Fighting the hydra

[Bruce_Abbott]

[From Bruce Abbott (971129.1815 EST)]

In Greek mythology, the hydra was a nine-headed creature that was nearly
impossible to defeat, because whenever you cut off a head, two immediately
grew to take its place unless the wound was cauterized. I seem to be
fighting the hydra myself these days, because whenever I cut off one
argument against my position, two arise to take its place. And my opponents
are so busy raising their objections that they sometimes lose sight of just
what it is they are opposing.

It all started with my question to Bruce Gregory: If incentives don't work
(as he seemed to be implying in his criticism of an economist's statement),
what were all those people doing with their hands on that Corvette? Bruce's
answer was to admit that incentives do work, but only for some of the
people, some of the time. "O.K.," I thought, "we're making some progress."
So incentives _can_ work, and I suggested some of the factors that might
determine when an offered incentive would or would not work. I then defined
incentive as "something a person wants," and noted that so-called
"conventional" psychologists do not view an incentive as something that
"forces" a person to do something, as in a stimulus-response reflex. This
led to Bill Powers to note that the dictionary definition of "incentive"
suggests an external "push." (Apparently this was intended as proof that
psychologists _do_ mean to convey the notion that incentives "make" people
do things, or if they don't mean that, that psychologists use terms in
unconventional ways, which of late has been treated as a mortal sin in this
forum.) Also referring to the dictionary, I countered by showing that the
term as normally used does not convey what Bill suggested it does -- i.e.,
an irresistible force "making" people do things. Meanwhile, Richard Marken,
adopting a diversionary tactic, was busy trying to make it _seem_ as though
I had stated that "conventional" psychologists really have been using
control theory all along. After raising this straw man in my name, he
attacked it with a series of questions asking why, if psychologists have
been using control theory along, they use terms like "incentive" when
control theoretic terms would be so much better. Hank Folson added his two
cents by suggesting that a proper understanding of control theory makes such
terms unnecessary in that they do not contribute anything beyond what can be
a systems analysis would indicate.

Bill Powers then returned to the fray by suggesting that the term
"incentive" may, according to control theory, actually cover somethat
different cases without distinguishing them, thus making the term
"misleading" and "perpetuating ignorance." Ever the obedient follower,
Richard Marken immediately took up this new argument, echoing Bill, as if
having two people say it would somehow make it more persuasive. However,
what Bill's position actually says is not that incentives do not work (!).
Instead, Bill offers a nice PCT _explanation_ for when incentives would be
expected to work or to fail to work. By and large these are the same
conditions "conventional" psychologists have identified, although most
commonly these are presented as emperically identified boundary condtions
rather than expectations based on the workings of a hypothesized underlying
mechanism.

Now I would be among the first to agree that understanding based on a
well-supported theoretical mechanism is hands-down better than a set of
empirical rules, even rules that yield the same prediction as could be
derived from knowledge of the mechanism and the starting conditions. But
what has been asserted heretofore is that _incentives don't work_. Now, if
incentives don't work, how is it that control theory can identify those
conditions under which they do?

Regards,

Bruce

[Bill_Powers1]

[From Bill Powers (9712129.1792 MST)]

Bruce Abbott (971129.1815 EST)--

Now I would be among the first to agree that understanding based on a
well-supported theoretical mechanism is hands-down better than a set of
empirical rules, even rules that yield the same prediction as could be
derived from knowledge of the mechanism and the starting conditions. But
what has been asserted heretofore is that _incentives don't work_. Now, if
incentives don't work, how is it that control theory can identify those
conditions under which they do?

What makes you think that a good model based on knowledge of the mechanism
and starting conditions would not do better than a statistical
generalization? A good model would predict not only when the so-called
incentive effect would be seen, but when it would not be seen, in each
individual. Are you saying that psychologists who study incentives can do
that?

Bruce, unless I overestimate you considerably, I am sure you would agree
with me that there is no physical object or occurrance that literally has a
magical special effect on a person. But the language of psychology gives
the appearance that this is exactly the opposite of what psychologists
believe. Psychologists speak of events and objects as if they can
instigate, cause, incite, arouse, direct, and alter the behavior of
organisms. Are you telling me that they are privately thinking that none of
these appearances is really correct, even while they use language that
strongly suggests this power of things over people? When a psychologist
publishes a paper on "The effect of X on Y," are you saying that this is
intended and understood to mean that X really is not literally having an
effect on Y? Is the language of psychology really so completely out of tune
with what psychologists actually think?

Do you even understand what I'm talking about?

Best,

Bill P.

[Bruce_Gregory9]

[From Bruce Gregory (971129.22.20 EST)]

Bruce Abbott (971129.1815 EST)

In Greek mythology, the hydra was a nine-headed creature that was nearly
impossible to defeat, because whenever you cut off a head, two immediately
grew to take its place unless the wound was cauterized. I seem to be
fighting the hydra myself these days, because whenever I cut off one
argument against my position, two arise to take its place. And my opponents
are so busy raising their objections that they sometimes lose sight of just
what it is they are opposing.

It all started with my question to Bruce Gregory:

No, it all started with my quote from Landsburg

"Most of economics can be summarized in four words: 'People respond to
incentives.' The
rest is commentary."

Some of us misguidedly believe that Landsburg actually meant what he said.
Fortunately, you are not so benighted. (Fuck you Isaac)

Bruce

[Bruce_Abbott]

[From Bruce Abbott (971129.2350 EST)]

Bill Powers (9712129.1792 MST) --

Bruce Abbott (971129.1815 EST)

Now I would be among the first to agree that understanding based on a
well-supported theoretical mechanism is hands-down better than a set of
empirical rules, even rules that yield the same prediction as could be
derived from knowledge of the mechanism and the starting conditions. But
what has been asserted heretofore is that _incentives don't work_. Now, if
incentives don't work, how is it that control theory can identify those
conditions under which they do?

