Commitment

CSGnet Archive CSG1999
1

[From Bruce Abbott (990330.1110 EST)]

Rick Marken (990330.0730) --

That would put CVs into the same class as god, the tooth fairy
and Bruce Abbott's commitment to PCT.

You're right about that last part. My commitment is to science; I have no
comitment to PCT. PCT is a _theory_ of behavior, not a faith, not some
ideal or belief system one should swear allegiance to, as if one were some
"defender of the faith." PCT as a theory of behavior has merit and deserves
serious attention, but it should never merit "commitment," except perhaps by
its author, who has some stake in whether or not it eventually gets
accepted by the larger scientific community.

Uncommittedly yours,

Bruce

2

[From Rick Marken (990330.1250)]

Bruce Abbott (990330.1110 EST) --

My commitment is to science; I have no comitment to PCT.
PCT is a _theory_ of behavior, not a faith, not some
ideal or belief system one should swear allegiance to, as if
one were some defender of the faith."

PCT is more than a theory. It is a scientific approach to
understanding behavior. When I say that you are not committed to
PCT I mean that you are not committed to a scientific approach
to understanding behavior. The conventional "scientific" approach
to understanding behavior _assumes_ that behavior is caused (or
selected) output. This is the _basic model_ of behavior in "scientific"
psychology. Yet this model is never tested or even
treated as a testable model. Instead, this model is enshrined in
the very methods that are used to study behavior; methods that
are called "scientific" but have little to do with science. PCT
scientists have done many tests of the basic model of behavior
used in "scientific" psychology and shown that it is the _wrong_
model when organisms exist in a closed loop relationship with
respect to their sensory inputs (which means that it is _always_
the wrong model).

A person who is committed to science is willing to revise their
model of a phenomenon based on evidence. The evidence shows that
the basic model of behavior used in "scientific" psychology is
wrong; behavior is not caused (or selected) output; rather,
behavior is controlled input. The fact that you are unwilling
to reject that basic model of behavior used in "scientific"
psychology (as evidenced by the contents of the latest edition
of your behavioral methods text, for example) shows that you are
not committed to a scientific approach to understanding behavior;
rather, you are committed to the _basic model_ of behavior used
in "scientific" psychology. Your commitment to this model is so
deep that you can't even see that this model _is_ a model.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

3

[From Bruce Abbott (990330.2240 EST)]

Rick Marken (990330.1250)

PCT is more than a theory. It is a scientific approach to
understanding behavior.

No, it's a scientific theory. No less. No more. It posits a particular
organization of the nervous system that may or may not be correct in various
ways.

When I say that you are not committed to
PCT I mean that you are not committed to a scientific approach
to understanding behavior.

How do you know? And if that's what you meant, why didn't you say precisely
that, instead of beating around the bush?

The conventional "scientific" approach
to understanding behavior _assumes_ that behavior is caused (or
selected) output. This is the _basic model_ of behavior in "scientific"
psychology.

So your assertion comes down to saying that I am committed to an open-loop
analysis of behavior, whether warranted or not. I would say that I am
"committed" to employing whichever analysis seems most appropriate in a
given case, given the empirical evidence.

You, on the other hand, are (in my view) "committed" to PCT, which means
that you will move mountains, if necessary, to make the available evidence
appear to fit the theory. It means that you will be hyper-critical of any
empirical result that is incompatible with the theory, to the point of
denying it any scientific merit whatsoever, while immediately accepting any
result that appears to offer support. That sort of partisanship is not my
idea of science, but hey -- to each his own.

Bruce

4

[From Rick Marken (990331.0940)]

Bruce Abbott (990330.2240 EST) --

No, it's a scientific theory. No less. No more. It
posits a particular organization of the nervous system that
may or may not be correct in various ways.

It also very explicitly includes a description of how effects
produced by the nervous system affect the inputs to that system.
PCT correctly represents the behaving system as two _simultaneous_
equations, one characterizing the nervous system and the
other characterizing the environment in which that system operates.

