"Blind men"'s rebuff, Another PCT Tech Report (Long!)

From Tom Bourbon (921212 22:55)

   Too bad your fate was in the hands of people with scientific bents --
they don't have much use for old fashioned fuzzy heads like PCT modelers.
Maybe we should just accept the fact that they already know everything we
try to say -- and then some. They just don't talk about it -- ever --
until they review a PCT manuscript, that is.
   The message is clear for those of us in psychology: If you want to
publish PCT in "real" journals, water it down and turn it into another
"perspective" or "framework," cite everybody they think is important,
and leave all of the serious issues alone. I am happy that sociology
and now education are fields in which at least some of the reviewers
are not so scientific.
   Maybe we can put out a special holiday set of Technical Reports --
two for the price of one, but suitable only for nonscientists.
   Too bad, buddy.
Tom Bourbon

[From Rick Marken (921212.1400)]

Well, as I predicted, the "Blind men and the elephant" paper
has been rejected by the scientifically inclined folks at
Psycholoquy. So I can't even publish in an e-mail journal.
It's pretty depressing; not one reviewer understood what the
paper was about. Virtually every one thought it was a
theoretical paper about control theory. It was not.
Every reviewer thought the ideas were in one way or another
"old hat" or too elementary. There were six reviewers;
only one recommended publication.

I suppose I can either continue as a lunatic fringe PCTer or
take a clue from these reviewers (whose opinions represent the
majority of real scientific psycholgoists) and accept the fact
that my work is horse dung (old horse dung at that) and that
psychology has moved well past whatever PCT might have to
contribute (I'll probably opt for lunatic fringe, of course).

I am attaching the reviews to this post. I know it's kind of a
long document (27K) but probabaly no longer than one of Bill
Powers' better efforts. For those who think that PCT is silly,
old fashioned, unscientific garbage, these reviews should prove
quite entertaining. For those who understand PCT and think it
is the most important scientific insight of the 20th (or any
other) century these reviews should prove quite entertaining.

Even though PCTers (well, at least one -- ME) are not allowed
to have a dialog with the real psychologists on Psycholoquy, at
least we can discuss these reviews amongst ourselves. Then
maybe we can see what we (or they) are missing.

Greg -- Do you think we could put "Blind men" out as a PCT Tech
Report? Along with "Models and their worlds"? Maybe this is a
better idea than trying to publish a journal (of unpublishable

Here they are, with a prelude by S. Harnad.



Dear Rick,

Below are the 6 referee reports on your ms. "The Blind Men and the
Elephant: Three Perspectives on the Phenomenon of Control."

The referees make some useful comments and suggestions for your future
development of these ideas, but the reports unfortunately do not give
me a basis for accepting your paper for publication in PSYCOLOQUY or
for recommending revision and resubmission. The principal problem is
that the ideas are not sufficiently new or specific to form a basis for
peer commentary that would be useful for the cognitive/biobehavioral
field as a whole.

I hope you will find the 6 thoughtful referee reports useful challenges
for fleshing out the predictive, empirical aspects of these ideas.

Thank you for allowing PSYCOLOQUY to consider your ms.

Best wishes,

Stevan Harnad
Editor, Behavioral & Brain Sciences, PSYCOLOQUY

Cognitive Science Laboratory | Laboratoire Cognition et Mouvement
Princeton University | URA CNRS 1166
221 Nassau Street | Universite d'Aix Marseille II
Princeton NJ 08544-2093 | 13388 Marseille cedex 13, France
harnad@princeton.edu | harnad@rrmone.cnrs-mrs.fr
609-921-7771 | 33-91-611-420

REPORT #1: Tom Zentall (ZENTALL@UKCC.uky.edu)

I have now had a chance to look over Richard Marken's ms. "The blind
men and the elephant: Three perspectives on the phenomenon of control".
On the one hand, it is far enough outside of my area of specialization
other hand, I tend to look favorably on any paper that tries to
integrate disparate areas of research. This paper presents a model that
purports to integrate S-R, R-Srf, and cognitive theories of behavior.
This is the kind of paper that asks to be published in an open
commentary format. It is provocative, relatively simple, and tries to
be inclusive. Even if others find it flawed in some way that I cannot
see, I think it will generate interest in the reader. For this reason,
I would recommend its acceptance.

