Domjan et al.

[Bruce_Gregory]

[From Bruce Gregory (2000.0917.1336)]

Bill Powers (2000.09.17.0634 MDT)

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.

We have virtually ubiquitous artificial control systems (thermostatically
controlled furnaces, cruise controls) that have no need to monitor their
outputs. I am baffled by the ability of researchers to ignore these
proofs-of-concept and continue to build unworkable models based on output
control. The only explanation that I have come up with is that the
thermostat seems to "control" the furnace by turning it on and off. But even
then it needs some way to know that it has been successful, not only in
issuing a command, but in seeing that the command has had the desired
effect. It seems obvious, but it obviously isn't, that in order to control
some variable, the system must have a way of monitoring the variable and not
simply its own actions. This is particularly the case since it is trivially
easy to test. Don't let the thermostat know what the temperature is and see
what happens. Since no one bothers to do this test, I infer that very very
few even understand what control is. I'm uncomfortable with this conclusion,
but I can't see any way around it.

BG

[Richard_Marken2]

[From Rick Marken (2000.09.17.1015)]

Bruce Abbott (2000.09.17.1100 EST)--

I "believe in" classical conditioning too -- as an empirically
demonstrated phenomenon

This is why it's not worth going through the agony, Bruce.
If you want to persue this Domjan model thing then go for it.
But from a PCT perspective, Domjan missed the point (of
behavior) before he even conceived his ill-conceived model.

Best

Rick

···

--

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

[Bill_Powers1]

[From Bill Powers (2000.09.17.1138 MDT)]

Bruce Abbott (2000.09.17.1100 EST)--

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).

I dispute your claim that there are any actual S-R mechanisms in the
_overall_ behavior of organisms (I say _overall_ to eliminate input-output
mechanisms that make up components of the system, such as sensory
nerve-endings and muscles). This is not just contentiousness -- I am laying
down a challenge, to be met, if you want to, in the usual scientific way. I
will not dispute a claim that there are _appearances_ that suggest such
mechanisms, but my claim is that these appearances are misleading: they are
interpreted to mean that one kind of mechanism exists, when (I claim) an
entirely different kind of mechanism actually exists.

You know what my proposal is, but let me spell it out anyway. An apparent
S-R mechanism is suggested when some environmental event takes place,
reliably followed by some action by the organism. The appearance is that
the environmental event is sensed so it gives rise to neural signals, and
that the neural signals then trigger off the action. Because the action
appears to be caused by the prior event, it is called a "response," and the
prior event that caused it is called a "stimulus."

The alternative is to say that the prior event is not itself sensed, but
that as a physical occurrance it tends to alter the state of some other
variable or variables that the organism is already controlling. Assuming a
constant reference signal, this disturbance gives rise to an error signal.
The error signal produces an action that has effects on the controlled
perception opposite to the effect of the prior event, thus limiting the
change in the controlled perception to a small fraction of what it would
have been without control. The result is that the disturbance largely
_fails_ to affect the variable it disturbes.

Under either interpretation, a prior event leads to a predictable action: a
"stimulus" leads to a "response." But if the control-system interpretation
is correct, there is no "stimulus-response mechanism." Instead, there is a
control system. Furthermore, the control system is acting continuously, so
that whether the disturbance be brief or continuing, the action will be the
appropriate one: a brief opposition to a brief disturbance, or a
continuously varying opposition to the effects of a continuously-varying
disturbance. The same control mechanism accounts for either kind of
response with no change in its organization.

···

===================================================================
Now let's consider your examples.

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.

What you describe here is a control system controlling the distance between
the bullet hole and the target, in a series of trials. The error signal is
converted into an adjustment of the aim of the gun. The actual pointing
direction is the cumulative sum of all successive angular adjustments, so
the output function is an integrator.

If no disturbances occur, the error can be brought to zero, so no further
adjustments will be needed. However, the likelihood of that is very small,
as a simulation containing a small amount of noise would show. Anyway,
because this is a control system rather than an S-R system, it will easily
adjust to changes in wind direction, miscalibration of the sights,
inaccurate clamping of the gun, and so on. A fixed environment is not
required for it to work.