So your assertion comes down to saying that I am committed to
an open-loop analysis of behavior, whether warranted or not.

You are committed to an open loop _model_ of living systems.
The analysis techniques you advocate (the one's described in
your methods textbook) are based on that commitment, which is
not warranted because you know that open-loop (including
sequential state) models of behavior are wrong.

I would say that I am "committed" to employing whichever
analysis seems most appropriate in a given case, given the
empirical evidence.

You have never used (or published research based on) an analysis
appropriate to studying the behavior of a closed- loop control
organization. You don't even explain how to perform such an
analysis in your textbook on behavioral research methods. So
you are clearly _not_ " 'committed' to employing whichever
analysis seems most appropriate in a given case, given the
empirical evidence". You are committed to the open-loop model
of behavior that is the basis of conventional psychology. Either
that, or you are just committed to having a textbook that sells.

You, on the other hand, are (in my view) "committed" to PCT,
which means that you will move mountains, if necessary, to make
the available evidence appear to fit the theory.

Fortunately, I don't have to. The available evidence fits the
theory to within a percentage point.

It means that you will be hyper-critical of any empirical result
that is incompatible with the theory

I certainly would be surprised to find an empirical result that
is incompatible with a control of input model of behavior. Do
you know of any. I've been asking for such evidence for years;
I've never seen any.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

5

[From Bruce Abbott (990331.1425 EST)]

Rick Marken (990331.0940) --

Bruce Abbott (990330.2240 EST)

No, it's a scientific theory. No less. No more. It
posits a particular organization of the nervous system that
may or may not be correct in various ways.

It also very explicitly includes a description of how effects
produced by the nervous system affect the inputs to that system.

I haven't seen that. Under PCT, how do effects produced by the nervous
system affect the inputs to that system? (Specific proposal please!)

How outputs affect inputs falls out of the physical organization of the
system; it is not an additional property of the theory. Furthermore, the
proper analysis of such a system is not a spcific proposal of PCT (which
merely posits that the system in question is a control system) but of
general control theory, for which the appropriate analytic tools were
developed. PCT is just an application of general control theory to a
specific proposed control structure.

PCT correctly represents the behaving system as two _simultaneous_
equations, one characterizing the nervous system and the
other characterizing the environment in which that system operates.

PCT correctly represents the behaving system in this way when the behaving
system can be correctly described in this way. When the system operates as
an open loop system, or in some other fashion (e.g., an equilibrium system),
it does not.

So your assertion comes down to saying that I am committed to
an open-loop analysis of behavior, whether warranted or not.

You are committed to an open loop _model_ of living systems.
The analysis techniques you advocate (the one's described in
your methods textbook) are based on that commitment, which is
not warranted because you know that open-loop (including
sequential state) models of behavior are wrong.

Amazing how you know me better than I know myself. I didn't have the
faintest idea! But I shouldn't be surprised, as you frequently claim to be
a mind reader. I think you really believe you are.

I would say that I am "committed" to employing whichever
analysis seems most appropriate in a given case, given the
empirical evidence.

You have never used (or published research based on) an analysis
appropriate to studying the behavior of a closed- loop control
organization.

Please review my extensive posts to CSGnet over the past four years,
especially those in which I post both computer models of control systems
and, where the study was empirical, data that the model has been fit to.
That should quickly disillusion you of the notion that I have "never used an
analysis appropriate to studying the behavior of a closed-loop control
organization."

As for publication, I have yet to find data that are well fit by a
control-system analysis, and some that appear to contradict one. You want
me to publish _those_?

You, on the other hand, are (in my view) "committed" to PCT,
which means that you will move mountains, if necessary, to make
the available evidence appear to fit the theory.

Fortunately, I don't have to. The available evidence fits the
theory to within a percentage point.

Your "available evidence" shows that when participants are placed in a
situation in which they must control something in order to perform the
required task, that they control it, and do it very well, thank you. This
is hardly surprising and it is no miracle that control-system equations do a
fine job of fitting model to data.