As a general comment, there are many theories that are based on the
principle of an optimal level of stimulation that an organism tries to
maintain (the author might want to reference, e.g., Fiske & Maddi,
1961; Zentall & Zentall, 1983; Zuckerman, 1979).

More specifically, it is probably reasonable to assume, as the author
has, in deriving equation (5) that s* = 0. It seems less reasonable to
assume that "system amplification" approaches infinity as it needs to
to derive equation (4). This may just show my lack of understanding of
control theory, but other readers may need to have this explained.

Finally, although I feel that this submission makes an important
contribution to the journal, I don't feel that my thoughts are
sufficiently developed to write a publishable commentary on this


some interesting questions which are probably worth discussing but I
INTELLIGENT TREATMENT OF THEM. I regret that I cannot be more

The central thesis of this manuscript appears to be that an animal's
behaviour is generated in order to keep certain sensory input variables
at (or at least as near as possible to) preset values (set-points).
This thesis is attributed by the author to Powers (e.g., 1978) and
perhaps this particular way of putting it is peculiar to him though it
AND A RATHER SIMPLE VERSION AT THAT since it restricts the type of
control system used to conceptualize behaviour to simple
servo-mechanisms and homeostats (NO MENTION IS MADE OF OPTIMAL CONTROL
I can see the author is not putting forward an original thesis and it
hardly represents a "new conception of behaviour" - the idea of
behaviour being part of a homeostatic mechanism is older even than
Wiener's cybernetics.

In my opinion, the manuscript's central thesis is a rather technical
way of expressing the "fact" that behaviour is (typically)
goal-directed and that talk of behavioural homeostasis and cybernetics
were, in the past, attempts to deal with goal directedness without
mentioning any teleological terms which were considered dirty words
before about 1970. In effect though, it says that goals are represented
by an animal as sensory set-points - which is typically the way goal
states are represented in servo-mechanisms and homeostats (though we
wouldn't want to say that servo-mechanisms have goal-states, merely
that a servo-mechanism is a tool for achieving a goal state). It is
obvious that an animal engaged in a goal directed activity must have
some way of determining whether the goal has been achieved and it will
typically do this by observing/sensing the state of it's relationship
to the environment or simply the state of this environment (because
goals are typically desired states of this kind). Indeed, I think that
the author's strict adherence to his talk of sensory set points and the
accompanying notation unnecessarily complicates matters since it
distracts one's attention from the simple fact that he is talking about
goal states which in our enlightened times he is perfectly at liberty
to do. It would be very much easier to talk about states of the
environment (which are represented by "sensory input variables") and
goal states (represented by sensory set points) - I'm not convinced
that the author's formalism goes far enough beyond this to warrant its
introduction. Consequently, I'm unconvinced that discovering what an
animal's sensory set-points are is any different from discovering what
its goals are. The latter is none other than a functional analysis of
behaviour which people have been trying to do for a long time.

It seems to me that the author's formalism not only fails to add
anything to a functional analysis in terms of goal states but actually
(1) strongly suggests that the actions which tend to reduce the
difference between the desired state and the current state are of a
single type which differ only in magnitude depending on the size of the
error signal. This is CLEARLY UNTRUE OF animal and human behaviour in
which the action taken to minimize such an error can vary widely -
there are typically lots of means to an end. Note further that the
assumption that the functions in equations (1) and (2) are linear is
ENORMOUSLY RESTRICTIVE; it allows the author to treat k.e, k.f and k.o
as numbers permitting the derivation of all the other equations which
appear in the manuscript. It is completely unclear what would happen if
the assumption of linearity was dropped (as it surely needs to be)
since it is vitally important for equations (3) to (7) which would then
no longer hold. For example, equation (5) cannot be said to establish
the "behavioural illusion" since it relies on the linearity

