Why PCT don't get no respect

[Fred_Nickols1]

[From Fred Nickols (2009.09.29.0835 EST)]

PCT clearly suffers from the Rodney Dangerfield syndrome.
I think the wounds are self-inflicted.

It is one thing to demonstrate the superiority of PCT; it is
altogether a different thing to tear down other edifices.

Too much time is spent attacking the old views. Why?
What does that gain?

Consider the behaviorists’ mantra that “Behavior
is a function of its consequences.” I have no problem with that
assertion, including its validity in light of PCT. Is not the closing of
a gap between perception and reference signal a “consequence” of
behavior? Is not some behavior “intended” to close such
gaps? Does not the closing of gaps confirm the actions that closed them
(or as some behaviorists might say, “reinforce” them)? Most
behaviorists I know would readily agree that the S-R model is kaput so why beat
a dead horse?

Rather persistent patterns in behavior can be observed just
about anywhere, including in the discourse on CSGNet so why harp on the
variability of behavior? We all know it varies. However, I for one
happen to believe that it varies within limits; we are restrained and
constrained by our repertory of behaviors and our belief systems.

I think I “get” PCT but I’ll happily
acknowledge that I will never fully grasp or appreciate it from an engineering
or scientific perspective. To me, it’s enough that it makes eminent
sense and is perfectly consistent with what I know of human nature and behavior
and the way the world works.

The cognitive (i.e., computational) psychologists might or
might not have something to offer; I’m not smart enough or well enough
read in that area to offer an opinion. I do know that I attach not one
shred of credence to the view that we “compute” our actions. I
do from time to take make “calculated” moves but that’s a
figure of speech referring to having thought things through and having tried to
ensure that things turn out as planned (which sometimes they do and sometimes
they don’t).

I did locate and read the Bourbon-Powers paper demonstrating
the inability of the S-R and cognitive views to match human behavior and found
it very interesting; however, I do not find it very persuasive. In sum,
what the paper says to me is that “We wrote a computer program that
operates in accordance with the principles of PCT and that program very
precisely mimics human behavior. We also wrote computer programs that
operate in accordance with the principles of S-R behaviorism and cognitive
psychology and those programs do not accurately mimic human behavior.
Therefore, PCT is superior.” Without being able to take apart and
study those programs from an informed perspective (which I can’t), I can’t
gauge their significance or import. All I can do is say, “If you
say so” (which is where I am with respect to all scientific research).
Like many ordinary people, when it comes to scientific research, about all I
can do is try to understand it, pay attention to what others are saying about
it, and test it against my own experiences of the world. Fortunately,
that’s not a problem for PCT or PCTers because ordinary people like me
aren’t the targets for persuasion; other scientists and researchers are
the targets. On that score, I can’t provide any insightful
counsel. But, for us ordinary folks, you need some whiz-bang,
knock-your-socks-off practical applications of PCT to carry the day. That’s
why I hang around; there’s got to be a pony or two in there somewhere.

Interestingly, the other feedback-based, systems view of
things (Jay Forrester’s stuff) also suffers from the Rodney Dangerfield
syndrome. Those folks can also often be heard lamenting the fact that
their powerful technology and know-how doesn’t get the respect or
attention it deserves. Even more interestingly, there are loads of
practical applications of Forrester’s stuff but they, too, suffer from being
hidden inside the computer where only the folks who put those programs in place
know how they really work. To the rest of us, it’s just another “black
box” – and we can’t talk to it or look it in the eye or read
its body language or run any other check for honesty or truthfulness. I’m
probably more inclined to trust what’s inside the box than some ordinary
folks because of my weapons systems background. I knew what that fire
control computer did and how it did it; I could take it completely apart,
spread its guts all over the deck in the plotting room; then reassemble, align
and test it to make sure it was working the way it was supposed to. The
final test, of course, was when we did what my weapons officer wanted: When
he said “Shoot” the guns went bang and the bullets hit the target.

I was a pretty good technician but I’m no engineer and
if you have to be an engineer to fully understand and appreciate PCT it’s
beyond me and zillions of others folks, too. If only the engineers can
get PCT, there’s a big problem because there are lots of engineers who
suffer from the Rodney Dangerfield syndrome, too.

For PCT to move center stage in things, it will have to
appeal to and catch the fancy of some powerful, well-placed people. If
that happens, there are hordes of researchers out there who will happily
conduct all manner of experiments and other forms of research to satisfy the
need powerful people have for a scientific basis for their actions. Gee,
maybe that’s the problem with PCT. PCT doesn’t afford you a
way off the hook for your actions; indeed, PCT very clearly suggests that you
do what you do to get what you want and that implies an extraordinary degree of
personal accountability. The powers that be (no pun intended) won’t
like that.

Anyway, that’s enough rambling. Y’all have
a great day.

Regards,

Fred Nickols

nickols@att.net

[Bill_Powers1]

[From Bill Powers (2009.09.29.0759 MDT)]

Fred Nickols (2009.09.29.0835 EST) –

FN: It is one thing to
demonstrate the superiority of PCT; it is altogether a different thing to
tear down other edifices.

Too much time is spent attacking the old views. Why? What
does that gain?

Consider the behaviorists’
mantra that “Behavior is a function of its consequences.” I have no
problem with that assertion, including its validity in light of
PCT. Is not the closing of a gap between perception and reference
signal a “consequence” of behavior? Is not some behavior “intended”
to close such gaps? Does not the closing of gaps confirm the
actions that closed them (or as some behaviorists might say, “reinforce”
them)? Most behaviorists I know would readily agree that the S-R
model is kaput so why beat a dead horse?

BP: I’ve been accused of beating that dead horse for some 50 years. It’s
still not dead, though it’s not exactly well. In the case you speak of,
behaviorists have claimed since the days of Watson in 1910 or so that you
don’t need to guess at what goes on inside an organism. It’s sufficient
to observe the stimuli acting on it, and the actions that are thereby
produced. That’s why it is called behaviorism: behavior is what a
scientist can observe; all the rest, according to Watson, is
“primitive superstition.” Watson and other behaviorists since
his era have spent a good deal of time attacking differing old and new
views, including my work, that suggest any modeling of the internal parts
of an organism that can’t be seen. They were doing that before I was
born; you can’t accuse me of starting the fracas.