It means that you will be hyper-critical of any empirical result
that is incompatible with the theory

Here's a nice, well-documented example that Bill P. provided of this sort of
response (you do the same, but this one comes to mind). Tinbergen and
Lorenz observed that if the egg of a graylag goose rolls out of the nest and
stops within reach of the goose that is incubating the eggs, she will stand,
reach out with her bill, place the bill on the side of the egg opposite her,
and use the bill to roll the egg back into the nest. The crucial
observation is that if the experimenter contrives to snatch the egg suddenly
away while the goose is in the process of retrieving it, she continues as if
nothing happens, pulling the bill along the ground toward her and tucking
the now imaginary egg between her legs before sitting down. This is an
observation by reliable observers, but it doesn't fit the PCT view of
things, so Bill's response was to deny the validity of the observation. As
I said, if the facts don't agree with the theory, dispute the facts, not the
theory.

I certainly would be surprised to find an empirical result that
is incompatible with a control of input model of behavior. Do
you know of any. I've been asking for such evidence for years;
I've never seen any.

There was this long-standing belief, based on PCT, that increasing the
response requirement on a reinforcement schedule would result in a
compensatory increase in response rate. I think you went so far as to state
that this clear prediction of PCT provided a critical test of PCT against
reinforcement theory, which predicted the opposite result (decreased rate of
responding with increasing response requirement). I ran the required study
and reinforcement theory won. The overwhelming response from the PCT gurus
was "oh." Nothing more came of it. However, it turns out that the
experiment provided a crucial test only for reinforcement theory. PCT can
account for the result (after the fact) no matter which way it turned out.

Regards,

Bruce

6

[From Bruce Gregory (990331.1512 EST)]

Bruce Abbott (990331.1425 EST)

There was this long-standing belief, based on PCT, that increasing the
response requirement on a reinforcement schedule would result in a
compensatory increase in response rate. I think you went so
far as to state
that this clear prediction of PCT provided a critical test of
PCT against
reinforcement theory, which predicted the opposite result
(decreased rate of
responding with increasing response requirement). I ran the
required study
and reinforcement theory won. The overwhelming response from
the PCT gurus
was "oh." Nothing more came of it. However, it turns out that the
experiment provided a crucial test only for reinforcement
theory. PCT can
account for the result (after the fact) no matter which way
it turned out.

This touches on the point I raised in my last post. It is not clear, to
me at least, than any evidence _could_ count against PCT as long as we
have the "out" that we have simply failed to identify the variable being
controlled. There is little doubt that control data can we well fit by a
control of input model. There is much greater doubt that _all_ behavior
can be explained by a control of input model. Since so few data have
been modeled this way, it is more of an article of faith. Perhaps that's
why Ken is so fond of PCT.

Bruce Gregory

7

[From Bill Powers (990401.1448 MST)]

Bruce Abbott (990331.1425 EST)--

There was this long-standing belief, based on PCT, that increasing the
response requirement on a reinforcement schedule would result in a
compensatory increase in response rate. I think you went so far as to state
that this clear prediction of PCT provided a critical test of PCT against
reinforcement theory, which predicted the opposite result (decreased rate of
responding with increasing response requirement). I ran the required study
and reinforcement theory won.

I'm sorry, but reinforcement theory lost, too. There was NO EFFECT. The
apparent effect was an artifact, which you discovered yourself. The reason
the "long-standing belief" existed in PCT was that I had been assured that
rats were capable of varying their bar-pressing rate, and I predicted the
effect to which you refer on that basis, and on the fact that increasing
reinforcement rate _apparently_ (from the uninterpreted data) went with
decreasing behavior. The prediction was wrong because the facts as
presented by researchers (you and others) were wrong. In fact, the rats who
produced these data either pressed as fast as they could or engaged in
behaviors other than bar-pressing.

The overwhelming response from the PCT gurus
was "oh." Nothing more came of it.