I failed to understand what the author was getting at in section 3.3.1
the "stimulus-response" view of control. The pupillary reflex, for
example, was one of the first types of behaviour to be subjected to a
control theoretic treatment that the author is advocating and this
treatment is widely accepted - the pupillary reflex is a
servo-mechanism and similar treatments were offered a very long time
ago for other reflexes e.g., the muscle stretch reflex - the idea that
active muscle force, F, is related to the difference between its
stretched length, x, and its unstretched length x* is described by an
equation of the form, F = k(x*-x) is basic stuff. The discussion of
Warren et al. (1986) is very curious - their treatment of the control
of running is basically control theoretic: they argue that the optic
variable tau determines the input to the muscles necessary to achieve
the goal of placing the feet correctly.

I don't really follow section 3.4.1 because the author does not discuss
learning which seems to me to be a central part of reinforcement theory.

I think that the three types of "view" that the author discusses
represent ways of attacking the problem of understanding goal directed
behaviour - we need to understand how goals are selected and how
methods for achieving these goals are arrived at (part of cognitive
psychology); we need to understand how perceptual information is used
to control on- goingbehaviour once that behaviour has been selected as
suitable for achieving the desired goal (the "stimulus-response" style
of the Warren et al. paper discussed by the author); and we need to
understand how the organisms learns what to do in order to achieve its
goals (the domain of reinforcement learning theory discussed by the
AN ORIGINAL CONTRIBUTION. In short, I cannot recommend that this
manuscript be published.

REPORT #3 Eliot Shimoff (shimoff@umbc4.umbc.edu)

Marken's "The Blind Men and the Elephant" seems like an attempt at a
grand unification theory, showing that S-R relations, reinforcement
theory, and cognitive psychology are all aspects a single process (a la
Powers [1978]).

Such attempts are not unprecedented; radical behaviorists have, for
example, often argued that cognitive phenomena can be best viewed from
an operant perspective, and cognitive psychologists have argued that
operant phenomena reflect underlying cognitive processes. Some of these
attempts have proven fruitful, in the sense that they have suggested
experiments or clarified muddy conceptual issues.

A serious attempt at unification must (in my opinion) (a) lead to
interesting NEW EXPERIMENTS, or (b) force CLARIFICATION of some muddy
concepts, or (c) make a SURPRISING PREDICTION (e.g., "If this theory is
correct, you should observe phenomenon X which is not predicted by any
other system").

Part of the problem may be that the VARIABLES ARE NOT PRECISELY
DEFINED. The precise measurement methodology for determining d (the
environmental variable) and r (the response variable) are far from
trivial. One might make a convincing case that the major distinction
between cognitive and radical-behavioral approaches is in the
definition of what counts as behavior (and how it is measured).

Marken suggests (I think) that the proper task of psychology is to
determine k.f, k.o, and k.e. But wouldn't those depend, for example, on
what r is and how it is measured?

Regrettably, the present paper doesn't seem to suggest new experiments,
clarify concepts, or predict any unexpected phenomena. I am not sure
what kinds of comments would be occasioned by its publication. Perhaps
it is all _too_ theoretical (in the pejorative sense); specifying new
experiments, clarifying old concepts, or making novel predictions might
make the paper more suitable for Psycholoquy.