What makes you think that a good model based on knowledge of the mechanism
and starting conditions would not do better than a statistical
generalization? A good model would predict not only when the so-called
incentive effect would be seen, but when it would not be seen, in each
individual. Are you saying that psychologists who study incentives can do
that?

On rereading my first sentence, I see that the clause beginning "even rules"
could be misinterpreted. What I meant to convey is that I would favor an
explanation in terms of well-supported mechanism over a set of empirically
derived "rules" _even if_ the latter yielded the same predictions as the
former. I am not asserting that this is the case. At the same time, I know
of no one who would assert that something offered as an inducement or
incentive to perform some activity has the ability to force that activity to
occur in an individual, no matter how strong an "incentive" one might
imagine it to be for that person. Plenty of work has been done to identify
under what conditions something will or will not function as an incentive
for an individual. However, in application, too many have developed a
naive, superficial understanding of incentives and have attempted a "one
size fits all udner all conditions" approach that ignores the research in
this area. They become disillusioned when this simplistic approach fails
and conclude that incentives "don't work."

Bruce, unless I overestimate you considerably, I am sure you would agree
with me that there is no physical object or occurrance that literally has a
magical special effect on a person. But the language of psychology gives
the appearance that this is exactly the opposite of what psychologists
believe. Psychologists speak of events and objects as if they can
instigate, cause, incite, arouse, direct, and alter the behavior of
organisms. Are you telling me that they are privately thinking that none of
these appearances is really correct, even while they use language that
strongly suggests this power of things over people?

I am certainly not claiming that psychologists are saying one thing publicly
and thinking another thing privately. Would you say that a disturbance to a
controlled variable instigates, causes, incites, arouses, directs, or alters
the behavior of the organism controlling that variable? Is that power
vested in the disturbance or does the organism have something to do with its
effects?

Do you even understand what I'm talking about?

Sure do. I'll be interested to see whether you understand what _I'm_
talking about, after I've received your answers to the above questions.

Best wishes,

Bruce

[Bill_Powers1]

[From Bill Powers (971130.0500 MST)]

Bruce Abbott (971129.2350 EST)--

Plenty of work has been done to identify
under what conditions something will or will not function as an incentive
for an individual.

Do you have any particular study in mind? It strikes me that such studies
with individuals could come very close to testing for controlled variables.
Depends on what sort of "conditions" are considered. An incentive must be
closely related to a controlled variable. A person will act to produce it,
when possible; disturbances of it will be resisted ("Hey, you said you'd
give me a dime, not a nickel"); the person will act only if it is possible
to perceive it ("Wait a minute, how do I know you'll really give my note to
Clinton if I contribute 50 Grand?").

However, in application, too many have developed a
naive, superficial understanding of incentives and have attempted a "one
size fits all udner all conditions" approach that ignores the research in
this area. They become disillusioned when this simplistic approach fails
and conclude that incentives "don't work."

Can you really blame appliers of theories which are stated as general
propositions for assuming that the theories will work when they're applied?
Not all people using theories have the luxury of applying them only to the
groups used by the researchers and under the same conditions. A businessman
is not interested in how college sophomores in a mix of 60 percent affluent
and 40 percent underpriviledged participants, with half of them being women
and half men, will change their judgments depending on whether they are
paid one dollar or five dollars. The problem with statistical
generalizations is that they are highly sensitive to "conditions," but
there is no theory to say how to adjust them in conditions different from
those in which they were found. The researcher who complains that his
findings are not being applied under the proper conditions is really
admitting that his findings have no value outside his laboratory.

I am certainly not claiming that psychologists are saying one thing publicly
and thinking another thing privately. Would you say that a disturbance to a
controlled variable instigates, causes, incites, arouses, directs, or alters
the behavior of the organism controlling that variable? Is that power
vested in the disturbance or does the organism have something to do with its
effects?

No, I would not say that. A disturbance as defined in PCT is capable only
of altering the value of a controlled variable. That is its ONLY causal
property relevant to behavior. A change in the controlled variable
(whatever the cause) has ONLY the effect of altering the perceptions
derived from it. A behavior is caused ONLY by the error signals (whatever
their origin) driving it.

To understand the relationships among these variables in terms of the
immediate casual relations involved in each part of the system, you have to
understand how the whole control system works. Any attempt to explain
behavior just by looking at the overall effects will lead only to illusions
and misinterpretations. The system does not work the way it appears to work.

Consider an example. Suppose a lathe operator turns out x pieces per hour.
You offer the incentive of a 10% salary increase if the operator will turn
out x + 10 pieces per hour. The lathe operator does so.

Now you replace the lathe with a more efficient model, so with the same
amount of effort, twice as many pieces per hour can be produced. Under the
new incentive, how many pieces per hour do you expect will be produced?

The answer, according to PCT, is very nearly x + 10. Does incentive theory
agree? And can you explain why this result is expected, in either case?

Best,

Bill P.

[Bruce_Abbott]

[From Bruce Abbott (971130.1320 EST)]

Bill Powers (971130.0500 MST)]

Bruce Abbott (971129.2350 EST)--

Plenty of work has been done to identify
under what conditions something will or will not function as an incentive
for an individual.

Do you have any particular study in mind?

No, none in particular (it's been quite some time since I had a look at this
literature). It generally falls under the umbrella of cognitive psychology,
so Rick Marken, who got his Ph.D. in cognitive psychology, may be more up on
this area than I am. The general area is choice, and in particular, Utility
theory, such as developed by Duncan Luce. Amos Tversky has done quite a bit
of work examining how people tend to perceive various options, and has used
the findings to explain why individuals may be willing to accept bets that
on rational grounds they should not (e.g., buying insurance, betting the
lottery).

It strikes me that such studies
with individuals could come very close to testing for controlled variables.
Depends on what sort of "conditions" are considered. An incentive must be
closely related to a controlled variable. A person will act to produce it,
when possible; disturbances of it will be resisted ("Hey, you said you'd
give me a dime, not a nickel"); the person will act only if it is possible
to perceive it ("Wait a minute, how do I know you'll really give my note to
Clinton if I contribute 50 Grand?").

