Grudging Acknowledgement

[From Bruce Abbott (2000.11.30.1130 EST)]

Bruce Gregory (2000.1130.0445)]

Much as I apparently hate Rick's guts, I have to admit that he doesn't
ignore difficult or potentially embarrassing questions.

I haven't seen anyone ask Rick any difficult or potentially embarrassing
questions.

I have seen Rick ask me questions that suggest to me that he hasn't been
paying close attention to what I've been saying, or perhaps he has and I
just haven't been clear.

For the record, I don't hate Rick or his guts, even though we've had our
sometimes rather spirited disagreements. But have a little patience -- I'm
not a retiree and still have work to do that sometimes must take priority
over CSGnet. At the moment I've got a serious deadline Monday I'm trying to
beat, and I'm going to be a little busy until I've finished what I need to
get done.

With respect to Rick's proposed test, here's question to stimulate thinking
-- is pressing the air the same as pressing the lever? Is it the act of
pushing down that was reinforced before Rick's "disturbance" was introduced,
or the act of pushing down the lever?

Bruce A.

[From Rick Marken (2000.11.30.0950)]

Bruce Abbott (2000.11.30.1130 EST)--

With respect to Rick's proposed test, here's question to
stimulate thinking -- is pressing the air the same as
pressing the lever? Is it the act of pushing down that
was reinforced before Rick's "disturbance" was introduced,
or the act of pushing down the lever?

This is _the_ question! And it is why we need a reinforcement
model.

I believe that the reinforcement theorist would say that it is
"pushing down the lever" that is reinforced, regardless of the
detailed responses (acts) that produce this result on each occasion.
I think this is the reason why reinforcement theorists invented
the concept of the "operant". They invented a word to explain away
an important problem for their theory and, at the same time,
make their theory (almost) unfalsifiable. The problem is the one
I mentioned in my previous post, viz., if what is selected by
reinforcement is "pushing down the lever" then how does the
reinforcement know which response (act) the animal should use
on any particular occasion to get the lever down? Each of the
acts that produced the result ("pushing down the lever") during
training is equally likely to occur after training and, thus, is
likely to occur in circumstances where it will not produce the
result ("pushing down the lever") that produces reinforcement.

The problem with the concept of the "operant" is that it assumes
that all responses (acts) -- all means of getting the lever down --
are created equal. There is no notion that the detailed acts that
produce a result (like "pushing down the lever") are precisely those
needed to compensate for disturbances (like changes in orientation
relative to the lever) that would prevent repetition of the result.
from the reinforcement theory point of view, pushing the lever down
with the right paw, the left paw, the mouth or the rump are just
different but equivalent ways of producing a particular result
(lever down) that produces reinforcement. But, as I mentioned in
my earlier post, these different acts are _not_ equivalent. Pushing
with the left arm, right arm, mouth or rump will only produce
"lever down" if the animal is in the appropriate orientation with
respect to the lever.

But I don't think that even the concept of the "operant" can save
reinforcement theory from my experiment. Even if we grant that
reinforcement magically selects "pushing down the lever", I think
it can only select that result using the acts that were reinforced
during the training session. That is, if the lever was pressed
only with the right paw and rump during training then the "pushing
down the lever" operant that was reinforced only included pushing
the lever down with the right paw and rump; pushing with the left
paw and teeth was never included in the operant because these
responses were never reinforced during training. So if, after
testing, circumstances are arranged such that only a never before
used response (act), such as pressing with the left paw, will
produce a particular result ("pushing down the lever") then that
result ("pushing down the lever") should not occur because "pushing
down the lever with the left paw" was never reinforced.

This is what my test of reinforcement would require. After training,
the result that produces reinforcement could only be produced by a
response (act) that was not previously reinforced (was not part of
the operant selected by reinforcement). So the reinforcement theory
prediction is that, in this case, the subject will not be able to
produce the result that produces reinforcement. If the subject can
produce the result using never before used acts (responses) -- acts
that were not part of the reinforced "operant" -- then reinforcement
theory would be rejected.