REPORT #4: Ed Fasse (edfasse@Athena.MIT.EDU)

Marken is suggesting that behavior is control. Engineering control
theory has been applied to study of behavior before, particularly in
the study of motor behavior. The idea that certain aspects of behavior
can be understood in terms of control is thus NOT NEW. The proposition
that behavior IS control may be new. In any case the following question
is interesting: To what extent is behavior control. Those are not
Marken's words, but that is the question he is asking as I understand
it. I would add (1) can all human behavior be understood in terms of
control? and (2) even if it can is this useful, i.e. will it lead to
new understanding about function of the nervous system? I have my
doubts about the first question, but even if it is possible to describe
every possible human behavior as a control problem this reformulation
will not always shed insight on processes in the brain. For example, it
may be possible to describe playing chess as a process of controlling
some "sensed aspect of the game" as Marken puts it. I don't think so,
but even if that's the case I doubt that that is an accurate
description of processes in the game. It's not a good description of
how computers play chess.

I think the question "to what extent can human behavior be described in
terms of control" (Not Marken's words, but I think his intent) is
worthy of consideration. I personally feel that to say that behavior IS
control is again saying that an elephant is a snake with a nostril that
eats peanuts.

I am not comfortable with all of the presentation however. Some of what
follows is criticism, the rest is comment. I will try to make the
distinction between the two clear.

1.0 Introduction

1.1 I think that it is valid to say that the goal of control is to
produce consistent results, but I DISAGREE that this always takes
place in the context of an unpredictably changing environment. It is my
understanding that most computer hard disk drives use stepping motors
which are well enough behaved that feedback control is unnecessary.
They operate open loop, and if there is a significant disturbance the
drive will fail unrecoverably. Similarly, the human speech production
system does not have to deal with the amount of environmental
disturbances that the limbs do as the acoustic properties of air don't
change much. That is my opinion. I think it is valid to say that the
goal of control is to produce consistent results in an unpredictably
changing environment. I don't have any problem with section 1.1 as is.

2.0 Closed-Loop Control

2.1 The first sentence, "The basic requirement for control is that an
organism exist in a negative feedback situation with respect to its
environment" is FALSE. Marken later implies that stability requires
negative feedback. This is not true in general. The range of feedback
gains that correspond to stable behavior depend on the underlying
dynamics of the system that is to be controlled. The system to be
controlled may be inherently stable, in which case a little negative
feedback will not destabilize it, and may enhance performance. For
multidimensional, linear feedback it is not the sign of the indivi-
dual feedback gains that count, but the properties of the resultant
feedback matrix. Anyway, it is not true that negative feedback =
stable, although that is often the case.

2.2 The definitions in this section are too VAGUE. Are sensory input
variables raw, unprocessed sensory information? Is this something I
might find in the primary sensory cortex? Or are these perceived
qualities or quantities? Let's say an individual sees a piece of
cheese, and that they decide to pick it up and eat it. Is the sensory
input the light coming from the cheese, the retinal image of the
cheese, the percept that it is an object, the percept that it is a
piece of cheese, or what? Are these quantities defined instan-
taneously or over time? Do they change continuously or discretely? Are
they indeed quantities, i.e. can we assign them numbers? Is a piece of
cheese s = 6 and a piece of cake s = 7? The definitions as is are so
vague that I can see applying them in any number of ways. If they
cannot be made more precise I don't see the utility, and I feel it will
not be possible to make them suffi- ciently precise while describing
all human behavior.

I have similar questions about the response variables. Is a response
something like a set of motor outputs, something I could measure from a
pool of motor neurons? Is it more complex like "man eats cheese"? I
think it is meant to be more complex, but if that's so, how is it going
to be quantified given the mathematical tools that are applied later
on? "Man eats cheese" is r = 3 and "Man throws cheese out window" is r
= 5? It's not clear that the mathematical tools that are used later are

What is an environmental variable? Is it truly extrinsic to the human,
or is it sensed. In section 3.1 Marken says that "The environmental
variable, d, is seen as a stimulus, such as a light or a sound." I
would think that the human only has knowledge of the environment via
the sensory input, s.

I'll pass on k.o, k.e and k.f for now.

