2000.09.15.1900)]

Bruce Abbott (2000.09.15.2005 EST)

I take it you have some qualms with the Domjan et al. paper? (:->
Care to be more specific?

I think the only way to reply to a paper this bad is to simply
hand the authors a copy of B:CP, MR, LCS and LCS II and the
CSGNet archives from 1990.

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: marken@mindreadings.com
mindreadings.com

[From Bill Powers (2000.09.16.1532 MDT)]

Bruce Abbott (2000.09.15.1300 EST)--

I'm particular interested in your comments on Section 3, Control Systems
Theory and Pavlovian Feed-Forward Mechanisms, although comments on any part
of the paper would be welcomed. It ought to be fun as Domjan et al. commit
some really glaring errors. Can you spot them?

It's hard to miss them, control of output being the main one. To me, the
biggest error is not producing any working models to show how the
"predictive" system they imagine would actually work. If they could show
that a system organized as they propose could actually control anything, I
might be a little more interested. But their model consists only of
assertions that control would be the result -- and they don't even say what
they mean by control.

I just don't have the heart to try to open up a debate with Domjan et. al.,
especially in Harnad's autocratic domain. 50 years is enough. You do it.

Best,

Bill P.

[9-16-2000 0945]

Question for Bill, Rick, or Phill, because you were all there.

It was a CSG conference early 90's that somebody gave a paper and
presentation on Pavlovian conditioning and PCT --- I can't remember who it
was -- I recall a Greek sounding name starts with a "P" ----

Mark

[From Bill Powers (2000.09.16.1147 MDT)]

Mark Lazare (2000.09.15) --

Mark, it would be nice if you put a header like mine at the start of your
posts. Makes it easy to see who wrote the post without scrolling to the
end, and in replying we can easily pick up the header to show who and what
we're replying to.

It was a CSG conference early 90's that somebody gave a paper and
presentation on Pavlovian conditioning and PCT --- I can't remember who it
was -- I recall a Greek sounding name starts with a "P" ----

I can only think of Wayne Hershberger. Domjan et. al. refer to
Hershberger's writing on Pavlovian conditioning in his "Conation and
control" book. We did have some discussions of it while Wayne was still
coming to CSG meetings. Could it have been one of Wayne's students?

I believe the upshot of the discussions was that Pavlovian conditioning
starts with a system that is controlling some basic variable. But I'd have
to re-read Hershberger to remember how we worked out the rest.

Best,

Bill P.

Sorry, I don't know. --Phil R.

···

On Sat, 16 Sep 2000, Lazare, Mark Crisis counselor, Phoenix AZ wrote:

somebody gave a paper and presentation on Pavlovian conditioning ...

[from Mark Lazare (2000.09.16.1245 PDT)

Thanks Bill and Phil for the feedback.

Mark

[From Bruce Abbott (2000.09.16.2225 EST)]

Bruce Abbott (2000.09.15.1300 EST)]

I'm particularly interested in your comments on Section 3, Control Systems
Theory and Pavlovian Feed-Forward Mechanisms, although comments on any part
of the paper would be welcomed. It ought to be fun as Domjan et al. commit
some really glaring errors. Can you spot them?

Mark Lazare (2000.09.16.1650) --

Michael Domjan

You wrote ---"Control systems theory developed as a discipline in
engineering, and although it has been used in the analysis of some biological
systems (e.g., McFarland 1971), it has not been extended to social behavior. "

. . . .

I hope in the future, if you use Control Theory in any of your papers, you do
it accurately.

Chewing out Domjan isn't quite what I had in mind. I had hoped that we (on
CSGnet) could go over his paper have a balanced discussion of the proposal,
con _and_ pro (yes, I think it offers some ideas worth thinking about with
an open mind).