I believe I argued strenuously for this view when, in another context, we
were calling incentives "reinforcers." (It would be nice to have some label
for them that did not carry those unfortunate theoretical implications. Any
suggestions?)

I am certainly not claiming that psychologists are saying one thing publicly
and thinking another thing privately. Would you say that a disturbance to a
controlled variable instigates, causes, incites, arouses, directs, or alters
the behavior of the organism controlling that variable? Is that power
vested in the disturbance or does the organism have something to do with its
effects?

No, I would not say that. A disturbance as defined in PCT is capable only
of altering the value of a controlled variable. That is its ONLY causal
property relevant to behavior. A change in the controlled variable
(whatever the cause) has ONLY the effect of altering the perceptions
derived from it. A behavior is caused ONLY by the error signals (whatever
their origin) driving it.

I agree, but you have to understand that the term "cause" as used in
psychology and in many other fields refers to something somewhat different.
Example, you enter your living room and find that one of the lights is on.
What is the cause of the light being on? In your view of cause, the light
is on because electric current is heating the filiment. This gives the
proximal physical cause of there being light radiating from the bulb.
However, the question really asks for a different explanation: what event
has occurred which explains why the bulb is now on rather than off? Did
someone flip the switch to "on"? Was their a short in the switch?
Examining the switch, you find it in the "on" position. Ah, it's not a
short: someone left the light on.

Given that you have a functioning control system, applying a disturbance to
the CV will result in output from the system that largely counteracts the
effect of the disturbance on the CV. In this sense the disturbance causes
the system's response. That is, manipulating the disturbance produces
highly systematic changes in system output, everything else being equal
(e.g., no change in reference level during the observations). Most of us
would say that under these conditions, manipulating the disturbance causes
systematic changes in output. I gather that you would interpret this
statement as a claim that the disturbance has a direct physical influence on
the output. If so, you would have a lot of people scratching their heads,
wondering how you could think that this is what they meant.

Getting back to my questions:

Would you say that a disturbance to a
controlled variable instigates, causes, incites, arouses, directs, or alters
the behavior of the organism controlling that variable?

An experimental psychologist would say "yes," but by that would not be meant
to suggest that the changes in system output result from a direct physical
effect of the disturbance on the output. Rather, she would be saying that
there exists under the test conditions a set of relationships through which
changes in the disturbance lead to changes in the output.

Is that power
vested in the disturbance or does the organism have something to do with its
effects?

Obviously, if the "effects" of the disturbance on the output must be
transmitted through a set of relationships within the organism (in this case
a circular loop of causality), then these "effects" depend as much on the
properties of the mediating system as on the properties of the disturbance.

No, I would not say that. A disturbance as defined in PCT is capable only
of altering the value of a controlled variable. That is its ONLY causal
property relevant to behavior. A change in the controlled variable
(whatever the cause) has ONLY the effect of altering the perceptions
derived from it. A behavior is caused ONLY by the error signals (whatever
their origin) driving it.

No it isn't, not by your usage of "cause." The error signal can only cause
some physical change in neural potential (or perhaps hormonal level). It
can't do a damned thing to behavior. (Do you see my point?) However, let's
not argue about whose definition of "cause" is "more scientific" or some
such palaber. People in different fields apparently use these terms
somewhat differently. If you don't recognize what those who use the term
actually mean by it (and apply a different definition), the result can only
be confusion. One needs to interpret someone's words according to the
definitions _they_ apply to them, whether one agrees with those definitions
or not.

Regards,

Bruce

[Bill_Powers1]

[From Bill Powers (971130.1143 MST)]

Bruce Abbott (971130.1320 EST)--

Would you say that a disturbance to a
controlled variable instigates, causes, incites, arouses, directs, or
alters
the behavior of the organism controlling that variable?

[I said no]

An experimental psychologist would say "yes," but by that would not be meant
to suggest that the changes in system output result from a direct physical
effect of the disturbance on the output. Rather, she would be saying that
there exists under the test conditions a set of relationships through which
changes in the disturbance lead to changes in the output.

I'm not talking about a direct physical effect of the disturbance on the
output. I'm talking about the neural model in which most psychologists have
believed for nearly a century, in which stimuli act on sensory nerves to
produce signals that are relayed through pathways simple or complex to the
muscles, in a direct lineal causal chain. There have been many examples
cited on CSGnet to show that this is exactly what the foremost thinkers of
our time believe and preach.

No, I would not say that. A disturbance as defined in PCT is capable only
of altering the value of a controlled variable. That is its ONLY causal
property relevant to behavior. A change in the controlled variable
(whatever the cause) has ONLY the effect of altering the perceptions
derived from it. A behavior is caused ONLY by the error signals (whatever
their origin) driving it.

No it isn't, not by your usage of "cause." The error signal can only cause
some physical change in neural potential (or perhaps hormonal level). It
can't do a damned thing to behavior. (Do you see my point?)

Yes, I see your point. Now let's see if you see mine. The error signal is
not caused by the perceptual signal, but by the perceptual signal and the
reference signal, equally. The perceptual signal is not caused by the
disturbance, but by the controlled variable which is affected equally by
the disturbance and the system's own actions.

Each time you come to a point where more than one variable can act on an
effect, it becomes less plausible to treat either input as a "cause." I
have listed only two points; in a hierarchy of control systems with many
systems at each level, there are hundreds of points to put it mildly where
such convergent chains meet. To trace just one causal path through the
system from one input to one output is simply a conceptual error, which all
beginning engineering students learn how to avoid if they want to pass the
course.

When you say "all else being equal" in analyzing the relation between a
disturbance and an output, you mention only the reference signal as one
variable capable of interfering the the effect of the input on the output.
But there is another variable that drastically affects the influence of the
disturbance on the error signal and hence on the output: THE OUTPUT. Are
you going to hold the output constant, too, so as to let the disturbance
have its proper effect? Obviously, that is impossible: if you do it, the
output will not change. The real relationship with the reference signal
held constant is not

output = f(input), but

output = f(input,output).