2.3. and 2.4: Equation (1) is a purely static model, as are the other
equations. These equations are not dynamic, that is they cannot make
predictions about evolution of the system with time. In section 2.1
Marken comments on stability, yet these models have no concept of
stable or unstable behavior. It might be useful to assume this for sake
of argument for equation (1) at times, but not for equation (2).
Equation (2) is a description of how sensory input changes depending on
responses and the environment. This relation will in general be
dynamic. r,s and d will be related by differential equations in the
continuous case, not by static, algebraic equations such as (2).

2.5-2.6: Equations (3) and (4) are static constraint relations. They
are not causal statements for r. Equation (1) is the causal statement
for r. Marken speaks of sensory energy and response energy. Does he
really mean thermodynamic energy, or does he mean energy-like
quantities as is often done in engineering? For example, the square of
magnitude is often related to energy, so it is sometimes loosely
referred to as energy. Saying that a little s lead to a big r does not
necessarily imply that a small sensory energy leads to a large response
energy. It depends on the energy functions. Also, saying that k.o can
be "quite a large number" is not in general meaningful. Here one can
argue that it is okay since Marken required that s and r be in the same
units, but in engineering control one usually refers to the loop gain
rather than the feedback gain as the relevant quantity. After all, what
is a large number? 0.1 metric tons/mm? 1000 Newtons/light year?

The energy discussion seems irrelevant. Marken wants justification to
let k.o approach infinity. His point is that as feedback gains increase
the steady state (static) behavior of the controlled system becomes
less dependent on the dynamics of the system which is being controlled.
This makes the controlled system behavior less susceptible to changing
dynamics. Fine. I would just say that and forget about the energy
argument, which I feel is misleading. If it is important I would
develop it better.

I do not like the last paragraph of section 2.6. Equation (4) is not a
causal statement, and thus cannot be used to show that the organism
acts to keep its sensory input constant by varying responses to
compensate for variations in the environment. The equation is correct
given Marken's assumptions, but the inferrence about behavior is not
valid. r = k.o (s*-s) is the causal statement. One could say "IT IS
AS IF the organism had access to information about the environment, d,
and that it was changing its responses based on this information.
However, information about the environment is only available via
sensory information, so the actual control law is r = function of

I am not happy with section 2. Control theory has been applied to
understanding behavior before, notably in motor control. I do not
feel that this presentation is particularly adept. It did not teach me
anything I didn't know before. Again, what I see as interesting is not
"Can aspects of behavior be understood in terms of control?". Yes, they
can. It's been done before.
What I see as interesting is "Can ALL behavior be understood in terms of

3.0 Three Views of Control

3.1-3.2: Again, equation (4) is not causal and should not be used for
making inferences. Equation (1) says stimulus causes response. Equation
(4) says that the response is compatible with the environmental
variable d. What does Marken mean to say that the environmental
variable is seen as a stimulus. How is the human aware of the
environment except via the sensory information s? Or is d sensory
information too?

"The value of s* can be constant or variable, it's value at any instant
being secularly determined by properties of the organism itself." A
change in s* has an observable change in behavior. How is it changing?
What processes are changing it? These processes are affecting behavior,
and since we are trying to model behavior, are we not then concerned
with modelling the change of s*? Is s* the response of some other
control process? Some planning process? s* affects behavior as much as
anything else, and it seems that it is being swept under the rug. How
can we model the process that generates s*?

3.3.1: I cannot really comment on this section as I am not familiar
with the references.

3.3.2: I think of a stimulus as being a sensory or perhaps perceptual
event. Marken seems to think of a stimulus as an external event. If
this is true, then it makes sense to speak of d as a stimulus. From my
perspective to say that d is a stimulus is nonsensical. From my
perspective the control law, equation (1) implied stimulus-response,
not equation (5).

3.3.3: I think the statement of this section is too strong and
misleading. Feedback makes the particular dynamics of the system that
is being controlled less visible. It does not make them invisible. I
would change "negative feedback" to "feedback", as the fact that the
feedback is negative is irrelevant. I have no idea what Marken means by
a "behavioral illusion" as I am not familiar with the reference. I
would like more explanation.