Mark, it's not clear to me whether you did much more than read the first
page and check the references before writing to Domjan. Do you have any
idea what phenomena the paper addresses? It's not social behavior per se,
but rather the possible importance of classical conditioning in the
expression of certain social behaviors in particular animal species.
Classical conditioning is being modeled as akin to Wiener's proposal in
_Cybernetics_ in which the input-output relations are modified by a
"compensator" in the light of the system's performance according to some
optimization criterion. It's not PCT, but it is an application of control
theory nevertheless.

I've heard negative evaluative comments about Domjan's paper, but as yet
there's been zero discussion of the actual proposal on CSGnet -- evidently
it is possible to "know" that it is all wrong without having the faintest
idea what the proposal is. That is the attitude of the zealot, not of the
scientist.

Bruce A.

[From Bruce Abbott (2000.09.16.2235 EST)]

Rick Marken (2000.09.16.1650) --

Domjan also, as editor
of JEP: Animal Behavior Processes, summarily rejected
one of my papers on the application of control theory (the
one on Chemotaxis, that I worte with Bill).

I doubt that "summarily" properly characterizes your treatment.
Customarily, the paper would have been reviewed by two independent
reviewers, to which the editor would have added his own comments and
decision with respect to publication. I hope you will share those
reviewers' and editor's comments with us -- I'd like to know what they had
to say.

Bruce A.

[From Bill Powers (2000.09.17.0215 MDT)]

Bruce Abbott (2000.09.16.2225 EST)

Theory and Pavlovian Feed-Forward Mechanisms, although comments on any part
of the paper would be welcomed. It ought to be fun as Domjan et al. commit
some really glaring errors. Can you spot them?

Chewing out Domjan isn't quite what I had in mind. I had hoped that we (on
CSGnet) could go over his paper have a balanced discussion of the proposal,
con _and_ pro (yes, I think it offers some ideas worth thinking about with
an open mind).

What we have to do, in that case, is discuss "classical conditioning,"
because that is really what Domjan's article is about. Nothing he says
about control systems is of any interest, because while he asserts that his
models could accomplish control, he doesn't demonstrate that they can. I
don't believe that any of his proposals about control would actually work.
Of course a demonstration would be worth a thousand "don't believe's_"

The problem is that Domjan seems to have believed in classical conditioning
through all of his professional life, so to him it is an established
reality not to be questioned as an interpretation of behavior. The idea of
explaining it in terms of a more basic model seems not to have occurred to
him. Is such an explanation what you have in mind, or did you just want to
go over Domjan's modeling effort with respect to control theory?

Best,

Bill P.

[From Bruce Gregory (2000.0917.0737)]

Bill Powers (2000.09.17.0215 MDT)

What we have to do, in that case, is discuss "classical conditioning,"
because that is really what Domjan's article is about. Nothing he says
about control systems is of any interest, because while he asserts that

his

models could accomplish control, he doesn't demonstrate that they can. I
don't believe that any of his proposals about control would actually work.
Of course a demonstration would be worth a thousand "don't believe's_"

I think that anyone who proposes a complex system of dubious workability
owes it to his readers to demonstrate that it not only works, but is
superior to simpler systems that undeniably work. This is my reaction to any
proposal that requires elaborate modeling including "cost-benefit" analysis.
Perhaps it can be made to work, but why bother? The chance that natural
systems are based on elaborate modeling seems to require a dramatic
demonstration rather than a simple unlikely claim.

BG

[From Bill Powers (2000.09.17.0634 MDT)]

Bruce Gregory (2000.0917.0737)]

I think that anyone who proposes a complex system of dubious workability
owes it to his readers to demonstrate that it not only works, but is
superior to simpler systems that undeniably work. This is my reaction to any
proposal that requires elaborate modeling including "cost-benefit" analysis.
Perhaps it can be made to work, but why bother? The chance that natural
systems are based on elaborate modeling seems to require a dramatic
demonstration rather than a simple unlikely claim.

