Other stuff

CSGnet Archive CSG1997
1

[From Bruce Abbott (970922.1725 EST)]

Bill Powers (970922.1316 MDT) --

Bruce Abbott (970922.1030 EST)

So the actor claims he is doing "the same thing" while
the observer claims that the actor is doing something different. The
observer is wrong.

Doesn't this depend on what they mean by "doing"?

Of course, and that is the point. What the observer means by "doing" is
what he can see the parts of the body doing in relation to the environment.

And in relation to the other parts of the actor's body.

What the actor means is what those activities accomplish that that actor is
perceiving.

I'm not sure I follow. What do you mean by "what those activities
accomplish that the actor is perceiving?" Is this what the actor intends,
or what the actor accomplishes?

If "doing" means what the actor is trying to accomplish, then the actor is
right and the observer is wrong: the actor is still "doing" the same thing.
If "doing" means what movements are being performed, then the actor is wrong
and the observer is right: the actor is "doing" something different. For
the observer, what the actor is "doing" (in the sense of trying to
accomplish) is an hypothesis to be tested, not an observation.

Yes, but it is not an hypothesis for the actor.

How does the actor know what she is doing? She cannot perceive her own
reference signal, right?

We can't really say who is right until we know what each means by the term
"doing." Perhaps they are both right.

No, they are not both right. There is no compromise between saying that the
sun moves around the earth and that the earth spins on its axis. In terms
of the grand scheme of science, one view is right and the other is wrong.

What the observer describes -- the actions of the actor -- are just as much
what the actor is "doing" as the actor's description in terms of goal. It
is purely a matter of how one defines "doing." The former type of "doing"
describes the actions, the latter the goal of those actions -- what the
actions accomplish (if successful). The former describes what the organism
is doing, the latter partly explains why. (The full account requires
knowing about the disturbances.) They are not competing accounts, as your
sun-centered and earth-centered accounts of the observations are. One is
description, the other, explanation.

You have just observed me touching the tip of my nose with the tip of the
index finger of my right hand, then withdrawing the hand. What am I doing?

Regards,

Bruce

2

[From Rick Marken (970923.1940)]

Bruce A.says:

We can't really say who is right until we know what each means by
the term "doing." Perhaps they are both right.

Bill P. replies:

No, they are not both right.

Bruce Abbott (970922.1725 EST), not willing to let conventional
psychology die with dignity, denies this, saying:

What the observer describes -- the actions of the actor -- are
just as much what the actor is "doing" as the actor's description
in terms of goal. It is purely a matter of how one defines
"doing."

Just as I thought [Rick Marken (970922.1430)]:wink:

By the way, for those who are into "real world" examples of control,
my prediction of Bruce A.'s behavior was possible because I know
many of the higher level perceptions Bruce A. is controlling for.
Agreeing that the observer's perspective is wrong and that the
actor's perspective is right would be as much of a disturbance
to one of Bruce's controlled perceptions (that the conventional
view of behavior is basically correct) as the idea that reinforcers
are controlled variables. And on that note I see that we now have
a response to _that_ disturbance:

Bruce Abbott (970922.2000 EST) --

I little thought about the matter [of a reinforcer being
a controlled perception] will show that this cannot be
correct. A controlled perception is a variable..

A good point. The controlled perception is really some function
of the pellets -- like their rate of occurrance. Since pellets
are called "reinforcers" then it is better to say that:

A reinforcer is an objective correlate of an aspect of a
controlled perception.

Mathematically, this is

p = f(re)

where p is a perceptual signal and re is some variable aspect
of the reinforcers (pellets), such as their rate, size, mass,
etc. and f() is the perceptual function that transforms the
variable aspects of the reinforcers into a perceptual signal.

But I still think that only someone desperate to salvage some
weird view of what is going on in operant conditioning
experiments could take much exception to my shorthand way
of saying the same thing: A reinforcer is a controlled perception.
My statement certainly captures the most important difference
between the PCT and EAB view of reinforcers: in PCT a reinforcer
is _controlled_; in EAB it _controls_.

Perhaps what you have in mind is that the reinforcer is some
particular state of the controlled variable

No. I think some physical aspect of the reinforcer(s) is an input
to the perceptual function that defines the perceptual variable
that is controlled.

a reinforcer must be that which reduces error between what I
have and what I want

This means that a reinforcer is _both_ an output and a disturbance
to a controlled perception since:

e = r - p and p = f (o+d)

so

e = r - f(o+d)

Assuming a fixed reference, r, then what reduces error, e, is
variations in o _and_ d. So a reinforcer is bar pressing rate
and non-contingent pellets.