So a pure-bred behaviorist would not admit that any of those gaps you
mention is an observable consequence of behavior. The consequences that a
behaviorist like Skinner meant are called reinforcers. When a rat presses
a lever by accident while moving randomly about in its cage, a piece of
food drops into a dish: that is an observable consequence of pressing the
lever. The next time the rat is in about the same situation, being
affected by the same “discriminative stimuli,” it is more
likely to repeat the same action and produce another edible consequence.
Thus behavior is controlled by its consequences (I don’t recall Skinner’s
ever using the term “is a function of”). To quote Skinner:
Reinforcement brings behavior under the control of discriminative
stimuli. Behaviorists think that is different from S-R theory. It’s
not.

This view of behavior, according to PCT, is exactly wrong. The
consequence is clearly controlled by the behavior; unless the lever is
pressed no food will appear. When the lever is pressed, the food does
appear. Appearance of the reinforcer is caused by, is a function of, the
behavior. And as we now know, varying the behavior is the means by which
a control system comes to control the occurrances of the reinforcer. And
randomly varying the various parameters of control is how the organism
comes to behave in that way – not because of any
“strengthening” effect of the so-called reinforcer. The
organism reorganizes until its behavior is controlling the local
environment to produce the consequences that the organism intends to
produce. A behaviorist hearing me say that would tell me to wash my mouth
out with soap.

Behaviorism is not just some pretty edifice that we ought to leave alone
just to be polite. It is a wrong conception of the nature of behavior,
and deserves to join phlogiston in oblivion. It is exactly opposed to PCT
at its very roots; it can’t coexist with PCT. I know full well that this
is a very difficult proposition for anyone who has admired behaviorism to
accept. But just fiddling around with the definitions of words is not
going to revitalized a horse that’s on the critical list. Watson and his
descendants like Skinner just had the worst of bad luck: they saw that a
certain idea made perfect sense when in fact it was completely wrong.
That happens now and then to everybody and everybody has a hard time
unbelieving something they once devoutly believed. In fact, I think that
very few succeed in doing that no matter how wrong they are. It’s just
too painful.

FN: Rather persistent patterns
in behavior can be observed just about anywhere, including in the
discourse on CSGNet so why harp on the variability of behavior? We
all know it varies. However, I for one happen to believe that it
varies within limits; we are restrained and constrained by our repertory
of behaviors and our belief systems.

BP: The persistent patterns you’re talking about are not patterns of
behavior. They are patterns of consequences controlled by whatever
behavior is needed at the moment to counteract disturbances while
producing the intended result. The actions of an organism do not repeat;
only their consequences repeat. The actions have to vary in exactly the
way needed to keep the consequences under control.

FN: I think I “get” PCT but I’ll
happily acknowledge that I will never fully grasp or appreciate it from
an engineering or scientific perspective. To me, it’s enough that
it makes eminent sense and is perfectly consistent with what I know of
human nature and behavior and the way the world works.

BP: The only barrier to “getting” PCT is a belief that
conflicts with it. It’s a simple idea that is easy to grasp – unless
grasping it brings into doubt some other idea that is considered
valuable. You can see it working with a pair of rubber bands or a few
coins scattered on a tabletop. The only problem is that at the same time
you’re seeing these simple demonstrations, something is whispering,
“But that’s a response to a discriminative stimulus; that other
thing reinforces the response; the contingencies in the environment
determine what actions will be generated; pulling on one end of the
rubber band is controlled by the knot’s moving away from the
dot.”
But you can’t have it both ways. Either that explanation is right, or the
PCT model is right. You can’t even describe the PCT analysis of behavior
without saying things that a behaviorist would consider to be exact
contradictions of behaviorism. The only way for a PCTer to avoid
attacking behaviorism is to say nothing about behaviorism or
PCT.

FN: The cognitive (i.e.,
computational) psychologists might or might not have something to offer;
I’m not smart enough or well enough read in that area to offer an
opinion. I do know that I attach not one shred of credence to the
view that we “compute” our actions.

BP: Neither do I, in the sense you mean.

FN: I do from time to take make
“calculated” moves but that’s a figure of speech referring to having
thought things through and having tried to ensure that things turn out as
planned (which sometimes they do and sometimes they
don’t).

BP: You haven’t calculated your actions; you’ve calculated the
intermediate consequences – perceived consequences, of course – that
are needed to bring about the final consequence. You can’t predict what
behaviors, what actions, will be needed to make each consequence occur in
turn. That will depend on the details, particularly on unpredictable
disturbances but also on little variations in the environment; where
things are, where you are, which way you and other things are moving. You
can plan walking along the deck of a ship at sea to get to the galley,
but in fact all you can really plan is moving along the deck in
that direction; the walking movements change all over the place as the
waves rock the ship. After a few weeks at sea, the sailor doesn’t even
notice the variations in the way he steps, but he probably never places
his feet the same way twice on his way to lunch.

When “things” don’t turn out as planned, do you just give up?
Or do you modify your actions and make them turn out that way anyhow? The
“things” you’re talking about are planned perceptions, not
planned actions.

FN: I did locate and read the
Bourbon-Powers paper demonstrating the inability of the S-R and cognitive
views to match human behavior and found it very interesting; however, I
do not find it very persuasive. In sum, what the paper says to me
is that “We wrote a computer program that operates in accordance with the
principles of PCT and that program very precisely mimics human
behavior. We also wrote computer programs that operate in
accordance with the principles of S-R behaviorism and cognitive
psychology and those programs do not accurately mimic human
behavior. Therefore, PCT is superior.”

BP: It’s not that the other models fail to mimic behavior accurately;
they generate the wrong behavior. The SR model says that movements
of the target are stimuli and movements of the subject’s hand are
responses determined by those stimuli. But when a random disturbance is
inserted between the hand movements and the cursor position, this model
would say that the hand movements should remain the same, since the
target movements remain the same. When the hand movements remain the
same, we see the cursor wobbling up and down relative to the target
because of the disturbance. But what we observe when the human being does
this task is that the cursor continues to track the target, and the hand
movements now differ from the target movements by an amount equal to (and
opposite to) the disturbance.

You don’t need to understand the computer program to understand that. I
can show you the S-R model behaving correctly, making the cursor move
exactly as the target moves. All three models do that much correctly. The
moment I introduce a disturbance of the cursor that offsets it from the
hand position (like a loose steering linkage), the S-R model starts to
behave wrong, because it fails to alter the hand movements to
compensate for the disturbance. The control system does alter its hand
movements and continues to make the cursor track the target.