Fig. 1 looks something like this, for those who haven't seen it:

Sensory Stimulus- Behavioral
Input ---> Respose ------->Output
(US) Actuator (UR)
                ^ |
                > >
            Comparator<------Monitor (C/B ratio calculator)
                ^
                >
        Cost-Benefit ratio
          instructions

Notice that the behavioral output has no effect on the sensory input, and
that the comparator compares the cost-benefit instructions with a
cost-benefit ratio calculated from monitoring the behavioral output. The
arrow entering the "stimulus-response actuator" from the comparator is
supposed to modify the Unconditional Response to the Unconditional Stimulus
in some manner -- presumably, just the manner that is needed to create the
effect we observe, whatever that is. This clearly is not a workable model.

I don't see how we could have a dialog with Domjan without mentioning his
model, but if we did mention his model it would not be in a way that he
would appreciate. I can't seen any point in discussing his paper with him.
He would just get defensive.

In PCT terms, the US elicits the UR because it disturbs something the
organism is controlling (in most cases I've heard of, it's not hard to
guess what). This was the gist of Hershberger's article. The CS is then
some sensory input that signals the occurrance of the US, so in some way it
creates a situation as if the US had occurred. My guess would be that the
imagination connection is involved, but as we're going outside the
boundaries of what we have modeled, other possibilities exist. If there's
anything interesting about this phenomenon, it would be in the phenomenon
of the organism's being able to treat a sensory input that accompanies a
disturbance as if it were a real disturbance. In complex situations this
would require reasoning power, but there may be a simpler process that has
an equivalent effect in animals that probably can't reason.

Of course if you think everything in behavior is due to classical
conditioning, then you can analyze anything as an example of classical
conditioning. But if your model of classical conditioning doesn't work, so
what?

Best,

Bill P.

[From Bruce Abbott (2000.09.17.1100 EST)]

Bill Powers (2000.09.17.0215 MDT) --

Bruce Abbott (2000.09.16.2225 EST)

Chewing out Domjan isn't quite what I had in mind. I had hoped that we (on
CSGnet) could go over his paper have a balanced discussion of the proposal,
con _and_ pro (yes, I think it offers some ideas worth thinking about with
an open mind).

What we have to do, in that case, is discuss "classical conditioning,"
because that is really what Domjan's article is about.

Yes.

Nothing he says
about control systems is of any interest, because while he asserts that his
models could accomplish control, he doesn't demonstrate that they can. I
don't believe that any of his proposals about control would actually work.
Of course a demonstration would be worth a thousand "don't believe's_"

In the interest of fairness, we need to keep in mind that Domjan et al. use
the term "control" in a wider sense than we do. When folks in Domjan's
field state that some variable "controls" some other variable, they mean
that manipulating the former results in some predictable, systematic change
in the latter, under certain specified conditions. An example of such usage
in ordinary discourse is the statement that the trigger-setting on a
variable-speed drill controls the speed of the drill. Defined in this way,
the term "control" comprises both open-loop and closed-loop systems.

I, too, have serious concerns about the workability of Domjan et al.'s
proposal, but the basic idea seems simple enough: change the input-output
relation of an S-R mechanism until the response to the stimulus is one that
is most beneficial to the organism (according to some criterion). This is
equivalent to clamping a rifle to a stand in a position that aims it toward
a target, firing the rifle, and then adjusting the rifle's aim on the basis
of the difference between target center and bullet hole. After a few
iterations of this, the rifle is accurately aimed and the bullet will hit
dead center every time, without need for further adjustment, so long as
conditions remain constant. Such a scheme might work in a biological system
where disturbances are rarely a significant factor: see steak, squirt so
much saliva into mouth prior to delivery of steak to mouth. Or if you are a
juvenile wolf, see big, bad alpha male, emit submissive gestures. (To avoid
misunderstanding: I would assume that those gestures would be performed by
closed-loop, negative feedback control systems. The S-R mechanism would
produce R by varying the appropriate reference for performing the behaviors
in the presence of S, the alpha male wolf.)