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.

So why isn't this a control system, too? We might guess that the salivation
is an act that makes chewing and swallowing easier, so it's part of a
complex of control actions that is called "eating." It's not the sight of
the steak that causes salivation, but the intention to eat. The intention
to eat is carried out by a series of actions in a particular order, with
salivation being one of the early actions. One way to test this would be to
see if satiety turns off the salivary "response to the sight of steak."
After a big steak dinner, will you still salivate if a new sizzling steak
is set before you? Or will you throw up?

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 highest-level closed-loop control system is the one that produces
submissive gestures as a means of controlling the behavior of the alpha
male. The gestures are not a response to the presence of the alpha male,
but an action taken to control the aggressiveness of the alpha male.

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.

That's a particularly apt simile for my purposes. What we appear to see is
that a magnet attracts iron filings to itself. But this simple observation
is not correct; there is no such thing as a one-way "attraction." The iron
filings attract the magnet exactly as powerfully as the magnet attracts
them. A much better way to express the situation would be to say that a
force develops between the magnet and the filings, such that both are urged
toward a common center. So the "observation" conceals an interpretation
that we unconsciously apply.

In a very similar way, we can see that a moth is "attracted" by a light at
night. But this "observation" is even more strongly influenced by our
interpretation of what we see happening. In fact the light exerts no force
on the moth at all, nor does the moth exert any force on the light. The
motions of the moth are created entirely by the moth's wings (and any
random breezes), and the moth approaches the light because it has a control
system for increasing the proximity of the light.

What is in question is not the phenomenon, but its explanation in terms of
mechanism.

I couldn't disagree more, as my comments above should bear out. Your
observations of the phenomenon are strongly influenced by what you believe
its mechanism to be. This is particularly true in the behavioral sciences,
where behavior is reported not as a simple series of coordinate changes,
but in terms of causes, effects, and functions. In your example above, you
illustrate this beautifully: you think you are observing submissive
gestures as a response to the mere presence of the alpha wolf: see wolf,
emit gestures. I'm sure you group the events in that way in your own mind,
so that you think you're observing the gestures as a response to the
stimulus of seeing the wolf. But you're not reporting an observation:
you're interpreting events. My interpretation of exactly the same events is
very different: I see a control system acting to have a particular effect
on the alpha male. See alpha male acting hostile, want to see alpha male
acting less hostile, produce submissive (or placatory) gestures until the
goal is accomplished. Of course if the male is not acting hostile, no error
is detected and no such gestures are needed.

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.

OK. Keep in mind what my comments are likely to be.

Best,

Bill P.

[Bruce_Gregory]

[From Bruce Gregory (2000.0917.1941)]

Bill Powers (2000.09.17.1138 MDT)]

That's a particularly apt simile for my purposes. What we appear to see is
that a magnet attracts iron filings to itself. But this simple observation
is not correct; there is no such thing as a one-way "attraction." The iron
filings attract the magnet exactly as powerfully as the magnet attracts
them. A much better way to express the situation would be to say that a
force develops between the magnet and the filings, such that both are

urged

toward a common center. So the "observation" conceals an interpretation
that we unconsciously apply,

Boy have you got that right. The notion that the baseball attracts the earth
with a force equal to the force exerted by the entire earth on the baseball
is so outlandish that most students never entertain it. The memorize
Newton's third law, but they are so steeped in cause and effect thinking
that they never see its implications.

BG

[Abbott_Bruce]

[From Bruce Abbott (2000.09.17.2000 EST)]

Just got home and have time only for a brief reply to this:

Bruce Gregory (2000.0917.1941) --

Bill Powers (2000.09.17.1138 MDT)

That's a particularly apt simile for my purposes. What we appear to see is
that a magnet attracts iron filings to itself. But this simple observation
is not correct; there is no such thing as a one-way "attraction." The iron
filings attract the magnet exactly as powerfully as the magnet attracts
them. A much better way to express the situation would be to say that a
force develops between the magnet and the filings, such that both are
urged
toward a common center. So the "observation" conceals an interpretation
that we unconsciously apply,

Boy have you got that right. . . .