FN: Without being able to
take apart and study those programs from an informed perspective (which I
can’t), I can’t gauge their significance or import. All I can do is
say, “If you say so” (which is where I am with respect to all scientific
research).

BP: Well, this is a problem and I’m not sure what to do with it. It comes
down to trust, I guess – you have to trust that both the behaviorist and
the control theorist are telling you honestly everything they did and
saw, not trying to hide anything and inviting inspection by others who do
have the required training. Generally I think that’s true.

But your common sense can still work. Think of driving a car. What makes
you turn the steering wheel back and forth even though you’re driving on
a straight level road? You can’t see anything happening, yet S-R theory
says that your steering movements have to be generated by some stimulus
acting on your sense organs. You won’t find any stimulus that corresponds
to each movement – because those movements are keeping a variable
crosswind from moving the car sideways enough to bother you. If there is
any stimulus, it is the small deviations of the car from the center of
its lane. But the response causes the same kind of effect: small
deviations of the car from the center of its lane, in the other
direction. So the stimulus and the response are measured by the same
aspect of the visible world. Either the crosswind alone or the steering
effect alone would cause far larger deviations, but these effects cancel,
leaving only a small residual effect. There’s no way to understand this
sort of closed-loop effect in terms of ordinary cause and
effect.

FN: Like many ordinary people,
when it comes to scientific research, about all I can do is try to
understand it, pay attention to what others are saying about it, and test
it against my own experiences of the world.

BP: That last one is what my demos are all about: having the experience
of controlling something. Words, mathematics, and programming can’t
provide that, though with programming you can give the person the ability
to control things on a computer screen.

FN: I was a pretty good
technician but I’m no engineer and if you have to be an engineer to fully
understand and appreciate PCT it’s beyond me and zillions of others
folks, too. If only the engineers can get PCT, there’s a big
problem because there are lots of engineers who suffer from the Rodney
Dangerfield syndrome, too.

BP: Yes. That’s why I keep saying don’t blame failures to communicate on
the reader. It’s really quite easy to think up demonstrations that are
perfectly understandable – all except for the modeling details. To
explain most things about PCT all you need is a pair of rubber
bands.

FN: Gee, maybe that’s the
problem with PCT. PCT doesn’t afford you a way off the hook for
your actions; indeed, PCT very clearly suggests that you do what you do
to get what you want and that implies an extraordinary degree of personal
accountability. The powers that be (no pun intended) won’t like
that.

Yep.

Best,

Bill P.

[BRUCE_STEPHANIE_ABBO]

[From Bruce Abbott (2009.09.29.1510 EDT)]

Bill Powers (2009.09.29.0759 MDT)

···

–

Fred Nickols (2009.09.29.0835 EST)

FN: It is one thing to demonstrate the superiority of PCT;
it is altogether a different thing to tear down other edifices.

FN: Too much time is spent attacking the old
views. Why? What does that gain?

FN: Consider the behaviorists’ mantra
that “Behavior is a function of its consequences.” I have no
problem with that assertion, including its validity in light of PCT. Is
not the closing of a gap between perception and reference signal a
“consequence” of behavior? Is not some behavior
“intended” to close such gaps? Does not the closing of gaps
confirm the actions that closed them (or as some behaviorists might say,
“reinforce” them)? Most behaviorists I know would readily
agree that the S-R model is kaput so why beat a dead horse?

BP: I’ve been accused of beating that dead horse for some 50 years. It’s still
not dead, though it’s not exactly well. In the case you speak of, behaviorists
have claimed since the days of Watson in 1910 or so that you don’t need to
guess at what goes on inside an organism. It’s sufficient to observe the
stimuli acting on it, and the actions that are thereby produced. That’s why it
is called behaviorism: behavior is what a scientist can observe; all the rest,
according to Watson, is “primitive superstition.” Watson and other
behaviorists since his era have spent a good deal of time attacking differing
old and new views, including my work, that suggest any modeling of the internal
parts of an organism that can’t be seen. They were doing that before I was
born; you can’t accuse me of starting the fracas.

BP: So a
pure-bred behaviorist would not admit that any of those gaps you mention is an
observable consequence of behavior. The consequences that a behaviorist like
Skinner meant are called reinforcers. When a rat presses a lever by accident
while moving randomly about in its cage, a piece of food drops into a dish:
that is an observable consequence of pressing the lever. The next time the rat
is in about the same situation, being affected by the same “discriminative
stimuli,” it is more likely to repeat the same action and produce another
edible consequence. Thus behavior is controlled by its consequences (I don’t
recall Skinner’s ever using the term “is a function of”). To quote
Skinner: Reinforcement brings behavior under the control of discriminative
stimuli. Behaviorists think that is different from S-R theory. It’s not.

BP: This
view of behavior, according to PCT, is exactly wrong. The consequence is
clearly controlled by the behavior; unless the lever is pressed no food will
appear. When the lever is pressed, the food does appear. Appearance of the
reinforcer is caused by, is a function of, the behavior. And as we now know,
varying the behavior is the means by which a control system comes to control
the occurrances of the reinforcer. And randomly varying the various parameters
of control is how the organism comes to behave in that way – not because of
any “strengthening” effect of the so-called reinforcer. The organism
reorganizes until its behavior is controlling the local environment to produce
the consequences that the organism intends to produce. A behaviorist hearing me
say that would tell me to wash my mouth out with soap.

I’m going to have to side with Fred on this one, Bill. I believe that in
your criticism of reinforcement theory, you are conflating two distinctly
different causal relationships. The first causal relationship is the one in
which the response produces the reinforcer. Here, the behavior (e.g., a lever
press) controls the consequence (e.g., food delivery). You present this fact as
if it is directly contrary to reinforcement theory, but no behaviorist I know
of would deny this fact. The lever press causes the food to be delivered.
Behaviorists refer to this relationship as a contingency.

The second causal relationship (the one asserted by
reinforcement theory) refers to the effect of experiencing that contingency on
the likelihood that the same response (the lever press) will occur again,
everything else being equal. Having experienced the contingency between
lever-pressing and food delivery, the hungry rat now presses the lever
frequently. As the reinforcement theorist would have it, the response has been
reinforced. This is exactly what one observes: behavior varies as a function of
its consequences, and neither Fred nor I see that as incompatible with PCT, in
so far as appearances are concerned.