This kind of relationship is simply unknown to psychologists; when they
speak of the effect of the input on the output, they can't possibly be
thinking of the correct relationship. The true input-output relationship
that you would see (at least briefly) if you cut the feedback loop would
look entirely different from the input-output relationship you would
observe with the feedback intact.

This is why I say that the appearance of a causal relation between a
disturbance and behavior is an ILLUSION. The actual relationship with the
loop closed is determined _primarily_ by the feedback function; it bears
almost no relation to the actual forward path through the organism.

When one naively varies an independent variable and observes changes in
behavior, isn't it normally assumed that the relation of the DV to the IV
is detemined by the forward path through the organism? That the observed
relationship tells us something about processes inside the organism in the
path between input and output?

I'm not talking here about different "perspectives" on behavioral
phenomena. We know how to prove that a control system is involved in a
behavior, when it is involved. From that alone, we can prove that the
input-output relationship observed in a simple input-output experiment
gives an incorrect picture of processes inside the organism. This is one of
the fundamental findings of PCT, a fact of nature that is new to
psychologists (even those who think they have tried to apply control theory).

However, let's
not argue about whose definition of "cause" is "more scientific" or some
such palaber. People in different fields apparently use these terms
somewhat differently. If you don't recognize what those who use the term
actually mean by it (and apply a different definition), the result can only
be confusion. One needs to interpret someone's words according to the
definitions _they_ apply to them, whether one agrees with those definitions
or not.

I hope I have made a little progress toward convincing you that this is not
just a matter of terminology. We're talking about the difference between
correct and incorrect concepts of how behavior works.

Best,

Bill P.

[Bruce_Abbott]

[From Bruce Abbott (971201.1035 EST)]

Bill Powers (971130.1143 MST) --

Bruce Abbott (971130.1320 EST)

Me:

Would you say that a disturbance to a
controlled variable instigates, causes, incites, arouses, directs, or
alters
the behavior of the organism controlling that variable?

[I said no]

An experimental psychologist would say "yes," but by that would not be meant
to suggest that the changes in system output result from a direct physical
effect of the disturbance on the output. Rather, she would be saying that
there exists under the test conditions a set of relationships through which
changes in the disturbance lead to changes in the output.

I'm not talking about a direct physical effect of the disturbance on the
output. I'm talking about the neural model in which most psychologists have
believed for nearly a century, in which stimuli act on sensory nerves to
produce signals that are relayed through pathways simple or complex to the
muscles, in a direct lineal causal chain. There have been many examples
cited on CSGnet to show that this is exactly what the foremost thinkers of
our time believe and preach.

Hmmm. Here is what you said:

No, I would not say that. A disturbance as defined in PCT is capable only
of altering the value of a controlled variable. That is its ONLY causal
property relevant to behavior. A change in the controlled variable
(whatever the cause) has ONLY the effect of altering the perceptions
derived from it. A behavior is caused ONLY by the error signals (whatever
their origin) driving it.

That doesn't sound to me like "the neural model in which most psychologists
have believed for nearly a century." That sounds to me like an unequivocal
statement that "cause" is to be defined purely in terms of an immediate,
direct physical influence of variable A on variable B. Let's try to resolve
_this_ issue before we go on to the more complex one concerning the nature
of the causal model being proposed.

No it isn't, not by your usage of "cause." The error signal can only cause
some physical change in neural potential (or perhaps hormonal level). It
can't do a damned thing to behavior. (Do you see my point?)

Yes, I see your point. Now let's see if you see mine. The error signal is
not caused by the perceptual signal, but by the perceptual signal and the
reference signal, equally. The perceptual signal is not caused by the
disturbance, but by the controlled variable which is affected equally by
the disturbance and the system's own actions.

Of course. But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output. If we hold disturbance constant
and vary the reference signal, we find that variation in reference signal
causes variation in output. If we vary both jointly (in the right way), we
can determine the independent contribution of each to the output. That is
the experimental method.

Each time you come to a point where more than one variable can act on an
effect, it becomes less plausible to treat either input as a "cause."

What the sort of anlaysis I am describing does is identify which variables
influence which. It is not aimed as identifying something that could be
described as _the_ cause of change in some other variable. Reference and
disturbance each can be shown to act as a "cause" of change in output in my
example, because other independent "causes" have been held constant during
the observations.

I
have listed only two points; in a hierarchy of control systems with many
systems at each level, there are hundreds of points to put it mildly where
such convergent chains meet. To trace just one causal path through the
system from one input to one output is simply a conceptual error, which all
beginning engineering students learn how to avoid if they want to pass the
course.

You have begged the question. No one is suggesting that only one causal
path be traced from one input to one output. As for the implication that I
am more naive than a beginning engineering student, did you really mean to
be condescending?

When you say "all else being equal" in analyzing the relation between a
disturbance and an output, you mention only the reference signal as one
variable capable of interfering the the effect of the input on the output.
But there is another variable that drastically affects the influence of the
disturbance on the error signal and hence on the output: THE OUTPUT. Are
you going to hold the output constant, too, so as to let the disturbance
have its proper effect? Obviously, that is impossible: if you do it, the
output will not change.

Obviously. But I might vary the feedback function and see what effect
_that_ has on the relationship between disturbance and output. Or I might
generate an output like that produced by the system and observe whether that
affects the system's own output. And so on.

However, we have moved onto a different issue (discovering the nature of the
system under study). The discussion was not about this, but about whether,
when experimental psychologists say, for example, that variations in
disturbance cause variations in output, that by this they necessarily mean
that there exists some direct physical lineal effect of the disturbance on
the output. My claim is that they do not. They may have developed a model
involving complex interactions of numerous variables that ultimately link
disturbance to output. Whether this model is lineal or includes feedback is
a different (and important) issue. But just because an experimental
psychologist concludes on the basis of an experiment that disturbance and
output are causally linked, one cannot conclude on that basis that she is
asserting a model of lineal causality for the system as a whole. To
determine that, you have to look at the model.