That is a lie. I retracted the control model before you even caught on that
there was something wrong with it. What you are doing here, Abbott, is
disgusting.

···
8

[From Bill Powers (990401.15012 MST)]

Bruce Gregory (990331.1512 EST)--

It is not clear, to
me at least, than any evidence _could_ count against PCT as long as we
have the "out" that we have simply failed to identify the variable being
controlled.

If you conclude that you have failed to identify a variable that is under
control by some behavior you want to explain, then you have to say that PCT
has not been shown to apply to that behavior and you can't explain it with
PCT.
I don't see your problem. In cases where you can identify all the
properties of a control system and the variable it is controlling,
naturally you can conclude that PCT does explain the observed behavior.

I don't quite grasp what you're driving at. Does it seem to you that PCT is
a general statement that every observable behavior of anything is control
behavior? If so, then a single instance of a non-control behavior would
prove that belief wrong. But to conclude from that that NO behavior is
control behavior, it seems to me, would be equally incorrect, especially
since we have formal ways of defining control behavior and testing for its
presence.

PCT is a theory that explains how control by organisms works. Once you've
established that control is going on, you turn to PCT for ways to explain
it. Or, perhaps, you turn to model-based control theory, or open-loop
control theory, or reinforcement theory, although you already know why I
wouldn't do that.

There is little doubt that control data can we well fit by a
control of input model. There is much greater doubt that _all_ behavior
can be explained by a control of input model.

Of course. But the only way to vindicate, or allay, this doubt is to design
the experiments and carry them out. There's no shortcut that I know of.

Best,

Bill P.

9

[From Bruce Gregory (990331.1736 EST)]

Bill Powers (990401.15012 MST)

I don't quite grasp what you're driving at. Does it seem to
you that PCT is
a general statement that every observable behavior of
anything is control
behavior?

Yes.

If so, then a single instance of a non-control
behavior would
prove that belief wrong.

Perhaps if I had a few examples of non-control behavior I could make the
distinction. My impression comes from Rick's insistence that behavior
_is_ the control of perception. Should I take it that you disagree with
him?>

But to conclude from that that NO behavior is
control behavior, it seems to me, would be equally incorrect,
especially
since we have formal ways of defining control behavior and
testing for its
presence.

I agree completely.

PCT is a theory that explains how control by organisms works.
Once you've
established that control is going on, you turn to PCT for
ways to explain
it.

Again, I agree completely.

Of course. But the only way to vindicate, or allay, this
doubt is to design
the experiments and carry them out. There's no shortcut that
I know of.

I hope Rick agrees with you.

Bruce Gregory

10

[From Rick Marken (990331.1500)]

Bruce Abbott (990331.1425 EST)-

Under PCT, how do effects produced by the nervous system affect
the inputs to that system?

Via the environmental feedback function and the perceptual input
function.

How outputs affect inputs falls out of the physical organization
of the system; it is not an additional property of the theory.

I disagree. The entire loop must be part of a correct model of
a closed loop system. Failure to properly include feedback effects
is the reason for the failure of "open loop" (cause-effect) models
of behavior. Failure to correctly _identify_ the correspondence
between loop variables in a living and an artificial control system
is one reason for the failure of non-PCT control theory models of
behavior.

PCT is just an application of general control theory to a specific
proposed control structure.

Yes. It's the _correct_ application.

When the system operates as an open loop system, or in some other
fashion (e.g., an equilibrium system), it does not.

Right. So far there is no evidence that living systems _ever_
operate as open loop systems. Indeed, how could they operate
open loop; the sensory organs are connected to appendages (head,
arms, legs, etc) that are caused to move by what is sensed by
these organs. The feedback connection between output and input
is strong, immediate and inexorable; it is as real a part of
living systems as the sensory organs and the body itself.

Me:

You have never used (or published research based on) an analysis
appropriate to studying the behavior of a closed- loop control
organization.

Bruce:

Please review my extensive posts to CSGnet over the past four
years

No thanks. Just send me some publications where you describe
studies of controlled variables (by that name).