This is a very new look for reinforcers, Bruce.I always
thought they were the pellets that arrived as a _result_ of
bar pressing;-)

Me:

But I think it's important to understand that a reinforcer is,
indeed, a controlled perception. Otherwise one might be tempted
to think that behavior modification approaches to dealing with
people actually have some merit.

Bruce:

I don't see the connection

You don't? Try meditating on this:

Behavior modifiers see reinforcers as _controllers_; we see them
as _controlled_. Behavior modifiers use reinforcers as controllers
of behavior. We think that's idiotic because reinforcers can't
control anything; if anything, they are controlled.

but at least I now understand why the good Dr. Marken -- who
apparently does see a connection between these two ideas --
wishes to deny the obvious.

What is the obvious that I am denying?

Best

Rick

···

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

3

[From Bruce Abbott (970923.0800 EST)]

Rick Marken (970923.1940) --

Bruce A.says:

We can't really say who is right until we know what each means by
the term "doing." Perhaps they are both right.

Bill P. replies:

No, they are not both right.

Bruce Abbott (970922.1725 EST), not willing to let conventional
psychology die with dignity, denies this, saying:

What the observer describes -- the actions of the actor -- are
just as much what the actor is "doing" as the actor's description
in terms of goal. It is purely a matter of how one defines
"doing."

This has nothing to do with conventional psychology vs PCT; it is just a
statement of fact. One cannot know who is right until one knows what each
means by the term "doing."

Just as I thought [Rick Marken (970922.1430)]:wink:

By the way, for those who are into "real world" examples of control,
my prediction of Bruce A.'s behavior was possible because I know
many of the higher level perceptions Bruce A. is controlling for.

What is your argument, Richard? Is it that one shouldn't accept my position
because I am "defending conventional psychology"? My reasons for advancing
this position (whatever they may be) have nothing to do with the truth or
falsity of my claim, which should be evaluated on its own merits. I don't
see such an evaluation here.

As for Bill Powers' distinction between the two views of "doing," I
understand and agree with his thesis that there is a right way and a wrong
way to _interpret_ the observed movements. The right way is
organism-centered, not observer-centered. But that is not what Bill _said_
(although it is surely what he intended to convey). He said that the
observer is wrong _merely for describing the agent's actions_. There is
nothing inherently wrong with describing the observed actions, as seen by
the observer; these descriptions are mere data. The problem is not with the
observations, but with their _interpretation_ -- "seeing" the sun go around
the earth, when in fact the earth is rotating.

A little thought about the matter [of a reinforcer being
a controlled perception] will show that this cannot be
correct. A controlled perception is a variable..

A good point. The controlled perception is really some function
of the pellets -- like their rate of occurrance. Since pellets
are called "reinforcers" then it is better to say that:

A reinforcer is an objective correlate of an aspect of a
controlled perception.

Mathematically, this is

p = f(re)

where p is a perceptual signal and re is some variable aspect
of the reinforcers (pellets), such as their rate, size, mass,
etc. and f() is the perceptual function that transforms the
variable aspects of the reinforcers into a perceptual signal.

A reinforcer (e.g., a food pellet) does not have a "variable aspect." It
may be delivered at a variable rate, but it does not "have" a variable rate
of delivery. It "aspects" are constants: constant size, constant weight,
constant composition, constant color. Constants are not variables.

But I still think that only someone desperate to salvage some
weird view of what is going on in operant conditioning
experiments could take much exception to my shorthand way
of saying the same thing: A reinforcer is a controlled perception.

Your "shorthand" hides a fatal flaw in your argument: reinforcers are not
variables. That is why I take exception to it.

My statement certainly captures the most important difference
between the PCT and EAB view of reinforcers: in PCT a reinforcer
is _controlled_; in EAB it _controls_.

Wrong. I have noted this mistake before. The term "control" in EAB does
not have the same meaning as the term "control" in PCT; the assertion above
misleadingly implies (by failing to note the distinction) that "control" has
the same meaning in both.