What we see is that the magnet is doing the attracting -- the filings will
not attract another pile of filings. And what we see in classical
conditioning is that a particular stimulus triggers a particular response,
under certain physiological conditions. I am well aware of the possibility
that the observed phenomenon may be produced by a control system, for which
the "stimulus" is a disturbance to the system's controlled variable and the
"response" is a change in output that serves to counter the effect of the
disturbance on that controlled variable. Or it could be that the
relationship is mediated by an open-loop mechanism. The observance of the
relationship is an observation; the explanation in terms of mechanism is an
interpretation. See the difference?

Of course, which interpretation is correct can be decided by appropriate tests.

Bruce A.

[Richard_Marken2]

[From Rick Marken (2000.09.17.1900)]

Bruce Abbott (2000.09.17.2000 EST)

Of course, which interpretation is correct can be decided by
appropriate tests.

Of course. So why waste time trying to build a control model
before you know whether or not you are dealing with a
control phenomenon?

The fact that the food-salivation relationship is part of
the process of controlling some variable, such as food texture,
(and I would be quite surprised if it were not) could have been
determined, by psychologists who study classical conditioning,
back in the 1970s, when Powers explained how to do the test for
controlled variables. But such tests were not done so we have
no idea which "interpretation", S-R or control, is correct.

Best

Rick

···

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

[Bill_Powers1]

[From Bill Powers (2000.09.18.1244 MDT)]

Bruce Abbott (2000.09.17.2000 EST)--

What we see is that the magnet is doing the attracting -- the filings will
not attract another pile of filings.

Nor will the magnet attract a pile of aluminum filings. "Attract" is simply
not the right word to use in reporting a physical process; it's a layman's
word, a word that reflects a vague theory as outdated as Aristotle.

And what we see in classical
conditioning is that a particular stimulus triggers a particular response,
under certain physiological conditions.

We've been around on this before. You don't observe any "triggering";
that's a model of an unseen mechanism. Can't you think of a way of
expressing the observations that doesn't favor _either_ S-R theory _or_
control theory? To say that a stimulus triggers a response is to assert a
theory, a model, in the very act of reporting what happened. It puts a spin
on the report. We do not observe either stimuli triggering responses, or
actions that correct the effects of disturbances. What we observe are
events or variations in the environment, and behaviors or actions created
by the muscles (or other motive equipment) of an organism. The whole
question is what mechanism relates the first observation to the second. I
was trying to show, in my post, how the same observations could equally
well be described in control-theory terms. I didn't mean to imply that that
was a correct way to report observations, either. I was trying to show a
different bias and how it could influence the interpretation of the actual,
atheoretic, observations. The idea was to show, by parallel, that your
description also contained a bias.

Once you have proposed a model, it becomes possible to test it: if the
model is correct and well-formed, you can predict its behavior under
different conditions, and see whether the prediction is born out. For
example, if the relation between the antecent and consequent events is one
of "triggering," then we should find that the response is independent of
the magnitude of the stimulus before and after the moment of triggering. We
should also find that there is no effect of the response on the triggering
effect of the stimulus. Of course you can state your own definition of what
you mean by "triggering;" I don't mean to put words in your mouth.

On the other hand, if you embed the model in the report of the observation,
what is there to test? You simply defend the observation as self-evident,
and the model with it.

I am well aware of the possibility
that the observed phenomenon may be produced by a control system, for which
the "stimulus" is a disturbance to the system's controlled variable and the
"response" is a change in output that serves to counter the effect of the
disturbance on that controlled variable. Or it could be that the
relationship is mediated by an open-loop mechanism. The observance of the
relationship is an observation; the explanation in terms of mechanism is an
interpretation. See the difference?

Of course I do. I'm trying to get you to see that _either_ way of
describing the situation is unwarranted except in terms of a model.
Choosing a model is not just a matter of preference or optional
interpretations. If one model is right, the other is wrong. It's
conceivable that both are wrong, but they can't both be right. We need to
start with a description of the phenomenon that does not inherently favor
one model or the other; only then can we separate what we actually observe
from what our mental models lead us to read into the observations. When you
say you observe a stimulus triggering a response, you are favoring the S-R
model from the beginning; in fact, you're assuming it.