As for publication, I have yet to find data that are well fit
by a control-system analysis

See _Mind Readings_ and collect some data at

http://home.earthlink.net/~rmarken/ControlDemo/Cause.html

or

http://home.earthlink.net/~rmarken/ControlDemo/Levels.html

and see how nicely the model fits it.

and some that appear to contradict one. You want me to publish
_those_?

You bet!! That would be extremely important!!

Tinbergen and Lorenz observed that if the egg of a graylag goose
rolls out of the nest and stops within reach of the goose that is
incubating the eggs, she will stand, reach out with her bill,
place the bill on the side of the egg opposite her, and use the
bill to roll the egg back into the nest. The crucial observation
is that if the experimenter contrives to snatch the egg suddenly
away while the goose is in the process of retrieving it, she
continues as if nothing happens, pulling the bill along the
ground toward her and tucking the now imaginary egg between
her legs before sitting down. This is an observation by reliable
observers, but it doesn't fit the PCT view of things, so Bill's
response was to deny the validity of the observation. As I said,
if the facts don't agree with the theory, dispute the facts, not
the theory.

There is not enough evidence here to tell whether or not the
"facts fit the theory". I suppose you take this observation to
mean that the goose's "retrieving behavior" is performed "open
loop". That may be. PCT suggests that you should first try to
determine whether or not this "retrieving behavior" is a side
effect of control of some perceptual variable or is an aspect
of a controlled perceptual variable itself. You have to do the
Test to find out. If the "retrieving behavior" is being produced
open loop then it will fail every test to determine whether it
is (or is a side effect of controlling) a controlled variable.

Me:

I certainly would be surprised to find an empirical result that
is incompatible with a control of input model of behavior. Do
you know of any.

Bruce:

There was this long-standing belief, based on PCT, that increasing
the response requirement on a reinforcement schedule would result
in a compensatory increase in response rate.

That would be true only if 1) reinforcement rate were a controlled
variable 2) the organism could influence this variable only by
varying response rate and 3) the organism _could_ successfully
control reinforcement rate by varying its reinforcement rate.

I ran the required study and reinforcement theory won.

Really? That is _big_ news. I think you should publish this
study in JPCT! That's an _incredible_ result. Maybe you could
publish a preview here on CSGNet?

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

11

[From Rick Marken (990331.1530)]

Bill Powers (990401.15012 MST)--

But the only way to vindicate, or allay, this doubt [about
whether or not control is involved in a particular behavior]
is to design the experiments and carry them out. There's no
shortcut that I know of.

Bruce Gregory (990331.1736 EST) --

I hope Rick agrees with you.

Are you kidding? I've been doing the experimental tests and
examining the evidence of control in behavior. You've been
sitting in your armchair wondering "whether there is any evidence
that _could_ count against PCT".

Adn I agree with Bill about that retina thing, too:-)

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

12

[From Bruce Gregory (990331.2002 EST)]

Rick Marken (990331.1500)

There is not enough evidence here to tell whether or not the
"facts fit the theory". I suppose you take this observation to
mean that the goose's "retrieving behavior" is performed "open
loop". That may be. PCT suggests that you should first try to
determine whether or not this "retrieving behavior" is a side
effect of control of some perceptual variable or is an aspect
of a controlled perceptual variable itself. You have to do the
Test to find out. If the "retrieving behavior" is being produced
open loop then it will fail every test to determine whether it
is (or is a side effect of controlling) a controlled variable.

This is an excellent example of what I was afraid of. Since there will
always be _some_ test that we have not performed, the possibility of a CV
can _never_ be eliminated. No test can rule out the possibility that this is
an example of control.

Bruce Gregory

13

[From Bruce Gregory (990331.2005)]

Rick Marken (990331.1530)]

Are you kidding? I've been doing the experimental tests and
examining the evidence of control in behavior. You've been
sitting in your armchair wondering "whether there is any evidence
that _could_ count against PCT".