3.4.1: I can't comment on this section.

3.4.2: "The sensory effect of a reinforcement can be assumed to be
directly propor- tional to its size and weight making k.e = 1." That's
just plain wrong. s = k.f (r) + k.e (d) is already linear with respect
to d. Assuming d is some parameter of size this expression already says
that the sensory effect is directly proportional to d. That constant of
proportionality is k.e, it certainly does not have to equal 1. Or else
I'm missing something here in a big way, but I don't think so.

3.4.3: Well, that's logically consistent, but don't think that's
actually what's going on.

3.5.1: Okay, but what's the point. Is this an introduction for 3.5.2?

3.5.2: But what is generating s*? Presumably some cognitive process or
another controller. The process that is generating s* has as much
influence on behavior as the controller. What is this process? One
cannot say control explains behavior until one explains the generation
of s*, which affects behavior.

4.0 Looking at the Whole Elephant

This section is fine. It is logically consistent with the rest of the
paper. The statements in it are contestable. "Organisms behave in order
to keep sensory inputs at these reference values (Powers, 1989)." That
is debatable, but that's the kind of statement that would be good to
discuss by the community at large.

5.0 Appendix

I think the derivation in this appendix is sufficiently obvious that it
does not warrent inclusion. It is correct.


So, I think the question "can ALL behavior be described as the result
of a control process" (not a quote, but I don't have italics) is
interesting. I feel the theoretical, mathematical analysis of the paper
is WEAK, and NOT PARTICULARLY ORIGINAL. I think it would be best to get
rid of most of it and focus on formulating definitions of the concepts
introduced in section 2.2 that are as precise as possible. I also think
it would be useful to spend more time on the process that generates s*,
rather than the VAGUE statement that "it's value is at any instant
being secularly determined by properties of the organism itself".
That statement doesn't tell me anything.


I've taken several long looks at [Marken's] paper on control and
feedback and don't feel I have anything helpful to the author. It
reminds me of the sort of off-the-wall discussing we used to do late in
the evening after studying for comps. There's been so much specific
research since then, I would have expected such speculations to take a
new tack.

But that's just my reaction. Other readers, who didn't have the
benefits of my fellow graduate students might profit from being
stimulated by such a paper.

REPORT #6: Vasant Honavar (honavar@iastate.edu)

The author raises a number of interesting issues concerning the the
central role of control as a determinant of behavior. I am not in a
position to evaluate the author's suggestion that psychological
studies of behavior ignore issues of control for the most part. Perhaps
other reviewers will address this issue. I will restrict my comments to
some general suggestions on how the paper could be strengthened by a
more thorough consideration of the principle thesis put forth by the

Cybernetics (the science of control and communication in humans and
machines), starting with the pioneering work of Wiener in US and
Lyapunov in the former USSR during the late fifties very much
recognized the importance of studying control. Indeed Lyapunov
characterized the evolution of living nature (and hence one might
argue, intelligent behavior) as a process of proliferation of, and
selection among, control systems of various types (see Honavar & Uhr,
1990 -- Coordination and Control Structures and Processes:
Possibilities for Connectionist Networks; Journal of Experimental and
Theoretical Artificial Intelligence 2 277-302 for a discussion of this
and a number of related iss> ues in the context of artificial and
natural intelligent systems). Similar considerations have motivated
contemporary work on models of intelligent behavior within the
dynamical systems framework.

I am sympathetic to the author's view that the phenomenon of control
needs to be studied in its totality. However, I would like to see a
STRONGER AND MORE COMPLETE ARGUMENT developed -- perhaps using specific
examples of psychological phenomena that can be better understood and
explained (than is permitted by current approaches in psychology) by
taking a holistic approach to the study of control. This is just one
suggestion and I am sure there are other ways to strengthen the paper.
I recommend that the paper be resubmitted for publication in a revised
form. If the paper is eventually accepted for publication in
Psycoloquy, I will be more than happy to write a commentary.