Is this discussion going to founder on this point as other discussions have
done before?

Of course, which interpretation is correct can be decided by appropriate

tests.

Yes. So let's start by describing the observations of so-called classical
conditioning (there's another model for you) in a way that does not imply
any mechanism. Only in that way can we see what mechanism is being offered
in opposition to control theory. Then, as you say, we can devise tests
appropriate to each proposed mechanism.

Shall we start again?

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2000.09.18.0130 MDT)]

Bruce Gregory (2000.0917.1336)--

Since no one bothers to do this test, I infer that very very
few even understand what control is. I'm uncomfortable with this conclusion,
but I can't see any way around it.

I'm uncomfortable, too. You should read that thing by Lewis:
http://www.theorem.net/lewis1.html

He says," Following Friedland [1986], we may call the period from 1868 to
the early 1900's the primitive period of automatic control. It is standard
to call the period from then until 1960 the classical period, and the
period from 1960 through present times the modern period."

While he discounts modern control theory as a model of organisms, he does
present it as a new era in control; in fact he says that it made space
flight possible, which I think is utter nonsense. Most control systems that
fly in space are classical control systems.

I think he is confusing mathematical analysis with physical design of
control systems. I don't find the mathematics of "classical" control theory
any more user-friendly than that of "modern" control theory, but that never
stopped me from designing successful control systems. The BIG difference
between classical and modern control theory is not in the mathematics, but
in the physical design, the "architecture." The classical control system is
the one we diagram in PCT. It can be designed mathematically for optimum
operation in certain restricted situations, but it can also be designed by
simulation with a bare minimum of mathematics or simply by cut and try. The
"modern" control system, physically, consists of a huge array of sensors, a
computer that can provide an exact model of the external world and do
extremely complex calculations about it, and a library of information about
physical processes. It is designed to compute the action necessary to bring
about a preselected result, and then execute it. It could _never_ be made
to work by cut-and-try. That is an enormous difference in physical design,
not to mention in the mathematics.

In fact the physical design of modern control theory is one that existed
_before_ classical control theory. It's simply the Sherrington model of the
brain, in which top-down commands are elaborated from level to level until
the final common pathway is reached. What modern control theory does is
show the actual implications of that model: they are horrendous, and should
suffice to show that Sherrington's model is completely impracticable.

If modern mathematical methods were applied to the input-comparator-output
model, I'm sure that many improvements in mathematical design could be
produced. The hierarchical model offers even more improvements which were
never to my knowlege tried during the "classical era." But the real
question is the basic architecture: is the modern approach really an
improvement, or is it simply a way of thought that has been given elaborate
development for no good reason? Is it, in fact, a way of holding on to
old-fashioned cause-effect thinking?

I am terribly handicapped here, and I'm sure you feel somewhat the same
way. My suspicion is that modern control theory was put together by people
who never did understand intuitively how the PCT-type model worked. I keep
running across statements to the effect that modern control theory can
accomplish everything that a classical control system can do, and more --
but examples are missing and what comparisons there are seem irrelevant.
And there are other tell-tale signs, such as the idea that modern control
systems are multiple-input-multiple-output (MIMO) systems, whereas
classical control systems are single-input-single-output (SISO) systems.
Who ever thought that a control system was an input-output system in the
first place? And who says a classical system can't be a MIMO system, or at
least control multiple variables? I keep running across little things like
that that make me really wonder how much these guys know about control.
Could I have invented a whole new way of thinking about control that
_neither_ the classical nor the modern control theorists have worked out?
It hardly seems likely, but I always felt that the classical texts were
missing the main point, too. The _engineers_, however, were probably a
different story.