Best

Rick

···

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

[From Bruce Abbott (2000.11.30.1300 EST)]

I can answer this one quickly, so . . .

Bruce Nevin (2000.11.30.1151 EST) --

Bruce Abbott (2000.11.30.1130 EST)

[...] is pressing the air the same as pressing the lever?

The same for whom? The observer, or the rat?

The rat.

Gotta go . .

Bruce A.

[From Bruce Gregory (2000.1130.1325)]

Bruce Abbott (2000.11.30.1130 EST)

I haven't seen anyone ask Rick any difficult or potentially embarrassing
questions.

I must be losing my touch.

BG

[From Bruce Gregory (2000.1130.1340)]

Bruce Abbott (2000.11.30.1130 EST)

For the record, I don't hate Rick or his guts, even though we've had our
sometimes rather spirited disagreements.

Different Bruce. I'm the one who hates Ricks guts. Actually, I'm not
partial to anyone's guts, not even my own.

With respect to Rick's proposed test, here's question to stimulate thinking
-- is pressing the air the same as pressing the lever? Is it the act of
pushing down that was reinforced before Rick's "disturbance" was introduced,
or the act of pushing down the lever?

Whichever one will "explain" the outcome of the new experiment.

BG

[From Bruce Nevin (2000.11.30.1151 EST)]

Bruce Abbott (2000.11.30.1130 EST)--

···

At 11:32 AM 11/30/2000 -0500, Abbott, Bruce wrote:

[...] is pressing the air the same as pressing the lever?

The same for whom? The observer, or the rat?

         Bruce Nevin

[From Bruce Nevin (2000.11.30.1441 EST)]

Bruce Abbott (2000.11.30.1300 EST)--

>>[...] is pressing the air the same as pressing the lever?

>The same for whom? The observer, or the rat?

The rat.

Simple and quick answer:

If the rat is "pressing" in order to control the same perception in both cases then pressing the air is the same as pressing the lever.

If the rat is pressing the air for one purpose and pressing the air for another, then pressing the air is different from pressing the lever.

What is the controlled variable?

         Bruce Nevin

···

At 01:01 PM 11/30/2000 -0500, Abbott, Bruce wrote:

[From Bruce Nevin (2000.11.30.1444 EST)]

Bruce Nevin (2000.11.30.1441 EST)--

If the rat is pressing the air for one purpose and pressing the

lever

for another, then pressing the air is different from pressing the lever.

Please excuse the typo.

···

At 02:41 PM 11/30/2000 -0800, Bruce Nevin wrote:

What is the controlled variable?

        Bruce Nevin

[From Bill Powers (2000.11.30.1701 MST)]

... The problem is the one
I mentioned in my previous post, viz., if what is selected by
reinforcement is "pushing down the lever" then how does the
reinforcement know which response (act) the animal should use
on any particular occasion to get the lever down? Each of the
acts that produced the result ("pushing down the lever") during
training is equally likely to occur after training and, thus, is
likely to occur in circumstances where it will not produce the
result ("pushing down the lever") that produces reinforcement.

This is a profound observation that I have not seen anyone mention before.
It may be the missing piece in this puzzle.

The problem with the concept of the operant is precisely that it does not
tie "behavior" to any particular action, but only to a class of actions all
of which have the same outcome. But there is no such class unless you
include variations in circumstances; one act, if repeated under ALL the
same circumstances, must always produce the same result in a deterministic
universe, and if ALL circumstances remain the same, a different act must
produce a different result. It follows that if there is a class of
different actions that produce the same result, each one can produce the
same result only under its own set of circumstances (you make this point
very clearly). Thus the class is an artifact constructed by looking at
different actions produced under just the right different circumstances
that they happen to have the same result.

What you have said here shows that if a reinforcement increases the
probability that an _operant_ will occur, it must reinforce all the
specific actions that are included under the definition of the operant:
that is, all of the actions that can produce the reinforcing result. And
this means that it cannot reinforce just the specific action that, under
current conditions, is required to produce the result. It will also
reinforce all the other actions that, under other circumstances, might
produce the same result, but in _this_ circumstance will _not_ produce that
result.