Sorry I don't meet with your approval. Can you answer one simple question?
Is behavior the control of perception? Is there no behavior that is not the
result of the exercise of control? Thanks.

Bruce Gregory

14

[From Rick Marken (990331.1900)]

Bruce Gregory (990331.2005)--

Can you answer one simple question?

Why is tonight different from all other nights? :wink:

Is behavior the control of perception?

The behavior of living systems, yes.

Is there no behavior that is not the result of the exercise
of control? Thanks.

There are surely some behavioral variables that are not controlled;
the downward acceleration of your body as you fall from a roof is
one. But when you are dealing with a system (such as a living system)
where what occurs at the input transducers is a direct result of
actions caused by what occurs at those transducers then you are
probably dealing with a control system.

PCT is designed to explain _control_. It's highly unlikely
that you will be able to show that any controlling is done
"open loop"; as we have shown, control is _necessarily_ a
closed loop process. So I don't think anyone will ever be
able to "disprove" PCT in the sense of showing that some kind
of controlling is an open loop process. I think what will be
disproved about PCT are specific assumptions of the model; the
regular, hierarchical structure; the assumption that specific
_types_ of perceptions are controlled at each level; the assumption
that reorganization is random, etc. Those are the aspects of
the model that will be disproved by experimentation.

I think the point of _basic_ PCT is that organisms _are_
control systems. They may not be structured exactly as
assumed in B:CP. But I think we have done enough basic research
to show that _no_ open loop model can possibly explain the
behavior of living systems.

Best

Rick

···

---
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

15

[From Bill Powers (990331`.2054 MST)]

Bruce Gregory (990331.2002 EST)--

This is an excellent example of what I was afraid of. Since there will
always be _some_ test that we have not performed, the possibility of a CV
can _never_ be eliminated. No test can rule out the possibility that this is
an example of control.

You misunderstand the purpose of the Test. The Test is meant to rule out
variables that are not under control by an organism. To show that a
variable which passes the test IS under control, you have to match a model
to experimental data, and show that the model predicts the actions of the
organism and the variations in the controlled variable over some period of
time and to some degree of accuracy. How long and with what accuracy is a
judgment one must make -- generally we pick the theory that predicts over
the longest time with the greatest accuracy. In the present state of PCT,
we don't ask very much -- one minute of time, and 10% RMS accuracy as about
as well as we can do. But that's better than any other theory could do with
the same data.

Best,

Bill P.

16

[From Bill Powers (990331.1943 MST)] back to March for a little longer.

Bruce Gregory (990331.1736 EST)--

I don't quite grasp what you're driving at. Does it seem to
you that PCT is
a general statement that every observable behavior of
anything is control
behavior?

Yes.

That is a misinterpretation of the term "theory" as it is used in PCT. The
theory in PCT is described by the equations and diagrams that lay out the
relationships among variables in a negative feedback control system. All
systems whose organization is correctly described by these equations and
diagrams are by definition control systems; the equations and diagrams are
the theory that explains how they work. This use of "theory" is the same as
in "electronic theory", the theory that explains how electronic circuits
work (for example, Ohm's Law), or "the theory of the moon" which is about 6
pages of dynamical equations from which the future motions of the moon can
be calculated (and the present motions can be understood). Control theory
is the theory that explains how control systems work. PCT is the theory
that explains how control systems that rely on perceptions work.

Perhaps if I had a few examples of non-control behavior I could make the
distinction.

The behavior of a doorbell is non-control behavior. When the button is
pressed, the doorbell rings, but the ringing has no effect on the button.
The behavior of a mountain stream is non-control behavior: it is pulled
downward by gravity and constrained by the local terrain; if the terrain
wears away and changes the stream's path, the stream does nothing that
would restore it to the former path. If I leap up very fast to answer the
phone, I may knock my cup of coffee to the floor. Knocking the cup of
coffee to the floor is non-control behavior. You can prove this by
preventing the cup from hitting the floor: I will not change my action to
counteract your disturbance and make the cup hit the floor anyway.