The problem is that Domjan seems to have believed in classical conditioning
through all of his professional life, so to him it is an established
reality not to be questioned as an interpretation of behavior.

I "believe in" classical conditioning too -- as an empirically demonstrated
phenomenon, it is as well established as any scientific observation, as well
establsihed as, for example, the attraction of iron filings to a magnet.
What is in question is not the phenomenon, but its explanation in terms of
mechanism. Domjan et al. have proposed a mechanism, although as we both
know it is rather vaguely stated (e.g., how does this "cost/benefit ratio"
computer work? Magic?).

The idea of
explaining it in terms of a more basic model seems not to have occurred to
him. Is such an explanation what you have in mind, or did you just want to
go over Domjan's modeling effort with respect to control theory?

Well, we've already done some work together on a "more basic model"; we
certainly could try to elaborate such a model further. Tell you what, when
I get a bit more free time I'll go over Domjan's paper line by line and
offer comments where I think appropriate. Perhaps that would be a good way
to begin.

Bruce Gregory (2000.0917.0737)]

I think that anyone who proposes a complex system of dubious workability
owes it to his readers to demonstrate that it not only works, but is
superior to simpler systems that undeniably work.

"Dubious workability" is a judgment for which you have offered no
justification. Opinions are fine, but you really need to explain their
basis in fact or reasoning.

I don't know of any simpler systems that "undeniably work" to account for
all the phenomena of classical conditioning. If you have one to offer, I'm
all ears.

This is my reaction to any
proposal that requires elaborate modeling including "cost-benefit" analysis.
Perhaps it can be made to work, but why bother? The chance that natural
systems are based on elaborate modeling seems to require a dramatic
demonstration rather than a simple unlikely claim.

That's one of my concerns, too, but the "why bother" bothers me. It may not
be well thought out, but at least it provides a starting point for thinking
about what mechanisms might be at work. Do you have a better proposal? Or
do you assume that the problem has already been solved? (Belive me, it hasn't.)

Bruce A.

[Mark Lazare (2000.09.17.09:30 PDT)]

In a message dated 9/16/2000 8:37:14 PM US Mountain Standard Time,
ABBOTT@IPFW.EDU writes:
<< Mark, it's not clear to me whether you did much more than read the first
page and check the references before writing to Domjan. Do you have any
idea what phenomena the paper addresses? It's not social behavior per se,
but rather the possible importance of classical conditioning in the
expression of certain social behaviors in particular animal species.
Classical conditioning is being modeled as akin to Wiener's proposal in
_Cybernetics_ in which the input-output relations are modified by a
"compensator" in the light of the system's performance according to some
optimization criterion. It's not PCT, but it is an application of control
theory nevertheless. >>

...akin to Wiener's proposal .....

Only that Domjan is using some of the same words -- the meaning and the
application is so far off from Control Theory and even farther from PCT it
would not be worth commenting on every sentence or idea he presents. Instead
of trying to reinvent the wheel, he would be better of going to the research
that uses Control Theory and PCT. Plus there is no real study being
offered, the figures of the alleged model would short circuit and crash in a
program written in an attempt to model the figures he included.

http://www.cogsci.soton.ac.uk/bbs/Archive/Figures/domjan.fig1.html

[From Bill Powers (2000.09.17.0634 MDT)]
Fig. 1 looks something like this, for those who haven't seen it:

Sensory Stimulus- Behavioral
Input ---> Respose ------->Output
(US) Actuator (UR)
                ^ |
                > >
            Comparator<------Monitor (C/B ratio calculator)
                ^
                >
        Cost-Benefit ratio
          instructions

Me:
Here is the Text describing the closed loop diagram he conceived of (it is
not like anything you would recognize in PCT or Control Theory, even with an
open mind)

Domjan et al.:
" Figure 1: A closed-loop feedback control system for the regulation of
behavior. Each square represents a separate system component, and the solid
arrows indicate how the components are functionally connected. System input
and output is depicted without squares. Open arrows show the input and output
projections. The stimulus/response actuator translates sensory input
(unconditioned stimuli - US) into behavioral output (unconditioned responses
- UR). The remaining components provide feedback control. They enable the
system to track and evaluate the cost/benefit ratio associated with the
current output response. This information is then used to adjust the
stimulus/response actuator so that future behavioral outputs result in lower
cost/benefit ratios."