It might be argued that a change in circumstances will alter a
discriminative stimulus, which could steer the reinforcement effect to the
appropriate behavior. But the circumstances that determine the effect of an
action are not tied to specific discriminative stimuli; many are totally
undetectable, and even if there are detectable ones, they cannot be
detected with enough _quantitative_ significance to do the necessary
fine-tuning of the actions.

The problem with the concept of the "operant" is that it assumes
that all responses (acts) -- all means of getting the lever down --
are created equal. There is no notion that the detailed acts that
produce a result (like "pushing down the lever") are precisely those
needed to compensate for disturbances (like changes in orientation
relative to the lever) that would prevent repetition of the result.

And that, of course, is the next major problem, which is that most
disturbances are unheralded by any stimuli at all except the change in the
result itself.

But I don't think that even the concept of the "operant" can save
reinforcement theory from my experiment. Even if we grant that
reinforcement magically selects "pushing down the lever", I think
it can only select that result using the acts that were reinforced
during the training session. That is, if the lever was pressed
only with the right paw and rump during training then the "pushing
down the lever" operant that was reinforced only included pushing
the lever down with the right paw and rump; pushing with the left
paw and teeth was never included in the operant because these
responses were never reinforced during training. So if, after
testing, circumstances are arranged such that only a never before
used response (act), such as pressing with the left paw, will
produce a particular result ("pushing down the lever") then that
result ("pushing down the lever") should not occur because "pushing
down the lever with the left paw" was never reinforced.

I can see problems here. It's possible for a rat to learn to control for
sitting on the lever as a means of making food appear. It follows that it
would not press the lever with its paws or teeth once sitting was firmly
established as the means. Of course this is a problem for the concept of
the operant, too, because it is unlikely that the rat will use any action
that is a member of the operant class but the specific one that it has
learned to use.

In order for the rat to use a different action to overcome a novel
disturbance (one that, say, prevented sitting from being effective but
allowed pawing to work) it would probably have to reorganize some more. We
would probably have much more luck if the disturbance required altering the
amount or direction of the specific kind of action that had been learned.
Given that a control system is what was learned, the right disturbance
should result in _immediate_ cessation of the action, or even reversal of
its direction, with no time for reinforcements to alter probabilities of
acting.

Of course if it proved that a rat would _immediately_ find a substitute
action that would have the same result, your case would be made in spades,
doubled and redoubled.

Best,

Bill P.

[From Chris Cherpas (2000.12.01.0900 PT)]

Bill Powers (2000.11.30.1701 MST)--

What you have said here shows that if a reinforcement increases the
probability that an _operant_ will occur, it must reinforce all the
specific actions that are included under the definition of the operant:
that is, all of the actions that can produce the reinforcing result. And
this means that it cannot reinforce just the specific action that, under
current conditions, is required to produce the result. It will also
reinforce all the other actions that, under other circumstances, might
produce the same result, but in _this_ circumstance will _not_ produce
that result.

In the world of the operant it is assumed that there is an
uncommitted pool of ("random") operant behavior. Reinforcement
supposedly selects the "behavioral atoms" (I know, mixed genetic
and physics metaphors) that produce it -- or its non-occurrence
selects against those that do not.

Variations will always remain because, relative to whatever occurred
at the precise moment of transition from (for example) the no-food to
the food state, there are variations in behavior that are more distant,
and hence, less committed to what produces the food. This is what
some, more generalized, versions of the matching law have tried to state.

Othewise, for one thing, "shaping" would be impossible. Speaking of
"random" operant behavior might seem to be a contradiction in terms,
but it is a necessary part of the concept.

2. Stereotypy is sometimes said to be one result of reinforcement.
Here we have the extreme where the amount of uncommitted operant
behavior is at a minimum.

Hope this helps more than it confuses!