In general, non-control behavior would be the production of actions which
have effects on the environment, but when those effects are disturbed the
action does not change to oppose the disturbance.

My impression comes from Rick's insistence that behavior

_is_ the control of perception. Should I take it that you disagree with
him?>

No.

Best,

Bill Powers

The behavior of control systems can be described, in a very shorthand sort
of way, as the visible (to others) part of the process by which the control
system controls its own sensory representation of the outside world. Thus,
in even briefer form, the behavior of control systems is the control of
their perceptions.

So if you observe an organism acting as a control system, you can say its
behavior is the control of its perceptions. If you can't tell whether it's
acting as a control system, you can't say that for sure -- although it's
reasonable to assume it given that in every case you have carefully
observed, you were able to identify the control process.

17

[From Bruce Gregory (990401.0541 EST)]

Rick Marken (990331.1900)]

PCT is designed to explain _control_. It's highly unlikely
that you will be able to show that any controlling is done
"open loop"; as we have shown, control is _necessarily_ a
closed loop process.

It is not highly unlikely, in that case. It is impossible.

I think the point of _basic_ PCT is that organisms _are_
control systems. They may not be structured exactly as
assumed in B:CP. But I think we have done enough basic research
to show that _no_ open loop model can possibly explain the
behavior of living systems.

I'm not quite as persuaded as you are. I do, however, agree that closed-loop
models are much simpler and do the job much better. I don't think anyone who
understands this would prefer an open-loop explanation. My take on PCT is
that it tells us that organisms should be treated as control systems until
we have solid evidence that shows otherwise. With this I am whole-hearted
agreement.

Bruce Gregory

18

[From Bruce Gregory (990401.0522 EST)]

Bill Powers (990331.1943 MST) back to March for a little longer.

Control theory
is the theory that explains how control systems work. PCT is the theory
that explains how control systems that rely on perceptions work.

I was unclear. I was referring to the idea that all behavior is controlled.

>Perhaps if I had a few examples of non-control behavior I could make the
>distinction.

If I leap up very fast to answer the
phone, I may knock my cup of coffee to the floor. Knocking the cup of
coffee to the floor is non-control behavior. You can prove this by
preventing the cup from hitting the floor: I will not change my action to
counteract your disturbance and make the cup hit the floor anyway.

Your behavior was controlled, but not by your perception of the cup. Then
the behavior Bruce A. describes of the gray goose _is_ controlled behavior
but not control of a perception we have identified,.

Bruce Gregory

19

[From Bill Powers (990401.0539 MST)]

Rick Marken (990331.1900)--

PCT is designed to explain _control_. It's highly unlikely
that you will be able to show that any controlling is done
"open loop"; as we have shown, control is _necessarily_ a
closed loop process.

Those who propose open-loop control models always forget one loop that is
present even in a so-called open-loop control system: the loop through the
person who adjusts the system for proper operation (or through the adaptive
process that does the same thing). Somehow, somewhere, the results of the
system's operation must be compared with the desired results, and the
output part of the system must be adjusted to correct the error, if any. An
open-loop "control system" is really just the output function of the actual
control system. Of course it's interesting that we can invent output
functions that require only occasional action to keep them producing the
effects we want -- as in the case of a digital wristwatch which we employ
to tell us the time. I have one that cost about $15 and requires adjustment
only once a year to stay within 15 seconds of the right time (a fluke, I'm
sure). Navigation-wise, that's about four miles of longitude at the equator
(placing anything easily within eyeshot of a person on the deck of a ship).

Going through this exercise is useful, because it reminds us that
hierarchical control does not happen just through varying reference
signals. Higher systems can also act on the parameters of lower systems,
changing their gains, their limits, and who knows what else. Higher systems
can thus do everything that other modelers attribute to such processes as
the Extended Kalman Filter. A higher system that adjusts parameters of
lower systems would monitor perceptions having to do with the detailed
performance of the lower systems -- for example, the stability of the lower
system, or its output sensitivity to error signals.