What’s wrong with the behaviorist view not that it has
the causal relationships reversed, but that it does not offer a correct
description of the internal mechanism responsible for the appearances (if
indeed it offers any mechanism at all). According to PCT, the food-deprived rat
is experiencing an error in some control system; reorganization then “hunts”
at random for a set of connections and parameter values that reduce the error.
In the operant chamber, perhaps the only available mechanism that will
reestablish control is one that brings about approaching and depressing
the lever, then picking up and consuming the food pellet that was delivered by
the lever-press. We observe an increase in lever-pressing (“reinforcement”
of lever-pressing) and this increase is maintained until the underlying
controlled variable approaches its reference level, after which lever-pressing
ceases until the consumed food-stores are used up. (Behaviorists refer to this
latter phenomenon as “satiation.”)

In reinforcement theory, the reinforcer simply “strengthens”
the behavior that produces it; reinforcers have this effect because in the
evolutionary past of the organism, the behavioral changes that occur as a
result of this so-called “strengthening” have usually been
adaptive, for example resulting in the discovery of behaviors that permit the consumption
of food when the animal has been deprived of food. What is lacking in
reinforcement theory is an acknowledgement that when learning is involved, what
the reinforcement process brings into being is a control system. And once the
control system exists, it operates in order to keep its controlled variable at
or near some internally specified reference level. The behavior is being driven
by error, not the reinforcer.

Reinforcement theory says that a response is selected (for
repetition) when it produces a reinforcer, where the reinforcer is reinforcing
because of the evolutionary history of the organism and the organism’s
current biological state; In contrast, PCT says that a response is (in effect)
selected when it reduces error in an intrinsic variable. With regard to the PCT
view, a key problem is why the reorganizing system doesn’t end up
reorganizing away all sorts of functional systems while randomly changing the
structure and parameters of the control hierarchy; the reinforcement mechanism brings
about a change in organization only when the change succeeds in bringing about
better control over the variable currently in error. There may be a baby in
there along with the bath water.

Bruce

[Bill_Powers1]

[From Bill Powers (2009.09.29.1755 MDT)]

Bruce Abbott (2009.09.29.1510 EDT) --

BP earlier: This view of behavior, according to PCT, is exactly wrong. The consequence is clearly controlled by the behavior ...

BA: I'm going to have to side with Fred on this one, Bill. I believe that in your criticism of reinforcement theory, you are conflating two distinctly different causal relationships. The first causal relationship is the one in which the response produces the reinforcer.

...

The second causal relationship (the one asserted by reinforcement theory) refers to the effect of experiencing that contingency on the likelihood that the same response (the lever press) will occur again, everything else being equal.

I have to think about this for a while, Bruce -- not because I have any trouble seeing the flaw in your arguments, but because if I just tell you what it is, I will be talking from one side of a wall and you will be talking on the other side, defending your position. We have worked together for years now, most recently with you getting all my demos into shape to run on Windows machines and knowing every detail of them as well as I do. If you work out what the flaw is yourself, just from what you understand about control systems, which is a lot, I won't have to convince you of anything.

I'll give you one hint. The clue as to what is wrong is in Skinner's concept of the operant. William Verplank put it this way:

"The first of these concepts is that of the operant response, defined as a part of behavior (a) that is recurrently identifiable and hence enumerable; and (b) whose rate of occurrence can be determined as a systematic function of certain classes of environmental variables. These parts of behavior, or actions, are what we can see an animal perform repeatedly. They are not simple muscle twitches or limb movements. Rather, they are meaningful, repeated actions. They constitute bar presses rather than leg extensions, the picking up of food rather than digital flexions, the speaking of words rather than laryngeal contractions." http://web.utk.edu/~wverplan/biblio18.html

That pretty firmly establishes the nature of the operant in relation to PCT: it is not a behavior, but a controlled variable affected by a whole class of outputs including behaviors that are exact opposites. Because of that, the nature of the operant is such that the supposed effect of a reinforcement on the organism, and on the organism's behavior in the form of the operant, is a physical impossibility.

I'm going to bed. Work on the above for a while. I'm betting you'll see what I mean.

Best,

Bill P.

[_Boris_Hartman]

Bruce !

BA : "Reinforcement theory says that a response is selected (for repetition)
when it produces a reinforcer, where the reinforcer is reinforcing because
of the evolutionary history of the organism and the organism�s current
biological state; In contrast, PCT says that a response is (in effect)
selected when it reduces error in an intrinsic variable. With regard to the
PCT view, a key problem is why the reorganizing system doesn�t end up
reorganizing away all sorts of functional systems while randomly changing
the structure and parameters of the control hierarchy; the reinforcement
mechanism brings about a change in organization only when the change
succeeds in bringing about better control over the variable currently in
error. There may be a baby in there along with the bath water."

BH : Sorry Bruce to interrupt again. I know that I have no respect from you
and either from Bill. The relationship sadly became reversed.
It seems to me that you are making the same mistake as you did in theme "Kid
who doesn't talk". Apply everything what you wrote to that case and you'll
see where your flow of thoughts have a "behavioral" barrier.

I "jumpded" into discussion because it's really time that "boned" ghost of
Behaviorism pass to Museum. I won't make more interuption as I know I have
"wrong picture" about what homeostasis is (according to Bill), or what error
in intrinsic variables are (if there are any "error" according to Martin),
so I'm not any more in position to judge HPCT as it's obviously that I don't
understand it.

But I'm working for 30 years with all kind of children in teaching and
social programs, programs with children-convicts and I've gone through
project of ESF (Europian Social Fund). It must be 4.000 children or
something like that, I've been dealing with in these years. I can say that
behaviorist theory just doesn't produce plausible results. Behaviorist
observations or appearance of behavior (on the wrong theoretical bases) are
usually misleading and mostly produce wrong conclusions in REAL LIFE and I
could say it makes much more damage than good. But maybe experiancies
doesn't count much in such a briliant theoretical discussions.

Sorry for my language. And please don't make distortions to my surname as
you did before.

[BRUCE_STEPHANIE_ABBO]

[From Bruce Abbott (2007.09.30.0650 EDT)]

Boris Hartman:

BH: Bruce !

