Acting Up

[Abbott_Bruce_Indiana]

[From Bruce Abbott (941214.1730 EST)]

(Note: I sent this yesterday and it was bounced back to me with a "HOST
UNKNOWN" message. However, the address appears to be correct as listed in the
header. Ah, well...)

Tom Bourbon [941214.1236]

Sequences of movements under feedback control? Or sequencs of peceptions
under feedback control? The difference is significant. Your next sentence
gives the impression you mean that the "acts" are under feedback control.
If so, there is still a problem.

This response ignores the point I was trying to get across. Although I admit
to being dense from time to time, I do understand the meaning of "perceptual
control"--that specific bits of behavior vary as needed in order to produce
desired (i.e., reference-state) perceptions. But contrast the following two
descriptions:

(a) The cat learned to pull the string.

(b) The cat learned to vary the reference levels of a number of perceptual
     control systems so as to perceive its paw grasping the string and
     pulling on it.

Alternative (b) may be more accurate (although it still finesses plenty of
detail) but is exceedingly clumsy in ordinary discourse. So, I write (a), and
ask you to allow that by (a) I intend to be understood as implying (b). I
suggested that we might have an easier time of it if we would agree on a
general name by which we can refer to these intended perceptions (references)
which must be achieved by means of variable behaviors. They might be called
"acts" or "operants" or some other name that will distinguish them from the
varying behavioral outputs by which perceptions are brought to their reference
levels.

My suggestion has the advantage of being consistent with the common-sense
meaning of such terms. If you are observed to be struggling with the twist-
cap on a bottle of your favorite Thunderbird wine and someone asks "what are
you doing?", you will probably reply that your are trying to twist the #$$%&##
cap off the bottle. It is what you are TRYING TO DO (goal and necessary
subgoals) that defines the act, not the specific (variable) behaviors through
which the goal is achieved.

If in the past you have succeeded in getting the cap off similar bottles by
going through a specific series of acts (grasping the bottle firmly around the
middle with one hand, grasping the cap firmly in the other, applying strong
counterclockwise torque to the cap), you will probably repeat those acts
(ACTS, not BEHAVIORS) here, although the specific behaviors you use to
accomplish those acts may vary.

Bill Powers (941214.1030 MST)

Once again I'm piggy-backing off someone else's citation of a post that
hasn't got here yet ( 16 hours after its time-stamp), in this case Gary
Cziko citing Bruce Abbott:

Too bad, because Gary's post left out everything I had to say by way of
clarification. I'm not arguing about how these things are accomplished, but
about how to describe these things in a natural way. The basic problem for me
is that I find it awkward to keep having to explicitly describe the whole
damned closed loop every time around. If different behaviors bring about the
same perceptual state then at some level of abstraction they are all the same
act, e.g., accepting a bid. If reorganization restructures the organism so
that raising the reference level for some perceptual variable now produces a
specific sequence of changes in other reference levels so as to carry out a
specific behavioral act (e.g., pulling on one's ear), then it is the act of
pulling on the ear that gets executed, regardless of disturbances so long as
they do not exceed the capacity of the system to compensate. And if you ask
me what I'm doing, I'm pulling my ear and, as a result, accepting a bid. In
saying this, I'm certainly not stating that specific behaviors (i.e., muscle
contractions) have been selected in the reorganization process. Honest. (:->

Bill, a reminder: you were going to have something to say regarding my post on
RC flying. Also, I've had no reply to my question about saccharine-motivated
learning and reorganization...

Regards,

Bruce

[Tom_BOURBON3]

Tom Bourbon [941215.1149]

[From Bruce Abbott (941214.1730 EST)]

Tom Bourbon [941214.1236]

Sequences of movements under feedback control? Or sequencs of peceptions
under feedback control? . . .

This response ignores the point I was trying to get across. Although I admit
to being dense from time to time, I do understand the meaning of "perceptual
control"--that specific bits of behavior vary as needed in order to produce
desired (i.e., reference-state) perceptions. . . ..

That was my understanding of your understanding. I'm just trying to get a
clearer idea about what you mean by "bits of behavior," "behaviors," and
"acts." I didn't ignore your point, at least I didn't try to. After the
passage you quoted above I went on to discuss the following statement from
you:

Successful acts would remain "selected" in the
behavioral output function, because they would halt reorganization.

In reply to that idea, I said:

"I think there would be no new "acts" (or
old ones, for that matter) "in" the output function. I think reorganization
would halt (a) when, in a previously existing system, new reference signals,
or new settings of parameters, resulted in control of the intended
perception; or (b) when new control loops, perhaps at a new level, resulted
in control of the intended perception." And so on.

For good reason, you often use the language of EAB. I'm simply trying to
better understand what you mean when you speak EAB in discussions about
perceptual control.
. . .