But back to the handicap. For me the big problem is that all these
theoreticians, classical and modern, have been great mathematicians, and I
can't, or haven't the patience to, follow their mathematical arguments.
There's too much shorthand and convention; I can look right at an equation
and not recognize what it's saying. So who am I to say there's anything
wrong with the approach? My gut tells me that this whole field is one
gigantic systematic delusion, but if I stood up and said that in public I
would be unable to back it up with chapter and verse, because I can't talk
the language. I feel aphasic.

I guess someone else is going to have to resolve these issues, someone who
sympathesizes with my idea of control but has a far better mathematical
ability. It will be an uphill battle, of course: the proponents of modern
control theory treat it like the Second Coming, and there is a perfectly
enormous investment of man-hours and reputation in it.

The payoff will also be enormous if it should turn out that you are right
in saying "I infer that very very few even understand what control is."
from interchanges on CSGnet with certain inviduals I came to the same
conclusion, but it's such an unlikely conclusion that I haven't the courage
to proclaim it; I could so easily be flattened by a sophisticated
mathematical rebuttal.

Best,

Bill P.

[Abbott_Bruce]

[From Bruce Abbott (2000.09.18.0720 EST)]

Bruce Gregory (2000.0917.1336) --

It seems obvious, but it obviously isn't, that in order to control
some variable, the system must have a way of monitoring the variable and not
simply its own actions. This is particularly the case since it is trivially
easy to test. Don't let the thermostat know what the temperature is and see
what happens. Since no one bothers to do this test, I infer that very very
few even understand what control is. I'm uncomfortable with this conclusion,
but I can't see any way around it.

The confusion comes about because of the way these systems are often
diagramed and labeled. Consider the following diagram of a voltage regulator:

     INPUT ---------[regulator]------+---------> OUTPUT
      20V ^ | 5V (regulated)

···

+-----------+
                           feedback

The input is some voltage of say around 20 volts. The output is 5 volts,
and will remain very close to 5 volts despite variations in load (within
limits). Not shown is the set point, which would be just above 5 volts.
This is a perfectly ordinary control system; note that as labeled in the
diagram it is controlling its output. It might make some sense to use this
same sort of description to describe an animal's behavior: imagine the
"output" to be a particular speed of running; one could say that (this
aspect of) the animal's running behavior is controlled. "Behavior" does not
have to mean "actions."

The above diagram is often compared to an open-loop equivalent:

     INPUT ---------[resistor]----------------> OUTPUT
      20V 5V (unregulated)

The voltage drop shown here assumes a given current flow through the
resistor. If the load changes, the output voltage changes with it.

Now the animal's running speed is not controlled, but varies with the grade
of the land.

As I have noted before, it is also possible to "control behavior" in the
sense that a particular pattern of movements is created, as perceived by
both the actor and the observer, despite minor disturbances that would
otherwise have altered the pattern significantly. For these reasons one has
to be careful when reading about the "control of behavior" not to fall into
the trap of automatically assuming that the author is talking about control
of output (actions); maybe so, but then again, maybe not.

Bruce A.

[Bruce_Gregory]

[From Bruce Gregory (2000.0918.0956)]

Bruce Abbott (2000.09.17.2000 EST)

What we see is that the magnet is doing the attracting

What we observe is the filings moving to the magnet. We interpret this
movement as the result of the magnet exerting a force on the filings.
Newton assures us, however, that the fillings are exerting an equal
force on the magnet. Unless of course, psychologists are not yet ready
to acknowledge the validity of Newton's Third Law. Considering what they
do with control theory, this would not surprise me one bit.

observance of the
relationship is an observation; the explanation in terms of
mechanism is an
interpretation. See the difference?

Do you?

BG

[Bruce_Gregory]

[From Bruce Gregory (2000.0918.11480]

Bill Powers (2000.09.18.0130 MDT)

Could I have invented a whole new way of thinking about control that
_neither_ the classical nor the modern control theorists have
worked out?

I must admit, this interpretation seems more and more believable, to me
at least.

The payoff will also be enormous if it should turn out that
you are right
in saying "I infer that very very few even understand what
control is."
>From interchanges on CSGnet with certain individuals I came to the

same

conclusion, but it's such an unlikely conclusion that I
haven't the courage
to proclaim it; I could so easily be flattened by a sophisticated
mathematical rebuttal.