BH: Sorry Bruce to interrupt again. I know that I have no respect from you
and either from Bill. The relationship sadly became reversed.
It seems to me that you are making the same mistake as you did in theme "Kid
who doesn't talk". Apply everything what you wrote to that case and you'll
see where your flow of thoughts have a "behavioral" barrier.

BH: Sorry for my language. And please don't make distortions to my surname
as you did before.

You must have me confused with someone else. We have never communicated
before now.

Bruce

[BRUCE_STEPHANIE_ABBO]

[From Bruce Abbott (2009.09.30.0730 EDT)]

Bill Powers (2009.09.29.1755 MDT) --

Bruce Abbott (2009.09.29.1510 EDT)

BP earlier: This view of behavior, according to PCT, is exactly wrong.
The consequence is clearly controlled by the behavior ...

BA: I'm going to have to side with Fred on this one, Bill. I believe
that in your criticism of reinforcement theory, you are conflating two
distinctly different causal relationships. The first causal
relationship is the one in which the response produces the reinforcer.

...

The second causal relationship (the one asserted by reinforcement
theory) refers to the effect of experiencing that contingency on the
likelihood that the same response (the lever press) will occur again,
everything else being equal.

BP: I have to think about this for a while, Bruce -- not because I have any
trouble seeing the flaw in your arguments, but because if I just tell you
what it is, I will be talking from one side of a wall and you will be
talking on the other side, defending your position. We have worked together
for years now, most recently with you getting all my demos into shape to run
on Windows machines and knowing every detail of them as well as I do. If you
work out what the flaw is yourself, just from what you understand about
control systems, which is a lot, I won't have to convince you of anything.

BP: I'll give you one hint. The clue as to what is wrong is in Skinner's
concept of the operant. William Verplank put it this way:

BP: "The first of these concepts is that of the operant response, defined as
a part of behavior (a) that is recurrently identifiable and hence
enumerable; and (b) whose rate of occurrence can be determined as a
systematic function of certain classes of environmental variables.
These parts of behavior, or actions, are what we can see an animal perform
repeatedly. They are not simple muscle twitches or limb movements. Rather,
they are meaningful, repeated actions. They constitute bar presses rather
than leg extensions, the picking up of food rather than digital flexions,
the speaking of words rather than laryngeal contractions."
http://web.utk.edu/~wverplan/biblio18.html

BP: That pretty firmly establishes the nature of the operant in relation to
PCT: it is not a behavior, but a controlled variable affected by a whole
class of outputs including behaviors that are exact opposites.

BP: Because of that, the nature of the operant is such that the supposed
effect of a reinforcement on the organism, and on the organism's behavior in
the form of the operant, is a physical impossibility.

BP: I'm going to bed. Work on the above for a while. I'm betting you'll see
what I mean.

I haven't forgotten that the operant is defined by what it accomplishes
(e.g., lever depressed) rather than by what the organism does to achieve it.
It is goal-directed behavior. The rat in the operant chamber already
possesses a set of control systems to actuate its muscles as needed to move
itself about and maintain posture while compensating for the effects of
disturbances. These are integrated in various ways into other control
systems that generate the patterns of muscle contractions required for
locomotion and positioning so as to accomplish a goal. To obtain food in the
operant chamber, the rat needs to learn what to do. Some body part must
contact the lever and exert some force against it. I'm not convinced that
rats always get the right idea -- that the lever needs to be pressed down --
but what they do learn usually succeeds in getting the lever depressed
enough to activate the switch and thus trigger food delivery. One could
imagine that reorganization in the PCT sense isn't involved at all. Existing
control systems are not being modified; rather, a pattern of
reference-setting is being memorized. The result, however, is a new control
system that allows the rat to control food delivery in the operant chamber.
The new system varies references to the existing movement systems to bring
about approach to the lever and the other acts that get the lever pressed.
The rat learns to press the lever because it has associated contacting the
lever with the subsequent immediate delivery of food, and because food
consumption is the means by which the rat can correct an error in another
control system -- the one associated with feelings of hunger.

That's as far as I can take my thinking at present as I have to get ready
for work. How am I doing so far?

Bruce

[_Boris_Hartman]

Hi Bruce !

I'm really sorry to confuse you with somebody else. Are you maybe the Bruce
Abbott who wrote the Synopsis of William T. Powers : PCT ? If you are, you
would make me a great pleasure if you could tell me, when did you wrote that
Synopsis.

Once more, sorry for the confusion.

Best,

Boris

[BRUCE_STEPHANIE_ABBO]

[From Bruce Abbott (2009.09.30.1005 EDT)]

Yes, I wrote the synopsis. I had to check the file date to determine when I wrote it. It was written on May 21st 2001.

Bruce

···

Sep 30, 2009 08:24:20 AM, CSGNET@LISTSERV.ILLINOIS.EDU wrote:

Hi Bruce !

I’m really sorry to confuse you with somebody else. Are you maybe the Bruce
Abbott who wrote the Synopsis of William T. Powers : PCT ? If you are, you
would make me a great pleasure if you could tell me, when did you wrote that
Synopsis.

Once more, sorry for the confusion.

Best,

Boris

[rsmarken]

[From Rick Marken (2009.09.30.0815)]

Bill Powers (2009.09.29.0759 MDT)–

BP: I’ve been accused of beating that dead horse for some 50 years. It’s
still not dead, though it’s not exactly well.

The “dead horse” you have been beating is the open-loop causal model of behavior, which made it’s first explicit appearance as Behaviorism but still lives on in all branches of psychology, as evidenced by the fact that all research in psychology is based on this model. This is the point I made in my overwhelmingly ignored “Revolution” paper.

So while Behaviorism may be a dead horse, the open-loop causal model, for which Behaviorism, Cognitive Psychology, Cognitive Neuropsychology, Self-Regulation Theory, etc, are merely synecdoches, is still the basic model of behavioral organization in psychology. Thus, Behaviorism – in the form of the open-loop causal model – is alive and well, even among dedicated members of CSG. This is why your well-meaning exhortation to “give up internal nit-picking, sniping, and nostalgia for the past,[and] turn our attention outward, and begin a project to persuade all the behavioral sciences that the control system model of living systems is the only correct one among all the others” is as big a fantasy as Obama’s fantasy of bipartisanship.

Best regards

Rick

···

–
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Bill_Powers1]

[From Bill Powers (2009.09.30.0826 MDT)]

Bruce Abbott (2009.09.30.0730 EDT) --

I haven't forgotten that the operant is defined by what it accomplishes
(e.g., lever depressed) rather than by what the organism does to achieve it.
It is goal-directed behavior.

Is it a behavior, or is it a goal, a reference condition that is brought into being by variations in lower-order actions? Suppose the rat is facing to the left of the lever in one case, and to the right of the lever in the next one. The motor action required to get to the lever in the first case is to turn right; in the second case it is to turn left. The rat gets to the lever in either case, so the operant could be defined as "motion toward the lever." However, there is only one reinforcement at a time, so which motor process gets reinforced -- turning left, or turning right? If all the motor acts included in the operant class get reinforced, what is the net result? If only some of them get reinforced, how does it happen that only the appropriate one occurs the next time, even if it wasn't the one reinforced the last time?

The rat in the operant chamber already
possesses a set of control systems to actuate its muscles as needed to move
itself about and maintain posture while compensating for the effects of
disturbances. These are integrated in various ways into other control
systems that generate the patterns of muscle contractions required for
locomotion and positioning so as to accomplish a goal.

In PCT, variables such as posture, locomotion, and positioning are controlled variables, as you say. They can be controlled as a means of controlling higher-level -- or just "other" -- variables. You say that the control systems controlling these variables "are integrated in various ways into other control
systems that generate the patterns of muscle contractions required for
locomotion and positioning so as to accomplish a goal." What does this mean? How, for example, would a control system required for positioning be "integrated into" the control systems that use muscle contractions to adjust limb forces? What does "integrated into" mean?

To obtain food in the operant chamber, the rat needs to learn what to do. Some body part must contact the lever and exert some force against it.

The rat may need to learn what to do, but how does that influence what it learns? How does a body part contacting the lever and exerting some force on it produce learning? And what is learned? In the example above, what has to be learned is that if you're left of the lever you should turn right, but if you're to the right of it, you should turn left. That's not a behavior; it's a program-level control system, isn't it? Exactly how does reinforcement act to reinforce this program, which isn't any particular behavior?

I'm not convinced that rats always get the right idea -- that the lever needs to be pressed down -- but what they do learn usually succeeds in getting the lever depressed enough to activate the switch and thus trigger food delivery. One could imagine that reorganization in the PCT sense isn't involved at all.

All right, let's say there is no change in organization in the rat's nervous system. How does it happen, then, that what the rat does in the same situation changes? I can agree that what the rat learns to control is not necessarily a "lever press", but only some posture or force or movement in relation to the apparatus that happens to generate a depression of the lever fairly reliably. How is that learned, if not through some kind of change in its neural networks, when the rat couldn't produce that consequence of its actions before and then becomes able to produce it?

Existing control systems are not being modified; rather, a pattern of
reference-setting is being memorized.

What is it (according to PCT) that generates that pattern of reference settings for lower control systems? Isn't it the output function in a higher-level control system? Isn't that more or less what the control systems in "ArmControlReorg" become able to do after a period of reorganization?

The result, however, is a new control system that allows the rat to control food delivery in the operant chamber. The new system varies references to the existing movement systems to bring about approach to the lever and the other acts that get the lever pressed.

I agree, the result is a new control system. But do you still say that this new control system comes into being without any kind of reorganization happening?

Also, I remind you that generating a memorized pattern of reference settings "to actuate its muscles as needed to move itself about and maintain posture while compensating for the effects of disturbances" can't possibly work. What memorized pattern of "reference settings to actuate muscles" do you use to drive to your office from home? A control system doesn't memorize a set of output actions that will do all these things: there is no one set of output actions that will accomplish that. A "pattern of reference settings" is the output of a higher control system which varies them (as you said yourself) according to changing events in the environment -- disturbances, and variations in the position, location, speed, and direction of movement of the body.

The rat learns to press the lever because it has associated contacting the
lever with the subsequent immediate delivery of food,

How does that "because" work? If the rat associates contacting the lever with the left side of its body with food, how does that teach it to contact the lever with the right side of its body the next time? The movements and forces involved are opposite to each other.

and because food consumption is the means by which the rat can correct an error in another control system -- the one associated with feelings of hunger.

But this "because" is just as empty of meaning as the previous one. How does the fact that food consumption corrects an error in another control system have any effect on the rat's behavior? You're assuming that reorganization occurs but you are also assuming it doesn't occur.

That's as far as I can take my thinking at present as I have to get ready
for work. How am I doing so far?

Grade for expressing the behaviorist view, A+. Grade for using what you know about PCT to discover the main problem, C-.

The main problem is that THERE IS NO BEHAVIOR AT ANY LEVEL OF ORGANIZATION SUCH THAT REPEATING IT WILL USUALLY PRODUCE THE SAME CONSEQUENCE. In fact, to make any consequence repeat, it is normally necessary that the behavior change from one instance to the next, not remain the same. Therefore INCREASING THE PROBABILITY OF A SPECIFIC BEHAVIOR IS THE WRONG METHOD FOR OBTAINING REINFORCEMENTS (that is, for controlling the intake of food or water, or creating any other state of the environment preferred by the organism).

It is very very hard to change a system concept even when you're trying to do that.

Best,

Bill P.

[_Boris_Hartman]

Hi Bruce !

Thank you for information. Synopsis is a very good work. Please accept again
my apology for confusion and accusations I made.

Best,

Boris

[BRUCE_STEPHANIE_ABBO]

[From Bruce Abbott (2009.09.30.1935 EDT)]

Bill Powers (2009.09.30.0826 MDT) --

Judging from your replies to my previous post, I'd say that I'm not doing a
good job communicating what I have in mind. I'll try to do better. And for
anyone else who may be following this thread, the example of a rat in an
operant chamber is simply a well-defined situation that could represent any
of a vast number of similar ones that individuals face in the real world, of
which finding your next meal when you have no idea how to get it is one
example.

Bruce Abbott (2009.09.30.0730 EDT)

I haven't forgotten that the operant is defined by what it accomplishes
(e.g., lever depressed) rather than by what the organism does to achieve

it.

It is goal-directed behavior.

BP: Is it a behavior, or is it a goal, a reference condition that is brought
into being by variations in lower-order actions?

The latter, of course.