I hope I have made a little progress toward convincing you that this is not
just a matter of terminology. We're talking about the difference between
correct and incorrect concepts of how behavior works.

What we're (or at least I'm) talking about is not jumping to confusions
based on an unappreciated difference between you and experimental
psychologists in the use of the term "cause." For some reason you have
drawn the conclusion that I see the difference between open-loop and
closed-loop systems as one of mere terminology. Come on, Bill, you know me
better than that.

Regards,

Bruce

[R_Marken]

[From Rick Marken (971201.0830)]

Bruce Abbott (971201.1035 EST) --

You're hydra meat, baby;-)

You said:

But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output.

This is FALSE!! Until you can figure out why this sentence is
false, you will not understand why the conventional (causal)
basis of psychological research is wrong. If you believe that what
you say here is true then there is no possible way you could have
understood my "Dancer..." paper (as you said you did when you
reviewed it). Bill explained in an earlier post why your statement
above is false. I'll just give you a hint. In a control loop,
output = f(disturbance, output), NOT output = f(disturbance).

It's a loop, boobabla; output is (in part) a function of _itself_!!
The hydra has slain you. Time to be re-born as a control theorist.

Best

Rick

···

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

[Bill_Powers1]

[From Bill Powers (971201.0929 MST)]

Bruce Abbott (971201.1035 EST) --

I'm not talking about a direct physical effect of the disturbance on the
output. I'm talking about the neural model in which most psychologists have
believed for nearly a century, in which stimuli act on sensory nerves to
produce signals that are relayed through pathways simple or complex to the
muscles, in a direct lineal causal chain. There have been many examples
cited on CSGnet to show that this is exactly what the foremost thinkers of
our time believe and preach.

Hmmm. Here is what you said:

No, I would not say that. A disturbance as defined in PCT is capable only
of altering the value of a controlled variable. That is its ONLY causal
property relevant to behavior. A change in the controlled variable
(whatever the cause) has ONLY the effect of altering the perceptions
derived from it. A behavior is caused ONLY by the error signals (whatever
their origin) driving it.

That doesn't sound to me like "the neural model in which most psychologists
have believed for nearly a century." That sounds to me like an unequivocal
statement that "cause" is to be defined purely in terms of an immediate,
direct physical influence of variable A on variable B.

Yes, that's what I mean. It's not the neural model in which most
psychologists have believed for nearly a century (which I think is wrong).
It's the model that most systems engineers use, insofar as they talk about
"causes" at all. In engineering, if you say that A causes B, you imply that
there is nothing else that can affect B at the same time.

Let's try to resolve
_this_ issue before we go on to the more complex one concerning the nature
of the causal model being proposed.

... But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output. If we hold disturbance constant
and vary the reference signal, we find that variation in reference signal
causes variation in output. If we vary both jointly (in the right way), we
can determine the independent contribution of each to the output. That is
the experimental method.

That method works only when there are no loops. In engineering, the
possibility of loops is always present, which is why this method has to be
used cautiously. You skipped over holding the output constant (to prevent
unwanted effects on the input) while you vary the disturbance. Why? Because
it screws up the whole procedure.

Each time you come to a point where more than one variable can act on an
effect, it becomes less plausible to treat either input as a "cause."

What the sort of anlaysis I am describing does is identify which variables
influence which. It is not aimed as identifying something that could be
described as _the_ cause of change in some other variable. Reference and
disturbance each can be shown to act as a "cause" of change in output in my
example, because other independent "causes" have been held constant during
the observations.

If the system you're investigating contains loops, the influences will be
misidentified by your method.

You have begged the question. No one is suggesting that only one causal
path be traced from one input to one output. As for the implication that I
am more naive than a beginning engineering student, did you really mean to
be condescending?

No, critical, because I think that deep down inside somewhere you know
better than to do this.

When you say "all else being equal" in analyzing the relation between a
disturbance and an output, you mention only the reference signal as one
variable capable of interfering the the effect of the input on the output.
But there is another variable that drastically affects the influence of the
disturbance on the error signal and hence on the output: THE OUTPUT. Are
you going to hold the output constant, too, so as to let the disturbance
have its proper effect? Obviously, that is impossible: if you do it, the
output will not change.

Obviously.

What a minute. Obviously? What I said is a fatal flaw in your idea of
holding everything else that can affect the input constant while you vary
just one of the influences, like the disturbance. The main other variable
that can affect the input while you're manipulating disturbances is the
output. But if you hold the output constant, you'll have no effect to
measure. If you break the connection from output to input, you'll get a
completely different relationship between the disturbance and the output,
and I mean completely and drastically different.

You really need to try this out in a simulation if you want to understand
what I'm taliing about. Make a simple integrating control system with a
disturbance function of 1 and a feedback function of between 1 and 100.
Plot the output against the disturbance. Then break the output-to-input
connection (set the constant to zero) and do the plot again. And finally,
with the loop closed, do the plot with the feedback function set to 2,4,8
and 16, with the output ingtegration factor set so the system remains
stable over that range (say, around 0.03). The object of this exercise is
to convince yourself that with the loop intact, what you see is primarily
the form of the environmental feedback function, not the forward properties
of the control system.

But I might vary the feedback function and see what effect
_that_ has on the relationship between disturbance and output.

That is exactly what I am urging you to do.

Or I might
generate an output like that produced by the system and observe whether that
affects the system's own output. And so on.

The only valid way to do this is to break the system down into subsystems
with well-defined inputs and outputs and NO LOOPS. You can then
characterize each subsystem as a purely causal system, describing it with
an input-output equation. However, to see what will happen with the system
intact, you must then solve the simultaneous equations, actually or by
simulation. You can't reason out the result just by tracing causal pathways.

If you generated an artificial "output" that affected the input, you would
have to prevent the system's own output from affecting the input at the
same time. Otherwise the effect on the system would not be due only to your
manipulated variable. Something would happen, but it would not reveal the
system's true response. Such an artificial output would simply be a
disturbance.