In PCT, to "control" means to bring a variable to a particular state and
keep it there by opposing the effects of disturbances on the variable. In
EAB, to "control" means to exert an influence. Interestingly, in
engineering the term "control" is often used in the EAB sense: a
"controller" is a device that, under appropriate conditions, influences the
state of some variable in a particular way. For example, a simple motor
controller supplies a variable power to the motor, thus influencing the
speed, torque, etc. of the motor (but only when power is being supplied to
the controller). In EAB, the reinforcer, through its contingent delivery
via some specified class of behavior, influences what behavior will be
observed, and its probability over time (but only when any required
establishing operations have been performed). This is far different from
claiming that a reinforcer brings a variable to a given state and keeps it
there by opposing the effects of disturbances on the variable. The latter
claim is obviously wrong; I suppose that it why it is so often attributed to
EAB in this forum.

Perhaps what you have in mind is that the reinforcer is some
particular state of the controlled variable

No. I think some physical aspect of the reinforcer(s) is an input
to the perceptual function that defines the perceptual variable
that is controlled.

So a reinforcer is not, in your view, something a person or animal wants?

a reinforcer must be that which reduces error between what I
have and what I want

This means that a reinforcer is _both_ an output and a disturbance
to a controlled perception since:

e = r - p and p = f (o+d)

so

e = r - f(o+d)

Assuming a fixed reference, r, then what reduces error, e, is
variations in o _and_ d. So a reinforcer is bar pressing rate
and non-contingent pellets.

This is a very new look for reinforcers, Bruce.I always
thought they were the pellets that arrived as a _result_ of
bar pressing;-)

It is a reinforcer because it has a certain effect on a controlled variable.
It reinforces a particular behavior only when it is delivered contingent on
that behavior. In the latter case, it is a disturbance to the CV that is
produced by the control system's own actions. Bar pressing is not a
reinforcer, as can be shown by eliminating the contingent delivery of the
reinforcer: bar pressing then collapses.

Regards,

Bruce

4

[From Bruce gregory (970923.1000 EDT)]

Bruce Abbott (970922.1725 EST)

You have just observed me touching the tip of my nose with the tip of the
index finger of my right hand, then withdrawing the hand. What am I doing?

Trying to make a rude gesture?

Sancho Panza

5

[From Bruce Gregory (970923.1025 EDT)]

Bruce Abbott (970923.0800 EST)]

It is a reinforcer because it has a certain effect on a controlled variable.
It reinforces a particular behavior only when it is delivered contingent on
that behavior. In the latter case, it is a disturbance to the CV that is
produced by the control system's own actions. Bar pressing is not a
reinforcer, as can be shown by eliminating the contingent delivery of the
reinforcer: bar pressing then collapses.

I assume "it" refers to the pellet. I liked it better when you
said that the reinforcer is part of the control loop. Now you
are sounding like an EABer. The pellet "reinforces a particular
behavior only when it is delivered contingent on that behavior."
What you say is true of the heat generated by the furnace and
its effect on the thermostat. However, removing the heat does
not cause the thermostat to behave differently because of its
limited reorganization skills.

I don't know why you object to bar pressing being a reinforcer.
If the rat is not hungry the pellet is no longer a reinforcer
either. Why do we single out one component of the control
loop for sanctification as a reinforcer?

Sancho Panza

6

[From Bill Powers (970923.0900 MDT)]

Bruce Abbott (970922.1725 EST)--

What the actor means is what those activities accomplish that that actor is
perceiving.

I'm not sure I follow. What do you mean by "what those activities
accomplish that the actor is perceiving?" Is this what the actor intends,
or what the actor accomplishes?

I intended the contrast to be between what the actor is perceiving and what
some outside observer might be perceiving to be "the" consequence of the
actions. The actor is attending to some kind of perception, and has a
reference level for that perception. The observer might be attending to
some other perception, and interpret its final state as the significant
result, incorrectly.

If "doing" means what the actor is trying to accomplish, then the actor is

right and the observer is wrong: the actor is still "doing" the same thing.
If "doing" means what movements are being performed, then the actor is

wrong

and the observer is right: the actor is "doing" something different.

We can't really say who is right until we know what each means by the term
"doing." Perhaps they are both right.

I think you miss the actual difference in the meanings of "doing". The
difference is not between the movements and the goal of the movements; it's
a matter of proposing a cause of the observed movements. When we say the
organism "does" something, we mean that the doing is the result of some
output from the organism, rather than an effect of some other agency. If I
seize your wrist and force your hand upward, the motion of your hand is
what _I_ do, not what _you_ do.