I think you have the evidence you need. The tracking data is modeled to
an extremely high degree of accuracy by the PCT model. No matter how
sophisticated a "modern control model" is, it doesn't have much room to
demonstrate its superiority (purchased at a relatively high cost in
computing overhead). It might be interesting to compare the performance
of a typical thermostatically controlled furnace with the performance of
a "modern control theory" controlled furnace. Again, there's not a hell
of a lot of room for improvement, as far as I can tell.

BG

[Bruce_Gregory]

[From Bruce Gregory (2000.0918.1207)]

Bill Powers (2000.09.18.1244 MDT)

Of course I do. I'm trying to get you to see that _either_ way of
describing the situation is unwarranted except in terms of a model.
Choosing a model is not just a matter of preference or optional
interpretations. If one model is right, the other is wrong. It's
conceivable that both are wrong, but they can't both be
right.

Please forgive me for picking this nit. I don't find it helpful to argue
that models are right or wrong. Models are more or less adequate
depending on the problem. True, some models are so inadequate that
nothing is lost by labeling them wrong, but right models (such as
Newton's Laws) have a way of turning out to be wrong. Newton's laws may
be wrong, but we use them every day and understand very well the domain
in which they work. We certainly don't discard old models that continue
to have their uses. This said, I agree with all the substantive points
you make. (My sensitivity about models is probably the outcome of
teaching the course on the Nature of Science, in which I talk about
facts and stories. Models are stories. Some stories are very powerful,
including Newton's. Some stories are virtually useless. I am by no means
a relativist, but am reluctant to think that we've finally got the
"right" story about anything.)

BG

[Bill_Powers1]

[From Bill Powers (2000.09.18.1347 MDT)]

Bruce Gregory (2000.0918.1207)--

Please forgive me for picking this nit. I don't find it helpful to argue
that models are right or wrong. Models are more or less adequate
depending on the problem. True, some models are so inadequate that
nothing is lost by labeling them wrong, but right models (such as
Newton's Laws) have a way of turning out to be wrong.

But Newton's laws have not turned out to be wrong. His laws of motion are
still exactly right; all that relativity did was to show that mass
increases with relative velocity, that geodesics are not straight in the
vicinity of matter, and that time is also affected. None of these factors
enters directly into Newton's laws of motion; but when we correct for them,
the laws still apply exactly as before.

Another way to see this is that Newton's laws remain qualitatively correct,
although their quantitative application requires the use of correction
factors that depend on velocity and mass/energy density.

There's another way to think the subject of modeling. I think that the
human organism is actually organized in some particular way. It may be a
lot more complex than our models are, but it is still a specific
organization and works in a specific way, however complex. Therefore one
model of it is _not_ as good as another; some models represent it better
than others do. We judge how well a model represents the real system by how
well the behavior of the model (at all levels of detail) matches the
behavior of the real system.

It is possible to make a model that predicts behavior to a certain extent;
it is possible to improve that model so it predicts better. At the end of
this process is a final model, the one that correctly represents the
organization and predicts its behavior under any specific circumstances to
the limits of measurement. By "correctly" I mean simply that the structure
and behavior of the final model is exactly isomorphic to the structure anhd
behavior of the real system.

I think we can tell bad models from good models, and good models from
better models. And if two models have contradictory properties, then at
least one of them must be wrong, since the real system can have no
self-contradictory properties.

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2000.09.18.1407 MDT)]

Bruce Abbott (2000.09.18.0720 EST)--

For these reasons one has
to be careful when reading about the "control of behavior" not to fall into
the trap of automatically assuming that the author is talking about control
of output (actions); maybe so, but then again, maybe not.

Don't you have to consider the source? Negative feedback control is
_always_ control of input, not action. But when most biologists,
psychologists, computer science experts, and neduroscientists speak of
control, they most commonly say they are describing control of actions,
even though we know the actual process consists of _varying_ actions to
produced consistent perceived consequences. If we can find a way to show
that what someone describes as a controlled "action" is probably actually a
controlled perception, we should not assume that the person who said it
knew that unless there is some pretty specific evidence that he or she did.