... just because an experimental
psychologist concludes on the basis of an experiment that disturbance and
output are causally linked, one cannot conclude on that basis that she is
asserting a model of lineal causality for the system as a whole. To
determine that, you have to look at the model.

WHAT model? The only model I know of that traditional psychologists use is
the input-output model. What are you talking about?

What we're (or at least I'm) talking about is not jumping to confusions
based on an unappreciated difference between you and experimental
psychologists in the use of the term "cause." For some reason you have
drawn the conclusion that I see the difference between open-loop and
closed-loop systems as one of mere terminology. Come on, Bill, you know me
better than that.

I thought so, too.

Best,

Bill P.

[Bruce_Abbott]

[From Bruce Abbott (971201.1215 EST)]

Rick Marken (971201.0830) --

Bruce Abbott (971201.1035 EST)

You're hydra meat, baby;-)

You said:

But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output.

This is FALSE!! Until you can figure out why this sentence is
false, you will not understand why the conventional (causal)
basis of psychological research is wrong. If you believe that what
you say here is true then there is no possible way you could have
understood my "Dancer..." paper (as you said you did when you
reviewed it). Bill explained in an earlier post why your statement
above is false. I'll just give you a hint. In a control loop,
output = f(disturbance, output), NOT output = f(disturbance).

It's a loop, boobabla; output is (in part) a function of _itself_!!
The hydra has slain you. Time to be re-born as a control theorist.

Funny, I don't _feel_ dead! Let's see, fingers moving, nope, I think I'm
still very much alive. Sorry to disappoint you, Hydra, but I'm not so
easily vanquished. I was talking about what is observed in the steady
state. There is a nice, systematic, IV-DV relationship between disturbance
and output if you observe the final output value at each disturbance value,
after the transition.

I know how simple proportional control systems work.

Chopping off yet another hydra head,

Bruce

[R_Marken]

[From Rick Marken (971201.1000)]

Bruce Abbott (971201.1215 EST) --

Sorry to disappoint you, Hydra, but I'm not so easily vanquished.

Boy, do I know that! I didn't really expect that you would be
vanquished. Don't think there is any way the Hydra can win this
fight.

PCT is like heaven. The people who want to get here are dying to
get in;-) Their old misconceptions about human nature have to die
before they can be reborn as control theorists. You seem to be
comfortable with your misconceptions. So enjoy!

I was talking about what is observed in the steady state. There is
a nice, systematic, IV-DV relationship between disturbance and
output

Yes. There is. But you said:

But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output.

This statement is false because disturbance variations do not
_cause_ output variations in a closed loop system. Your statement
would have been true if you had said "disturbance variations are
systematically related to output variations" (as you did the
second time around) or "disturbance variations _appear to_ cause
output variations, but don't".

I know how simple proportional control systems work.

The only thing you don't seem to know is how living control
systems work;-)

Chopping off yet another hydra head

I am confident that you will prevail against the hydra, preserving
your ingnorance of PCT. But the fight is fun even if the end result
is known.

Best

Rick

···

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

[Bruce_Abbott]

[From Bruce Abbott (971201.1700 EST)]

Bill Powers (971201.0929 MST) --

Bruce Abbott (971201.1035 EST)

That doesn't sound to me like "the neural model in which most psychologists
have believed for nearly a century." That sounds to me like an unequivocal
statement that "cause" is to be defined purely in terms of an immediate,
direct physical influence of variable A on variable B.

Yes, that's what I mean. It's not the neural model in which most
psychologists have believed for nearly a century (which I think is wrong).
It's the model that most systems engineers use, insofar as they talk about
"causes" at all. In engineering, if you say that A causes B, you imply that
there is nothing else that can affect B at the same time.

In psychology and many other sciences, you don't imply that. That was my point.

... But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output. If we hold disturbance constant
and vary the reference signal, we find that variation in reference signal
causes variation in output. If we vary both jointly (in the right way), we
can determine the independent contribution of each to the output. That is
the experimental method.

That method works only when there are no loops. In engineering, the
possibility of loops is always present, which is why this method has to be
used cautiously. You skipped over holding the output constant (to prevent
unwanted effects on the input) while you vary the disturbance. Why? Because
it screws up the whole procedure.

It seems to work fine when manipulating the reference and disturbance and
observing the resulting changes in output. Manipulating the feedback
function (discussed later in the post) could include holding the output
constant, so I have not actually "skipped over" it. The method works only
when there are no loops? What other method is there, than to manipulate
some variable under controlled contitions and observe what happens to other
variables in the system?

If the system you're investigating contains loops, the influences will be
misidentified by your method.

I suspect you must be thinking of a much cruder approach than I have in mind.

When you say "all else being equal" in analyzing the relation between a
disturbance and an output, you mention only the reference signal as one
variable capable of interfering the the effect of the input on the output.
But there is another variable that drastically affects the influence of the
disturbance on the error signal and hence on the output: THE OUTPUT. Are
you going to hold the output constant, too, so as to let the disturbance
have its proper effect? Obviously, that is impossible: if you do it, the
output will not change.

Obviously.

What a minute. Obviously? What I said is a fatal flaw in your idea of
holding everything else that can affect the input constant while you vary
just one of the influences, like the disturbance.

I didn't say anything about holding everything else constant that would
affect the _input_. (This would require holding the output constant.) When
I said I would hold other conditions constant while varying the disturbance,
I did not include the output, which is after all the dependent variable
whose changes are to be accounted for. However, blocking the output's effect
on the input (manipulating the feedback function) was a strategy I allowed
for, and this certainly would be revealing. It would show that the
relationship between disturbance and output is strongly moderated by
feedback from the output.

The main other variable

that can affect the input while you're manipulating disturbances is the
output. But if you hold the output constant, you'll have no effect to
measure. If you break the connection from output to input, you'll get a
completely different relationship between the disturbance and the output,
and I mean completely and drastically different.

Yeah, I think I just implied that.