Me:
This describes a single closed-loop feedback control system according the
author.

I apologize if I come off like the zealot, not the scientist. I just can't
believe Domjan et al. actually wrote:

"Control systems theory developed as a discipline in engineering, and
although it has been used in the analysis of some biological systems (e.g.,
McFarland 1971), it has not been extended to social behavior. "

Me:
He later cited Hershberger, W. A. (1990). Control theory and learning theory.
American Behavioral Scientist, 34, 55-66, but MISSED THE OTHER 10 PAPERS IN
THAT SAME EDITION. With regard to "social behavior", the papers in that ABS
edition progressed from the individual to dyads to collective action of
groups to Economics.

How could a scientist miss that?
Answer; I don't belive he did miss it but ---
That was not what he was controlling for.....

[From Bill Powers (2000.09.17.0215 MDT)]
The problem is that Domjan seems to have believed in classical conditioning
through all of his professional life, so to him it is an established
reality not to be questioned as an interpretation of behavior. The idea of
explaining it in terms of a more basic model seems not to have occurred to
him.

Mark Lazare

In a message dated 9/17/2000 9:10:12 AM US Mountain Standard Time,
ABBOTT@IPFW.EDU writes:

<< Do you have a better proposal? Or do you assume that the problem has
already been solved? (Belive me, it hasn't.) >>

This Paper was done in detail and Presented at one of the PCT meatings in Co,
early 90's.

I am still Looking for it --- May be Dag Forsell has a copy. I will keep
looking

Mark Lazare

[From Rick Marken (2000.09.17.0900)]

Bruce Abbott (2000.09.16.2235 EST)--

I doubt that "summarily" properly characterizes your treatment.

True. Poor word choice.

I hope you will share those reviewers' and editor's comments
with us -- I'd like to know what they had to say.

I'll check my files tomorrow and give you a report.

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: marken@mindreadings.com
mindreadings.com

[From Bruce Gregory (2000.0917.1329)]

Bruce Abbott (2000.09.17.1100 EST)]

>Bruce Gregory (2000.0917.0737)]

>I think that anyone who proposes a complex system of dubious workability
>owes it to his readers to demonstrate that it not only works, but is
>superior to simpler systems that undeniably work.

"Dubious workability" is a judgment for which you have offered no
justification. Opinions are fine, but you really need to explain their
basis in fact or reasoning.

Like Bill, I don't think the proposed control system will work. True, I
haven't proved it won't work, but I don't think that's my responsibility.
The authors need to _show_ that the system works, not simply postulate that
it works. At least that's the way it' done in the physical sciences.

I don't know of any simpler systems that "undeniably work" to account for
all the phenomena of classical conditioning. If you have one to offer,

I'm

all ears.

What phenomena are you thinking of that can't be modeled with closed loop
control systems? Salivation? Surely that can't be an example. Pupil
dialation? Not likely

>This is my reaction to any
>proposal that requires elaborate modeling including "cost-benefit"

analysis.

>Perhaps it can be made to work, but why bother? The chance that natural
>systems are based on elaborate modeling seems to require a dramatic
>demonstration rather than a simple unlikely claim.

That's one of my concerns, too, but the "why bother" bothers me. It may

not

be well thought out, but at least it provides a starting point for

thinking

about what mechanisms might be at work. Do you have a better proposal?

Or

do you assume that the problem has already been solved? (Belive me, it

hasn't.)

Again, exactly what phenomena are we trying to model?

BG