There us certainly enough to keep us occupied in the concept of control
through variation of reference signals. However, the world will not wait
for us to finish that job -- there are always those who scream "WHAT ABOUT
ADAPTATION?"
Well, _this_ is what about adaptation, but I leave that subject to others.

Best,

Bill P.

20

[From Bill Powers (990401.0435 MST)]

Bruce Gregory (990401.0522 EST)--

If I leap up very fast to answer the
phone, I may knock my cup of coffee to the floor. Knocking the cup of
coffee to the floor is non-control behavior. You can prove this by
preventing the cup from hitting the floor: I will not change my action to
counteract your disturbance and make the cup hit the floor anyway.

Your behavior was controlled, but not by your perception of the cup.

I think you'd better say what you mean by "your behavior." I was not
controlling my perception of the cup, but my actions affected my perception
of the cup (and the cup itself, we assume).

Behavior is not controlled by perception in any case I know of. Did you say
what you meant to say? Perception is not varied as a means of bringing
behavior to any preselected state, although very often we can see behavior
being varied to bring perception to a preselected state.

Then
the behavior Bruce A. describes of the gray goose _is_ controlled behavior
but not control of a perception we have identified.

I can see saying "control behavior," meaning actions being used to control
something. But I do wish we could drop this ambiguous term "controlled
behavior" or "control of behavior." Every time you say that, it would take
a paragraph to straighten out what you mean -- control of the action itself
at a lower level, or control of the results of the action at the same or a
higher level. Often, when we refer to someone's behavior we mean the
actions we can see someone else performing, and perhaps some of the
immediate environmental effects of the actions. But at least as often, we
mean the consequence of a set of varying actions, such as "scratching
behavior." It's much clearer if we just describe what we see happening,
keeping actions separated from their consequences at the same level.

The goose performs actions of extending its neck, lowering its bill, and
retracting the bill in such a way that if there were a rounded object under
it, the object would be rolled or dragged toward the goose. Those movements
were apparently regular enough that we might supposed they were under
control by the goose, or so we have been told (I've never seen this
phenomenon myself, or even a movie of it).

At another level, we might also suppose, given the circumstances, that
there was an intention at a higher level to bring an object (originally an
egg) back toward the goose. However, the geometry is such that the goose
might not be able to see the egg once the bill is over it, so the pulling
back toward the nest is done blindly. This conjecture is supported by the
fact (as reported here, but see below) that the experimenter was able to
remove the egg or substitute a smooth stone without any objection from the
goose. What we don't know is how well the goose can feel the object it is
pulling with its horny bill. If it can't feel the object very well, or at
all, it must rely on controlling the position and motion of its bill, which
it clearly can control.

I wonder what I would do, if I were the goose, if I reached out for the egg
I saw rolling away, and encountered thin air. I think I would trying
sweeping my bill inward several times, hoping to find the egg by feel. If I
were rather dim-witted (bird-brained?), I might give up and settle back on
the nest, assuming I had succeeded. An observer might misinterpret the
search for the egg as an automatic "response" executed without regard to
its success. It all depends on the perceptual bias of the observer.

I'm having some doubts about the description of Tinbergen's experiment. As
I recall it, a darning egg was substituted for the real egg, and the goose
pulled the darning egg in and sat on it. That would be much simpler to
explain, of course. But maybe both experiments were done. Mary's off trying
to find the original description in our limited library. We may have to
wait until we can get to a real library, unless someone out there can find
the original source of the story. There are lots of apocryphal stories
floating around, even among scientists! It's always easier to make a point
if you can invent an appropriate experimental result (like rats pressing
bars at varying rates).

I think that experimental results get stale, so the experiments need to be
done again after some lapse of time, say 5 years. Aside from the advantages
of replication under the eyes of a different experimenter, this is a good
way to correct for the way facts tend to drift.

Best,

Bill P.