You really need to try this out in a simulation if you want to understand
what I'm taliing about. Make a simple integrating control system with a
disturbance function of 1 and a feedback function of between 1 and 100.
Plot the output against the disturbance. Then break the output-to-input
connection (set the constant to zero) and do the plot again. And finally,
with the loop closed, do the plot with the feedback function set to 2,4,8
and 16, with the output ingtegration factor set so the system remains
stable over that range (say, around 0.03). The object of this exercise is
to convince yourself that with the loop intact, what you see is primarily
the form of the environmental feedback function, not the forward properties
of the control system.

I don't need to do the simulation, Bill, I know what the Behavioral Illusion
is, and how to demonstrate it. Somehow you've gotten into an argument with
me about what can be learned about a system by systematically manipulating
its variables and observing the results. I know that it is easy to be
mislead by those results if one does not understand how to approach
closed-loop systems analytically. (And I agree that a lot of experimental
psychologists don't, although there are some notable exceptions.) But I
don't know any other way to discover a system's organization than to study
it systematically, manipulating variables and observing changes in others,
and taking it apart if possible to examine the functional properties of the
components. Of course, if you already have a model to test (such as a
control model), you can measure the relevant variables in the real system
(if you can get at them) and observe whether their changes correspond to
those of the model under the same conditions.

But I might vary the feedback function and see what effect
_that_ has on the relationship between disturbance and output.

That is exactly what I am urging you to do.

Or I might
generate an output like that produced by the system and observe whether that
affects the system's own output. And so on.

The only valid way to do this is to break the system down into subsystems
with well-defined inputs and outputs and NO LOOPS. You can then
characterize each subsystem as a purely causal system, describing it with
an input-output equation. However, to see what will happen with the system
intact, you must then solve the simultaneous equations, actually or by
simulation. You can't reason out the result just by tracing causal pathways.

I agree. If you think I was saying something else, then I am sorry for
being so unclear.

... just because an experimental
psychologist concludes on the basis of an experiment that disturbance and
output are causally linked, one cannot conclude on that basis that she is
asserting a model of lineal causality for the system as a whole. To
determine that, you have to look at the model.

WHAT model? The only model I know of that traditional psychologists use is
the input-output model. What are you talking about?

There are all sorts of models out there, for different situations. I
already mentioned utility theory, which proposes that people attempt to
balance the subjective costs or risks against subjective estimates of
probable benefit (yielding decisions that frequently do not accord with the
objectively best bet). Negative feedback models are certainly no stranger
to John Staddon and many others. Several models I know of assume that
behavioral output reflects some sort of equilibrium value among two or more
variables. Just because you don't know about them does not mean that they
do not exist. (Absence of evidence is not evidence of absence.) However,
most of these models are models of behavior and not models of the behaving
system, and there is the great advantage of control theory over these
would-be competitors.

All the best,

Bruce

[Martin_Taylor4]

[Martin Taylor 970112 18:55]

Rick Marken (971201.0830)]

Bruce Abbott (971201.1035 EST) --

You're hydra meat, baby;-)

You said:

But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output.

This is FALSE!!
...
I'll just give you a hint. In a control loop,
output = f(disturbance, output), NOT output = f(disturbance).

I'm afraid I don't see how you reconcile this with the statement that
varying the disturbance DOES NOT cause any variation in the output.

How does the output stay constant when the disturbance changes while
the reference is constant, in a loop with non-zero loop gain?

Mildly curious:-)

Martin

[Bill_Powers1]

[From Bill Powers 9971201.1616 MST)]

Bruce Abbott (971201.1700 EST)--

That method works only when there are no loops. In engineering, the
possibility of loops is always present, which is why this method has to be
used cautiously. You skipped over holding the output constant (to prevent
unwanted effects on the input) while you vary the disturbance. Why? Because
it screws up the whole procedure.

It seems to work fine when manipulating the reference and disturbance and
observing the resulting changes in output. Manipulating the feedback
function (discussed later in the post) could include holding the output
constant, so I have not actually "skipped over" it.

If you hold the output constant, how are you going to observe tbe effect of
the disturbance on the output? If you don't hold the output constant, the
effect of the disturbance on the input will not be what it seems to be. And
if you break the external feedback connection, the relation of disturbance
to output will change radically.

The method works only
when there are no loops? What other method is there, than to manipulate
some variable under controlled contitions and observe what happens to other
variables in the system?

The other method is to measure or propose units which behave as local
input-output functions, construct a system model using them, and predict
mathematically what the behavior will be. That is how essentially all
engineering analyses and testing of physical theories are done. I have
never seen or heard of this method being taught to students of psychology.
Do you describe it in your methods book?

You say you're familiar with the Behavioral Illusion and understand how it
works. I'm not convinced that you have understood all its ramifications.
But I sense such resistance and protectiveness of your chosen profession
that I'm just getting depressed by this effort. I'm fighting a losing
battle here, and I really don't have the energy to keep it up.

Best,

Bill P.

[Myrexsw]

i.kurtzer (971201)

[From Bruce Abbott (971129.1815 EST)]

In Greek mythology, the hydra was a nine-headed creature that was nearly
impossible to defeat, because whenever you cut off a head, two immediately
grew to take its place unless the wound was cauterized. I seem to be
fighting the hydra myself these days, because whenever I cut off one
argument against my position, two arise to take its place.

This is a infuriating situation, but one that for me describes even better the
position of conventional psychology as the hydra.

It all started with my question to Bruce Gregory: If incentives don't work
(as he seemed to be implying in his criticism of an economist's statement),
what were all those people doing with their hands on that Corvette?

But is the conclusion "that incentives must be present" based on a meaningfully
predictive theory that explicates what incentives are in relation to behavior?

Was it derived that incentives were in place since that theory which entailed
the accurate prediction of behavior included as one of its terms, "incentive" ?
  That seems to be the first grounds by which this term is to be regarded.
"Incentive" is justified as an explanatory term in behavior when it is a term
within a sucessful theory.
Do current or past behavioral theories make predictions of that specificity and
accuracy to be considered successful? Are there any theories of behavior that
do not include this term and are predictively sucessful?

i.