As long as you leave out disturbances, the appearance is that all movements
are the organism's doing -- that is, effects of nervous system outputs. But
a simple physical analysis shows us that all observable results of muscle
tension are a joint effect of muscle tensions and other forces and
constraints acting on the limbs. When a rat reaches a paw out in
preparation for pressing a bar, all the forces being generated by its
muscles are aimed upward, holding the paw up against gravity. When the rat
presses downward, the downward force is due in part to the rat's muscle
forces, and in part to the weight of its foreleg. And when the lever moves
down, its position is the resultant of the net downward force produced by
muscles and the weight of the foreleg, minus the restoring force of the
spring under the lever, plus the unbalanced part of the weight of the
lever. Even the fact that the lever moves downward is misleading; it looks
as though the rat's output force is a vector aimed tangentially to the arc
along which the point of contact with the lever moves, but in fact it can
be aimed in any direction as long as there is a downward component. The arc
of movement is created by the physical constraint on the lever's movement,
and does not indicate the direction of the rat's output force. Even the
point where the downward movement ends is dictated by the stops under the
lever, not by the point where the rat would otherwise stop moving its paw
downward.

The problem with your "descriptive" approach to doing is that it is
entirely too superficial. While claiming to be simply a description of
observations, it is such a cursory description that it leaves out most of
what is actually happening. It attributes major aspects of the observed
"movements" to the organism, when the organism is not reponsible for them
at all. This is especially true when variable disturbances are present, and
ignored. The outcome, the observed movement, can remain exactly the same
even though the actual outputs of the organism are varying through their
entire possible range, from maximum positive to maximum negative. What then
is it that the organism is "doing"? Is the doing the variation of output
forces over the entire range, or is it the resultant, which is remaining
regular and repeatable? If the latter, then we know immediately that we are
observing a controlled result, not an action produced by the organism.

What you are saying here is that a vague and informal usage of the term
"doing" based on superficial observation is just as good an alternative as
a precise technical usage based on detailed observation. But these are not
just alternative ways of describing the same observations. If you see a
woman holding her arms out to the sides and waving them rapidly up and down
in varying patterns, is this what she is "doing?" Or is she maintaining her
balance on the high-wire? If you see a man holding a rifle and gradually
moving it from side to side, stopping in a particular position, is this
what the man is doing, or is he centering a target in the sights? These are
by no means equivalent descriptions of the same thing.

You have just observed me touching the tip of my nose with the tip of the
index finger of my right hand, then withdrawing the hand. What am I doing?

You have described a consequence of action, not an action. The action is
quite complex. First the arm must be supported against gravity. Then there
must be angular accelerations at the shoulder and elbow joints that start
moving the hand from its initial position (any initial position) along a
line from the former finger position toward the nose. The supporting forces
must vary as the radius from shoulder to the center of mass of the arm
changes. As the fingertip approaches the nose, the whole moving assembly
must decelerate, and as the fingertip and nose start to compress from the
contact, the motion must stop. Then the process must reverse to withdraw
the hand.

The _result_ of all these variable actions is "touching the tip of your
nose." What you are "doing" in the sense of action is not touching the tip
of your nose, but varying the positions of the segments of your arm.
Touching the tip of your nose is the perceived outcome of these actions.
And if, while you are in process of carrying out these actions, I were to
reach out and apply forces to your arm, your own forces would vary exactly
as required, and the fingertip would get to the nose anyway. The only
aspect of your behavior that is then entirely your doing is the controlled
perception of fingertip on nose. If you were doing this blindfolded, I
could reach out and touch your nose and fingertip at the same time, and
your motions would stop. (I just tried this with Mary, and it works
beautifully. "I didn't think my nose was out _there_!" she said afterward.).

Best,

Bill P.

7

[Hans Blom, 970923e]

(Rick Marken (970923.1940))

By the way, for those who are into "real world" examples of control,
my prediction of Bruce A.'s behavior was possible because I know
many of the higher level perceptions Bruce A. is controlling for.

You finally admit it: you're a model-based controller after all ;-).

Bruce:

a reinforcer must be that which reduces error between what I
have and what I want

Rick:

This means that a reinforcer is _both_ an output and a disturbance
to a controlled perception ...

It has indeed been observed that a disturbance (something incidental
and not causal) can be a reinforcer. Some chickens (or was it doves
or whatever) develop the strangest habits while being conditioned.
One chicken would, in one instance that I remember vaguely, far more
frequently than random stand on its right leg while turning its head
to the lower left over almost 180 degrees. Human observers interpret
this as the chicken having learned to perform these funny postures
"in order to" get the food, even though its delivery is random.