Best regards,
cc

[From Bruce Gregory (2000.1201.1554)]

Chris Cherpas (2000.12.01.0900 PT)

In the world of the operant it is assumed that there is an
uncommitted pool of ("random") operant behavior. Reinforcement
supposedly selects the "behavioral atoms" (I know, mixed genetic
and physics metaphors) that produce it -- or its non-occurrence
selects against those that do not.

Variations will always remain because, relative to whatever occurred
at the precise moment of transition from (for example) the no-food to
the food state, there are variations in behavior that are more distant,
and hence, less committed to what produces the food. This is what
some, more generalized, versions of the matching law have tried to state.

Othewise, for one thing, "shaping" would be impossible. Speaking of
"random" operant behavior might seem to be a contradiction in terms,
but it is a necessary part of the concept.

2. Stereotypy is sometimes said to be one result of reinforcement.
Here we have the extreme where the amount of uncommitted operant
behavior is at a minimum.

Bruce A., this is what I mean by a just-so story.

BG

[From Bruce Abbott (2000.02.1155 EST)]

Bruce Gregory (2000.1201.1554) --

Bruce A., this is what I mean by a just-so story.

I agree. It is an explanation designed to fit both the theoretical view of
the person who developed it and the existing observations. It's important
to be able to find a satisfactory (logically consistent and consistent with
previous observations) way in which one's current view can explain the data,
because if you can't, there may be something wrong with the theory. If
there is no way to test the explanation, it's a just-so story. It it's
testable but no one has tested it, then it remains a just-so story until it
has been submitted to empirical test.

Although labeling an explanation as a "just-so story" has become fashionable
recently as a way to mindlessly dismiss it as worhless, there's nothing
wrong with the latter sort of just-so story, unless it is asserted to be
true without empirical backing. Here's another example of a just-so story
of the empirically testable variety: HPCT.

Bruce A.

[From Bjoern Simonsen(2000.12.06.1325 GMT+1)]

from Bill Powers (2000.12.04.0433 MST)

I'm beginning to get that impression, too, since everything I propose that
might distinguish reinforcement theory from reorganization (or any other)
theory seems to have something wrong with it. Of course that may be the
truth, but it's also possible, as Bruce Abbott tells us, that behaviorists
believe that reinforcement is simply the name for an observable aspect of
behavior, so they don't see any need for "testing" it. Do you "test" your
idea that the sky looks blue?.

I have followed the discussion "Grudging Acknowledgement". Here my basis
will be
"as Bruce Abbott tells us, that behaviorists
believe that reinforcement is simply the name for an observable aspect of
behavior, so they don't see any need for "testing" it."

As I have written earlier my first step to CSG was the essay "Behavior,
Purpose and Teleology" written by Rosenblueth, Norbert Wiener and Bigelow.
In this essay they defined the behavioristic study of natural events.

My purpose writing this letter is to show my argumentation why we cant put
reinforcement theory and reorganization up against each other.

R.,W.and B wrote:
" Given any object, relatively abstracted from its surroundings for study,
the behavioristic approach consists in the examination of the output of the
object and the relations of this output to input. By output is meant any
change produced in the surroundings by the object. By input, conversely, is
meant any event external to the object that modifies this object in any
manner"

I'll put this in concrete terms saying that f(x) = ax1 + bx2 + ......
Where x1, x2, ..... are the external events (variables) and f(x) is the
object.
The behavioristic way to describe "the world"/reality is to tell how the
objects vary, varying
the external events.

An example in this site is:

Blowing smokes

[From Bruce Abbott (2000.11.28.2200 EST)]

As a possible counterweight to all my recent nonsense about reinforcement,
here's something I encountered in a textbook I'm using this semester in my
Learning class:

. . . DeGrandpre, Bickel, Hughes, and Higgens (1992) reviewed 17 studies
that manipulated the nicotine content of cigarettes, and they plotted
economic demand functions to show the effects of changing nicotine levels.
Their analysis showed that if smokers are given reduced-nicotine cigarettes
in place of their usual cigarettes, they smoke more cigarettes per day, and
thereby expose themselves to more cigarette smoke (and greater health
risks). Conversely, if smokers are given cigarettes with higher nicotine
level levels, they smoke fewer cigarettes per day.
[Mazur, 1998, pp. 228-229]

Here the group of the 17 students is the "Given any object, relatively
abstracted from its surroundings for study". De Grandpre et all. varied the
input "the external events that modified the object in any manner".