BP: Suppose the rat is facing to the left of the lever in one case, and to
the right of the lever in the next one. The motor action required to get to
the lever in the first case is to turn right; in the second case it is to
turn left. The rat gets to the lever in either case, so the operant could be
defined as "motion toward the lever." However, there is only one
reinforcement at a time, so which motor process gets reinforced-- turning
left, or turning right? If all the motor acts included in the operant class
get reinforced, what is the net result? If only some of them get reinforced,
how does it happen that only the appropriate one occurs the next time, even
if it wasn't the one reinforced the last time?

You seem to believe that I'm defending the behaviorist definition of the
operant as a class of behavior. I'm not. What's getting established is a
reference, initially for being near or in contact with the lever. The
control system that brings this about (despite differences in starting
position and disturbance patterns) in turn sets references for
next-level-down control systems and so on down the hierarchy. The behavior
that results is approach to the lever if the rat begins at a distance that
requires approach, and perhaps contact with the lever.

The rat in the operant chamber already possesses a set of control
systems to actuate its muscles as needed to move itself about and
maintain posture while compensating for the effects of disturbances.
These are integrated in various ways into other control systems that
generate the patterns of muscle contractions required for locomotion
and positioning so as to accomplish a goal.

In PCT, variables such as posture, locomotion, and positioning are
controlled variables, as you say. They can be controlled as a means of
controlling higher-level -- or just "other" -- variables. You say that the
control systems controlling these variables "are integrated in various ways
into other control systems that generate the patterns of muscle contractions
required for locomotion and positioning so as to accomplish a goal." What
does this mean? How, for example, would a control system required for
positioning be "integrated into" the control systems that use muscle
contractions to adjust limb forces? What does "integrated into" mean?

See above. I put it badly, but what I had in mind was the operation of the
ordinary control hierarchy. Apparently my words failed to communicate that
adequately.

To obtain food in the operant chamber, the rat needs to learn what to
do. Some body part must contact the lever and exert some force against it.

BP: The rat may need to learn what to do, but how does that influence what
it learns? How does a body part contacting the lever and exerting some force
on it produce learning? And what is learned? In the example above, what has
to be learned is that if you're left of the lever you should turn right, but
if you're to the right of it, you should turn left. That's not a behavior;
it's a program-level control system, isn't it? Exactly how does
reinforcement act to reinforce this program, which isn't any particular
behavior?

I'm not arguing for a reinforcement interpretation, although apparently you
believe that I am. I'm trying to understand the process involved as informed
by PCT, although I suspect that something different from the classic
PCT-style reorganization may be involved.

I'll try to address your questions. Having a body part contact the lever and
exert some force on it does not produce learning, so I do not have to
explain how it does so. Being hungry is a state the rat has already learned
how to deal with: start looking for food and when you find some, consume it.
I'm sure that a PCT model of this can be constructed, although I won't do so
here. We might call this a "hunger control system"; its reference perception
is zero hunger. The food deprivation imposed by the experimenter has created
an error (hunger is greater than zero), so the system is now generating
foraging behavior by manipulating the references of next-level-down control
systems. In the course of this activity the rat chances to encounter the
lever, closing its switch. The apparatus then delivers a food pellet,
allowing the program to execute the next phase of control: approaching,
picking up, and consuming the pellet.

The discovery of food is the goal that the foraging system was organized to
accomplish. Consuming the food is the means by which hunger will be brought
back to its reference level. I haven't thought completely through the
details of a PCT model that accomplishes this -- in fact, writing this train
of thought down is helping me to see the problem more clearly. At this point
all I can say is the following: (a) it is hunger that establishes the goal
of discovering food, (b) the fact that food is currently unavailable
constitutes error in the system that produces foraging behavior as the means
of correcting this error, and (c) the consuming the food produced is the
means by which the error in the hunger system gets corrected.

This brings me to your last question. The delivery of food does not
"reinforce" a program; it associates the achievement of the food-foraging
system's goal (perception of available food) with the activity-generated
perceptions that occurred just before the pellet appeared in the food cup
--what the rat perceives itself to have been doing.

Such associations generally require more than one experience of such a
relationship. Which if any of those time-varying perceptions must be
reproduced in order to produce another pellet? It may take several such
coincidences before the rat's brain can isolate the critical perceptions
that must be reproduced, especially since variable starting positions will
result in variations in the perceptions existing at the time of food
delivery. Those perceptions become the reference perceptions for the
relevant control systems, and those systems produce behaviors that bring
those perceptions back to their reference values.

I'm not convinced that rats always get the right idea -- that the
lever needs to be pressed down -- but what they do learn usually
succeeds in getting the lever depressed enough to activate the switch
and thus trigger food delivery. One could imagine that reorganization
in the PCT sense isn't involved at all.

BP: All right, let's say there is no change in organization in the rat's
nervous system. How does it happen, then, that what the rat does in the same
situation changes? I can agree that what the rat learns to control is not
necessarily a "lever press", but only some posture or force or movement in
relation to the apparatus that happens to generate a depression of the lever
fairly reliably. How is that learned, if not through some kind of change in
its neural networks, when the rat couldn't produce that consequence of its
actions before and then becomes able to produce it?

I would speculate that there is some kind of look-back memory function that
scans back for perceptual values that were present at the time the pellet
was delivered, and an averaging mechanism that allows the relevant
perceptions to be extracted from the irrelevant ones across repeated pellet
deliveries. I'm not sure whether traditional ecoli-style reorganization
would be necessary. It may be that establishing the proper reference values
for existing control systems would be sufficient, and this requires only the
storage, averaging, and retrieval of reference values from memory. Although
encoding memories requires changes in brain "wiring," I assume that these
would not qualify as reorganization as currently defined. But perhaps
reorganization would also be required; maybe you have a clearly picture here
than I am currently able to generate.

I'm going to skip over the next several paragraphs of your reply as I think
I've already addressed the questions they pose.

That's as far as I can take my thinking at present as I have to get
ready for work. How am I doing so far?

BP: Grade for expressing the behaviorist view, A+. Grade for using what you
know about PCT to discover the main problem, C-.

Ouch.

BP: The main problem is that THERE IS NO BEHAVIOR AT ANY LEVEL OF
ORGANIZATION SUCH THAT REPEATING IT WILL USUALLY PRODUCE THE SAME
CONSEQUENCE. In fact, to make any consequence repeat, it is normally
necessary that the behavior change from one instance to the next, not remain
the same. Therefore INCREASING THE PROBABILITY OF A SPECIFIC BEHAVIOR IS THE
WRONG METHOD FOR OBTAINING REINFORCEMENTS (that is, for controlling the
intake of food or water, or creating any other state of the environment
preferred by the organism).

Ouch, ouch. I know what the main problem is. My proposal doesn't entail
repeating a behavior (by which I assume you mean an action, as opposed to an
act). I'm talking about repeating certain perceptions. Perceptions are
repeated by setting references for perceiving them in the appropriate
control systems. They are in this way repeated despite differences in
starting position or disturbances, so long as the systems have the means to
counteract these effects.

BP: It is very very hard to change a system concept even when you're trying
to do that.

Ouch again. Perhaps I'm fooling myself, but I don't see anything here that
resembles a defense of reinforcement theory. Instead, I've been trying to
solve the problem of how, when the rat is hungry, the delivery of a food
pellet can bring about the repetition of behavioral acts (like
lever-pressing) that succeed in producing the pellet. In doing so, I've
appealed only to mechanisms described in B:CP (hierarchy of control systems,
reference levels retrieved from memory), or to compatible mechanisms (an
averaging mechanism). You may be able to reason through this better than I
can and can see flaws that are not apparent to me at present. Fine; it's
just a proposal offered for debate: if you think it's unworkable (or
unnecessary), I'm interested in hearing your reasoning to that end.

Bruce

[Bill_Powers1]

From Bill Powers (2009.09.30.1837 MDT)]

From Bruce Abbott (2009.09.30.1935 EDT) –

BA: I don’t see anything here
that

resembles a defense of reinforcement theory.

BP: The aspect of reinforcement theory I’m focusing on is the idea that
reinforcements increase the probability of repeating a particular
behavior (whether spontaneously or as a response to a discriminative
stimulus or any other way).

BA: Instead, I’ve been
trying to

solve the problem of how, when the rat is hungry, the delivery of a
food

pellet can bring about the repetition of behavioral acts (like

lever-pressing) that succeed in producing the pellet.

BP: Yes, this is exactly the problem I have in mind. What you call a
behavioral act is what Skinner called the operant, and what I call a
controlled variable. A behavioral act (etc) is defined as a consequence
of some kind of motor behavior, where the kind of motor behavior is
defined only as the class of all prior actions that could, under any
circmstances likely to occur, generate the required consequence.

This raises the question of how anything could influence the organism in
such a way as to make more frequent not a particular motor action, but a
particular consequence that many different motor actions can bring about.
I boiled that down to an example in which the act of pressing a lever can
be brought about on different occasions by moving either to the left or
to the right. In any given instance only one of those motor actions will
work, yet reinforcing either action (making it more probable) would not
be appropriate when the opposite direction of action is the next action
needed. Skinner tried to get around this problem, which I’m sure he must
have at least subconsciously recognized, by inventing the operant which
contains both directions of movement as causes, and thus lets him off the
hook as to which direction will be produced on any one occasion. But
unless we just say that the correct particular direction is selected on
each occasion by magic or ESP, we can’t explain how it is selected. That
is the fatal flaw in Skinner’s concept of reinforcement.

I think you’re trying to assimilate the concept of reinforcement into PCT
by saying that what is reinforced is not a motor action, but a
higher-level action of setting a reference level for a pressed lever –
for an “act.” The pressed lever occurs accidentally during a
search program, and the reinforcer that results supposedly increases the
probability that that reference condition (for whatever the critical
lower-level consequence is) will be generated again. The lower-level
system that causes the lever to be pressed is a control system, so it
automatically produces the right or left movements needed to correct any
difference between the reference signal and the perceptual signal, the
perception of a lever moving downward, or of something else happening
that causes the lever to move downward if only as a side effect. That
takes care of the defect in Skinner’s idea of an operant.

Notice that reorganization theory follows essentially the same route, but
the explanation is the mirror image of the reinforcement explanation.
Rather than saying that setting a given reference becomes more probable
when the reinforcer occurs, reorganization theory says that
changing the reference signal becomes less likely when the
reinforcer occurs. Of course we can’t go on using the term reinforcer any
more, because the significant consequence does not make anything more
likely to occur; it makes subsequent changes less likely to occur. It
does this, in my model, by reducing the intrinsic error that caused
reorganization to start in the first place. That is the connection with
evolutionary history that you propose.

I’ll grant you this: if there is a logical system already in existence
that can reason out the relationship between an act and its consequence,
there can be a systematic way of selecting the required state of the
right controlled variable to produce what we call a reinforcer (or to
remove an aversive stimulus, an unwanted effect). Such an intelligent
system could, as you suggest, recall what action was going on when the
desired reinforcer appeared, and what the surrounding circumstances (or
“discriminative stimuli”) were, and recreate that reference
condition the next time the lack of food or whatever occurs and the same
discriminative stimuli are present. I would venture to say that most
systematic ways of correcting errors come about in this way – through
replacing a crude built-in way of acquiring new skills by far more
efficient ways.

But I have not tried to explain how systematic and logical learning takes
place because I think those things have to be learned before they can be
used. Reorganization theory is aimed at explaining how the correct
behaviors can be learned before there are any significant systematic or
intelligent ways of solving problems, or in organisms that never do
develop such capabilities. You have seen the result in Chapters from 7 on
in LCS3. Reorganization theory requires an absolute minimum of
intelligence or reasoning power – really none at all. Yet it works, as
you have no doubt concluded by now along with me, infallibly.

BA: In doing so, I’ve appealed
only to mechanisms described in B:CP (hierarchy of control systems,
reference levels retrieved from memory), or to compatible mechanisms (an
averaging mechanism). You may be able to reason through this better than
I can and can see flaws that are not apparent to me at present. Fine;
it’s just a proposal offered for debate: if you think it’s unworkable (or
unnecessary), I’m interested in hearing your reasoning to that
end.

BP: OK, you have it.

By the time you make all the changes in behaviorism that bring it into
line with PCT, what you end up with is PCT, not behaviorism. PCT is not a
descendant, but a mutation.

Now, my old friend, I turn it over to you. What we need now is a program
that illustrates the behavioristic view, and then through a series of
simple and easily comprehended modifications transforms it into the PCT
view. Do that, and you will become famous.

Best,

Bill P.

[Bjorn]

[From Bjorn Simonsen (2008.10.03,1145 EU ST)]

I think PCT and many other theories and models don’t go inside people I talk to because they don’t listen. They just think about what theyshall say themselves.

bjorn