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2000.09.18.1414 MDT)]

Bruce Gregory (2000.0918.11480]--

No matter how
sophisticated a "modern control model" is, it doesn't have much room to
demonstrate its superiority (purchased at a relatively high cost in
computing overhead). It might be interesting to compare the performance
of a typical thermostatically controlled furnace with the performance of
a "modern control theory" controlled furnace. Again, there's not a hell
of a lot of room for improvement, as far as I can tell.

You're right. But the MCT people like to emphasize the ability of MCT to
handle multiple-input-multiple-output control processes, and also to handle
adaptation. They think that classical control theory can't handle either
one. I think it's pretty clear that that idea is wrong.

Best,

Bill P.

[Bruce_Gregory]

[From Bruce Gregory (2000.0918.1640)]

Bill Powers (2000.09.18.1347 MDT)

There's another way to think the subject of modeling. I think that the
human organism is actually organized in some particular way.
It may be a
lot more complex than our models are, but it is still a specific
organization and works in a specific way, however complex.
Therefore one
model of it is _not_ as good as another;

I never said that one model is as good as another.

some models
represent it better
than others do.

Indeed.

We judge how well a model represents the real
system by how
well the behavior of the model (at all levels of detail) matches the
behavior of the real system.

I thought I had said that.

It is possible to make a model that predicts behavior to a
certain extent;
it is possible to improve that model so it predicts better.
At the end of
this process is a final model, the one that correctly represents the
organization and predicts its behavior under any specific
circumstances to
the limits of measurement.

That's what I'm disputing. I'll change my mind when such a model exists.
(Of course all we can know is that up to now the model has never let us
down. That was true of Newton's model for at least two hundred years. So
how long should we wait before declaring the model is "final"?)

By "correctly" I mean simply that
the structure
and behavior of the final model is exactly isomorphic to the
structure and
behavior of the real system.

So far we haven't developed any such models. Clearly you are more
optimistic than I am.

I think we can tell bad models from good models, and good models from
better models.

I agree.

And if two models have contradictory
properties, then at
least one of them must be wrong, since the real system can have no
self-contradictory properties.

Such as being both a particle and a wave?

BG

[DTSDTO]

[Mark Lazare 2000.09.18.1400]

If Mike Domjan want to subscribe to the CSGnet who would he Write to?

Mark Lazare

[Bruce_Gregory]

[From Bruce Gregory (2000.0918.1722)]

Mark Lazare 2000.09.18.1400

If Mike Domjan want to subscribe to the CSGnet who would he Write to?

Why, Rick Marken, of course. He's the head of our welcoming committee.

BG

[Richard_Marken2]

[From Rick Marken (2000.09.18.1530)]

Mark Lazare (2000.09.18.1400) --

If Mike Domjan want to subscribe to the CSGnet who
would he Write to?

He should send the following message:

Subscribe CSGnet Michael Domjan

in the body of an e-mail to:

listserv@postoffice.cso.uiuc.edu

The complete instructions are at:

http://www.ed.uiuc.edu/csg/CSGNetsubscribe.html

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
MindReadings.com mailto: marken@mindreadings.com
www.mindreadings.com

[Abbott_Bruce]

[From Bruce Abbott (2000.09.18.2210 EST)]

Bruce Gregory (2000.0918.11480

I think you have the evidence you need. The tracking data is modeled to
an extremely high degree of accuracy by the PCT model. No matter how
sophisticated a "modern control model" is, it doesn't have much room to
demonstrate its superiority (purchased at a relatively high cost in
computing overhead). It might be interesting to compare the performance
of a typical thermostatically controlled furnace with the performance of
a "modern control theory" controlled furnace. Again, there's not a hell
of a lot of room for improvement, as far as I can tell.

In most cases you could do nearly as well modeling tracking data just by
assuming that the CV equals the target's varying value. Why, the margin for
improvement from there is so small that again, "there's not a hell of a lot
of room for improvement." So there's no need to bother with control theory,
modern or classic! (Now how's that for a reductio ad absurdum?)

Bruce A.