But contrast the following two
descriptions:

(a) The cat learned to pull the string.

(b) The cat learned to vary the reference levels of a number of perceptual
    control systems so as to perceive its paw grasping the string and
    pulling on it.

Alternative (b) may be more accurate (although it still finesses plenty of
detail) but is exceedingly clumsy in ordinary discourse. So, I write (a), and
ask you to allow that by (a) I intend to be understood as implying (b). I
suggested that we might have an easier time of it if we would agree on a
general name by which we can refer to these intended perceptions (references)
which must be achieved by means of variable behaviors. They might be called
"acts" or "operants" or some other name that will distinguish them from the
varying behavioral outputs by which perceptions are brought to their reference
levels.

I understand your point. When we write about our modeling, some of us draw
the distinction between "action" and "behavior," the former referring to the
moving and twitching of body parts, the latter to those environmental
events that are analogues of the controlled perception. To use your
example as an example:

. . . If you are observed to be struggling with the twist-
cap on a bottle of your favorite Thunderbird wine

ACTIONS

and someone asks "what are
you doing?", you will probably reply that your are trying to twist the #$$%&##
cap off the bottle. It is what you are TRYING TO DO (goal and necessary
subgoals) . . .

BEHAVIOR [but I doubt that in this example the highest(?) level goal is to
remove the cap, then stop ]

not the specific (variable) behaviors through
which the goal is achieved.

ACTIONS

It is often (nearly always?) the case that if we ask a person, "what are
you doing?," the answer will be something other than the actions we see.
They are doing something we can't see.

Any words we use to describe this distinction will carry a lot of baggage
from previous incarnations in behavioral science. "Operant" certainly
carries its share.

Later,

Tom

[Peter_Burke]

[From Peter Burke 941216.1000)]

[From Bruce Abbott (941214.1730 EST)]
If in the past you have succeeded in getting the cap off similar bottles by
going through a specific series of acts (grasping the bottle firmly

around the

middle with one hand, grasping the cap firmly in the other, applying strong
going through a specific series of acts (grasping the bottle firmly

around the

middle with one hand, grasping the cap firmly in the other, applying strong
counterclockwise torque to the cap), you will probably repeat those acts
(ACTS, not BEHAVIORS) here, although the specific behaviors you use to
accomplish those acts may vary.

All of this discussion which is focussed on keeping the inputs
separate from the outputs, and not confusing the two in any
discussion, fails to conisder the gloss that is made when one
says that the outputs do not matter as long as the inputs are
controlled. Of course the outputs matter, because some of them
do not control the inputs, and some push the inputs further
away from the reference value. The question is, how are those
outputs selected. Its one thing to say they are selected
because they control the inputs, but it is another thing
entirely to show exactly how the error signal accomplishes this
proper selection of behavior. It may not be reinforcement that
selects the behavior, but the behavior is nevertheless selected
by some mechanism. Let's see some disucssion of this issue!

Peter

[_Martin_Taylor]

[Martin Taylor 941216 13:50]

Peter Burke (941216.1000)

Of course the outputs matter, because some of them
do not control the inputs, and some push the inputs further
away from the reference value. The question is, how are those
outputs selected. Its one thing to say they are selected
because they control the inputs, but it is another thing
entirely to show exactly how the error signal accomplishes this
proper selection of behavior. It may not be reinforcement that
selects the behavior, but the behavior is nevertheless selected
by some mechanism. Let's see some disucssion of this issue!

The error signal accomplishes nothing except to modulate the amount of
output from the elementary control unit whose error signal it is. The
"selection of behaviour" is determined by the connections that output
signal has to lower level ECUs, where it contributes to their reference
signals. Each of these lower-level ECUs has some kind of perceptual
input function, totally unknown to the higher one, and irrelevant to its
operation, but each has sensory/perceptual input from many different sources.

The high-level ECU has only ONE scalar perceptual signal to control. That
signal is a function of MANY inputs, and therefore there are a large number
of different perceptual contexts that can lead to any given value of the
higher-level perceptual signal. When there is a higher-level error that
is, say, positive, it induces an output signal that may contribute positively
to some of the lower-level references, negatively to others. But depending
on the lower perceptual contexts, the resulting error in any particular
lower-level ECU may be negative sometimes and positive at other times, for
the SAME reference signal. These lower-level errors affect the outputs
from their own ECUs, so depending on the perceptual context, the higher-level
error signal causes sometimes one set of behaviours and sometimes another
at the lower levels. But no matter what, the result is still that the
higher-level loop gain is negative, so long as control is maintained (by
definition).

How can the system be constructed so that the gain remains negative even
when the resulting actions may vary from time to time? That's not
reinforcement, that's reorganization. And it has been much discussed
here and in the various writings of Powers and others often cited here.

Martin