[Myrexsw]

i.kurtzer (971201)

[From Bruce Gregory (971129.22.20 EST)]

Some of us misguidedly believe that Landsburg actually meant what he said.
Fortunately, you are not so benighted. (Fuck you Isaac)

i'll keep that in mind.

i.

[David_Center]

Hello All,

I'm new to this listserve and new to PCT, so please excuse my ignorance.

First, I would say on the issue of what do traditional psychologists
mean by the term "incentive" Who cares? Aren't you trying here to
develop an alternative to traditional psychology? Let them play their
game and get on with your own game, i.e., define "incentive" from a PCT
perspecitve and move on. Reading the messages gives me the feeling that
I've hooked into a debate among a group of hostile lexicographers. I'll
try to give you a few of my thoughts on the issue. My major reluctance
is that someone is going to start quoting a dictionary back at me. I'm
very busy and don't have time for it. Y'all must all be retirees or
neglecting your "real" jobs. Being a novice, I'll be easy pickings for
someone's ego.

Now that I've probably pissed half of you off. Fools rush in...

There seems to be some minor agreement that "incentive" from a PCT
perspective means "something someone wants," that is, a reference value
(or whatever you call it). I'll call it X. If someone wants X and doesn't
have it, there must be a perceived error signal. If they aren't trying
to correct the perceived error, it must be that they have categorized it
as not correctable by any available means. Then someone introduces into
the environment a contingency that makes correcting the error signal
possible or even relatively easy using means that are available. Now,
we see actions directed at correcting the error signal and bringing the
perceived state of reality into alignment with the reference value. It
seems to me that what is being called an "incentive" isn't X but rather
the contingency that makes X attainable by available means. In short,
"incentive" isn't a thing or event but rather a relationship between what
someone wants, the means available for correcting the error signal, and
the perceived probablility of reducing or eliminating the error signal.

A quick example from my experience. I worked with a kid once, in an
educational setting, who frequently provoked other students. I knew that
one thing he wanted was to shoot basketball(X). Since that wasn't part
of our routine and had never happened during school hours (a very low
probability event), he had a large error signal between X and the
perceived state of things. He also had no available way of modifying the
error signal (no means). Thus, he took no actions that were readily
interpretable as attempts to modify the error signal. I put a bowl of
15 poker chips on his desk and told him that each one was worth a minute.
I told him that each time he verbally provoked someone I would take one
of the poker chips. I also told him that at the end of the day, before
the bus came, he could go shoot baskets for as many minutes as he had
chips left (contingency). Now the means to correct the error signal were
available because verbal provoking was under his control and he knew that
being excused early was under my control. Therefore, he refrained from
provoking, retained the chips and got to shoot baskets(the error signal
was reduced). At the time, I called shooting basketball a reinforcing
response or activity reinforcer and I could just as easily called it an
incentive in the conventional psychological sense. I suppose from a
PCT perspective, I might now say that I made it possible, through the
contingency, for him to correct his error signal with means that were
available to him.

Well, I probably screwed up the language and concepts something awful and
I'm sure someone will take me to task for it, but I don't really care. I
was swimming up stream when I got here and I can keep on swimming. In the
event that I get anything approaching a rational response, I have a couple
of other equally naive comments on previous messages. Later.

[R_Marken]

[From Rick Marken (971201.1900)]

Bruce Abbott said:

But when we hold, say, reference signal constant, then variation
in disturbance causes variation in output.

I said:

This is FALSE!!
...
I'll just give you a hint. In a control loop,
output = f(disturbance, output), NOT output = f(disturbance).

Martin Taylor (970112 18:55) says:

I'm afraid I don't see how you reconcile this with the statement
that varying the disturbance DOES NOT cause any variation in the
output.

I don't have to reconcile it. I never said that the disturbance
DOES NOT cause _any_ variation in output. I said that the
statement "variation in disturbance causes variation in output"
is false because the disturbance DOES NOT cause variation in
output: the simultaeous effect of disturbance _and_ output on
the controlled variable causes variation in output.

How does the output stay constant when the disturbance changes
while the reference is constant, in a loop with non-zero loop
gain?

Got me?

Best

Rick

···

--

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

[Bruce_Abbott]

[From Bruce Abbott (971201.1145 EST)]

David Center (971201.2118) --

First, I would say on the issue of what do traditional psychologists
mean by the term "incentive" Who cares? Aren't you trying here to
develop an alternative to traditional psychology? Let them play their
game and get on with your own game, i.e., define "incentive" from a PCT
perspecitve and move on.

Great idea, David, but I tried that approach. Not acceptible -- no such
thing as "incentive" from a PCT perspective, or so I'm told.

Y'all must all be retirees or neglecting your "real" jobs.

For me, the latter. Bill P. is retired, though.

There seems to be some minor agreement that "incentive" from a PCT
perspective means "something someone wants," that is, a reference value
(or whatever you call it). I'll call it X. If someone wants X and doesn't
have it, there must be a perceived error signal. If they aren't trying
to correct the perceived error, it must be that they have categorized it
as not correctable by any available means. Then someone introduces into
the environment a contingency that makes correcting the error signal
possible or even relatively easy using means that are available. Now,
we see actions directed at correcting the error signal and bringing the
perceived state of reality into alignment with the reference value. It
seems to me that what is being called an "incentive" isn't X but rather
the contingency that makes X attainable by available means. In short,
"incentive" isn't a thing or event but rather a relationship between what
someone wants, the means available for correcting the error signal, and
the perceived probablility of reducing or eliminating the error signal.

Geez you're smart, David.

Well, I probably screwed up the language and concepts something awful and
I'm sure someone will take me to task for it, but I don't really care. I
was swimming up stream when I got here and I can keep on swimming. In the
event that I get anything approaching a rational response, I have a couple
of other equally naive comments on previous messages.

Well, for what it's worth, you can swim in my stream any day. Someone else
will have to do the taking-to-task; what you say makes great sense to me.

Regards,

Bruce