Initially, it is impossible to discriminate between consistent
regularities and irregular disturbances. I've noticed the same thing
in an early adaptive blood pressure controller: when it started the
infusion and the blood pressure _rose_, it "formed the opinion" that
the drug infused was one that _raised_ blood pressure. We, objective
bystanders with a vastly more knowledge and equally vastly better
perceptual apparatus, of course knew that the drug was a blood
pressure depressant, but that incidentally at the time the infusion
was started the surgeon made his first incision while anesthesia and
pain suppression were still insufficient.

What did we learn from this? Well, to provide the computer with some
"common sense", i.e. knowledge that it would not have and could not
possibly have acquired by the time control needed to be good. In this
case we -- irresistably -- told it: "whatever your own observations
may tell you, the loop gain that you will employ for control must be
negative and at least so and so much". A completely overpowering form
of teaching, reminiscent of some forms of social control...

It is only through learning over a longer time (far more examples
than just one or a few) and in a constant context that disturbances
cancel out and only the average effect remains. It is this average
that we then accept as "the" effect of the drug.

From long ago I remember the CAT (Computer of Average Transients)

that was used in animal experiments. It worked on a similar principle
in establishing the response (in the EEG) of a proffered stimulus (a
click in the ear or an electric current pulse to the skin). A single
response is not visible in the EEG at all: its amplitude is some 100
times lower than the background EEG. Yet by synchronously averaging
over hundreds or thousands of trials could a well-defined systematic
effect be demonstrated.

Could learning sometimes be really as simple as averaging out the
noise?

Greetings,

Hans

8

[From Rick Marken (970923.1240)]

Me:

By the way, for those who are into "real world" examples of control,
my prediction of Bruce A.'s behavior was possible because I know
many of the higher level perceptions Bruce A. is controlling for.

Hans Blom (970923e) --

You finally admit it: you're a model-based controller after all ;-).

Science is, indeed, an actual example of model based control.
The controller (scientist) invents (imagines) a model (f=ma)
whose behavior will produce certain observable results (like
eliptical planetary orbits). Some of these results fit
observations (like the eliptical orbits) that have already
been made; other results have not yet been observed; so we
make an observation (do an experiment ) and see if the
actual results match those predicted by the model. If they
do, we stick with the model and keep testing; if they don't
we try to revise the model so that it accounts for all existing
observations. So a scientists acts pretty much like you
Kalman filtering model based control system, refining the
model by comparing actual to expected effects of actions
(experimental manipulations) on the world.

The model based control process that is science is something that
can be done (as far as I know) by only one kind of organism --
humans. And, indeed, it seems that it can be done by only
a subset of humans at that. Therefore, it seems queer to propose
this very unique method of controlling as a general model of
the controlling (purposeful behavior) done by living systems.
Even more queer is the fact that this scientific approach to
controlling is being advocated by a person who has no interest
in using it himself. (If you were actually doing model based
control, Hans, you would revise your model based on evidence;
since you don't do this, I have to conclude that your own model
based control model does not apply to you; open loop control
seems more likely;-))

Best

Rick

···

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

9

[From Bruce Abbott (970923.1850 EST)]

Bruce Gregory (970923.1025 EDT) --

Bruce Abbott (970923.0800 EST)

It is a reinforcer because it has a certain effect on a controlled variable.
It reinforces a particular behavior only when it is delivered contingent on
that behavior. In the latter case, it is a disturbance to the CV that is
produced by the control system's own actions. Bar pressing is not a
reinforcer, as can be shown by eliminating the contingent delivery of the
reinforcer: bar pressing then collapses.

I assume "it" refers to the pellet. I liked it better when you
said that the reinforcer is part of the control loop.

I don't intend to change the definition. However, it is convenient to
continue to call the pellet "the reinforcer" even when not currently part of
a control loop, for the same reason that it is convenient to call a slice of
bread "food" even though it is not currently being digested in someone's
stomach. The relevant property of the pellet is its ability to affect the
CV in a particular way; any object with this property will serve as a
reinforcer when made contingent on the actions of the system.

I don't know why you object to bar pressing being a reinforcer.
If the rat is not hungry the pellet is no longer a reinforcer
either. Why do we single out one component of the control
loop for sanctification as a reinforcer?

Bar pressing per se does not reduce the error in the relevant control system
when this error exists. Pellet delivery does. When the rat is not hungry,
there is no error for the pellet to diminish, and then, like the bar press,
the pellet is not a reinforcer.

Regards,

Bruce