For the Behaviorists the main goal is the "examination of the output of the
object and the relation of this output to input" They have missing
information if they try to tell why the output happens. They can only tell
us the facts about input and output.
Therefore I think Bruce Nevin is correct when he writes

[From Bruce Nevin (2000.11.30 11:02 EDT)]

Many S-R studies cannot be analyzed in terms of control. Primarily, this is

because data on

individual performance is not available. Only statistical results are

reported.

R.,W.and B continues:
"The above statement of what is meant by the behavioristic method of study
omits the specific structure and the intrinsic organization of the object.
This omission is fundamental because on it is based the distinction between
the behavioristic and the alternative functional method of study. In
functional analysis, as opposed to a behavioristic approach the main goal is
the intrinsic organization of the entity studied, its structure and its
properties; the relations between the object and the surroundings are
relatively incidental."

I'll put this in concrete terms saying that f(y) = my1 + ny2 + ......
Where y1, y2, ..... are the description of internal structures and
properties and f(y) is the object.
The functional way to describe "the world"/reality is to tell how the
objects are structured and about their properties.

In PCT the "main goal is the intrinsic organization of the entity studied,
its structure and its properties; the relations between the object and the
surroundings are relatively incidental."

In PCT there may be disturbances. These are relations between the object and
the surroundings. But remember it is the perceptions of these disturbances
that work.

I'll go back to an earlier account [From Bill Powers(940917.0600 MDT)],

"The theory of control offers an explanation in terms of perceptual

signals, closed causal

loops, and mathematical properties of such systems. These entities, while

perfectly experience

able in the mind, are not experiences to be explained. We are saying IF

such an organization existed in the nervous system, THEN the experiences we
are trying to explain would follow. The theory proposes the existence of
entities in the world hidden from direct experience, perhaps not >all of
them hidden forever, but certainly hidden now."

I think we should cut out the competition between those who work with
reinforcement theory and ourselves who work with reorganization. It is two
different approaches which supplement each other.

Here I need some help (if there is help to get).
I have repeatedly tried to explain in a simple mathematical way that these
two approaches cannot be linked.
I assume that f(x) = ax1 + bx2 +...... and f(y) = my1 + ny2 +...... . Here
f(x) # f(y) and f(y) is also different from x1, x2, ..... Here are too many
unknown variables and I have too few equations. If I let a PCT-system be
disturbed by, for instance, ax1 I will have a problem to describe f(y),
because there is a reference in the picture.

I am inclined to say that people working with a behavioristic approach
cannot insist that people working with a functional approach are wrong and
vice versa. They are describing two non compatible worlds.

Let me digress from the issue. When people back in history tried to describe
the world they did it in two ways. Depending if their description was
behavioristic or functional, they described the world differently. Their
knowledge was gathered in different subjects. And basically some subjects
were deeply rooted in either the behavioristic or functional approach. I
think medicine, geology, chemistry etc. had a functional approach and
physics, psychology, sociology etc. had a behavioristic approach.
Later we got psychiatry, social medicine, geophysics etc. trying to mix the
two approaches. And I am not sure if they have succeeded. I have studied
geophysics myself and the mixing of physics and geology is successful.
However, I think a geophysicist is sometimes a physicist and other times a
geologist.

Back to the top.
I don't think the behaviorists have any need for "testing" their output. How
should they do it?

Do you "test" your idea that the sky looks blue?.

No but when I see a blue sky (seldom where I live) I put on sun glasses. It
is not because of the input or because of the disturbance. There may be a
special reference.

Bj�rn

···

Subject: RE: Experience, Reality and HPCT. Here he supports this writing: