Perceiving vs. Controlling

CSGnet Archive CSG1992
1

[from Gary Cziko 920609.2100]

Martin Taylor (920609 12:35) said:

JGT's results that you can
learn to perceive only what you actively control would argue that the
learning is itself a feedback loop. You learn to perceive by doing, and
you learn to do by perceiving. It doesn't work one way OR the other. You
are buiding an ECS (I would guess) for the percept, and only by controlling
it incrementally better do you learn to perceive it.

But can't one learn to perceive all kinds of things one can't control? A
speaker of North American English can understand without much difficulty
all kinds of worldwide dialects and accents of English but couldn't come
close to sounding the same way. I can tell (usually) an Australian from an
Englishman, but there is no way I could imitate their "accents" so that
someone else could tell which one I was trying to produce.

So whil I can understand Bill Powers's statement that can't control if one
can't perceive, but I don't understand the claim for the converse. What am
I misunderstanding about JGT and this claim?--Gary

···

------------------------------------------------------------------
Gary A. Cziko Telephone: (217) 333-4382
Educational Psychology FAX: (217) 244-0538
University of Illinois E-mail: g-cziko@uiuc.edu
1310 S. Sixth Street Radio: N9MJZ
210 Education Building
Champaign, Illinois 61820-6990
USA
------------------------------------------------------------------

2

[Martin Taylor 920610 13:30]
(Gary Cziko 920609.2100, Rick Marken 920609.2030, Bruce Nevin 920610 0908)

Gary:

But can't one learn to perceive all kinds of things one can't control? A
speaker of North American English can understand without much difficulty
all kinds of worldwide dialects and accents of English but couldn't come
close to sounding the same way. I can tell (usually) an Australian from an
Englishman, but there is no way I could imitate their "accents" so that
someone else could tell which one I was trying to produce.

So whil I can understand Bill Powers's statement that can't control if one
can't perceive, but I don't understand the claim for the converse. What am
I misunderstanding about JGT and this claim?--Gary

There's no problem here, except the oft repeated one of mistaking a reference
value for the existence of an ECS that uses that reference. The percepts
that allow you to distinguish accents or dialects are the same (but with
different values) as those that allow you to produce your own accent and
dialect. No new kinds of ECS need be developed ("kind" means having a specific
perceptual function transforming specific inputs into a value that can be
compared with a reference value).

···

=============
Rick:

But can't one learn to perceive all kinds of things one can't control?

It seems so to me too. In fact, if JGT were right, I'd be functionally
blind. I can't control my kids, my wife, my mother. How did JGT
come up with this stuff?

Maybe if he'd left that part out of his theory he might be a
household word today. Even Freud didn't say anything THAT silly.

Confused in California

But you can control for perceiving the presence of your wife (as opposed to
a generic woman). Again, we have the distinction between reference values
and ECS types. And I don't remember JGT EVER saying anything really silly
except on purpose (which he did quite often). You ARE functionally blind
for the kinds of things you haven't learned to control for. Can you, with
your eyes closed, identify the material, the location and the size of a
rectangular object 15 cm high placed on a table at arm's length in from of
you? I doubt it, and the reason is that the perception involves control for
something of a kind new to you. But JGT showed you can learn this perceptual
ability, if you learn control of it. Having someone tell you what is there is
no good. Reaching out and feeling while you are hearing the echoes from the
object is one way to learn (perhaps the only way). When you do that, you are
learning by developing a control system that would permit you to select
for perception of particular objects.

Bruce:

My take on this: In translating JGT's work to CT terms, MMT has used the
word "control" where it doesn't belong.

Here's a stumbling try to make the relevant distinction: JGT showed
that you can learn to perceive only what you can interact with, where
"interact" entails a feedback relationship through the environment, but
where "feedback relationship" does not always entail control.

There's a subtle thought. I think Bruce is right, in a way. Initially,
when you are learning to perceive something of a new kind, you cannot control.
What you are doing is learning to control. I think you are building a new
ECS, but equally it might be changing the function of an existing ECS that
had a different but related perceptual input function. I have to acknowledge
once again that JGT did not conceive things in terms of the same kind of
control hierarchy that we take for granted in this group. Sometimes I put
my own interpretation in PCT terms on what he actually said, without realizing
that I am doing it.

General comment:

Of what use is a percept that cannot be controlled?

The resources devoted to creating the percept must at best be excess baggage
when it comes to the survival or effectiveness of the perceiver, and at worst
the uncontrollable percept may be distracting and dangerous. At the same time,
it is equally clear that almost all perceptions are, at any moment, not being
controlled. This is the reason for alerting functions (such as the movement
detectors in the visual periphery), and is at the base of concepts such as
"situation awareness."

Gary and Rick mistake the perception of variations within a type for the
perception of new types. An ECS is concerned with the variation within a type
(be it sensation or principle), but it cannot handle sensory abstractions
that are of a different type. I take "learning to perceive" as "developing
an ECS that can work with variations on a type." That, within H-PCT, implies
that the ECS can be provided with reference values. Learning what can be
done with the new percept is something else again. So, as Bill says, learning
to control requires that there be a percept to be controlled; and I say,
learning to perceive requires learning to control that perception.

Maybe this doesn't clear up the question. But it may help.

Martin

3

[From Rick Marken (920610 12:30)]

Martin Taylor (920610 13:30) says:

      The percepts
that allow you to distinguish accents or dialects are the same (but with
different values) as those that allow you to produce your own accent and
dialect.

This says it very nicely. As you can see from my earlier post today, I
caught onto this after embarassing myself by suggesting that JGT might
be saying something silly. In fact, it seems that he is saying
something quite deep; philosophically, I suppose, JGT is an
empiricist -- he's saying that we LEARN to perceive (perceptual functions
are not built-in at birth). I think Bill P. made a similar suggestion
in BCP -- when he suggested that a perception like "straightness" could
be learned by creating a perceptual function that produces a signal
that is invariant with respect to certain outputs (such as translational
movements of the eye). I think this is what JGT had in mind. The
perceptual function is constructed to serve control but, once constructed,
is available as an indication of the degree to which that perception
is present in experience.

I don't know if JGT pointed this out, but Bill P. noted that this control
based approach to perceptual learning does not lead to "arbitrary"
perceptual functions. The process should converge on certain types of
perceptual functions. This convergence is a result of the fact that
control takes place in a "real world" -- the place that scientists are
trying to model-- which is both the the input to our perceptual functions
and the constraint on how we can influence perception. It is a property of the
real world that "straightness" is invariant with respect to translation.
So, if a control system needs to perceive a constant signal as the eye
is moved left to right, it will "have to" construct a perceptual function
that is sensitive only to excitation of receptors that fall on a straight
line (in the plane of eye movement). This means that so called "a priori"
perceptions (I think that was Kant?) might not have to be "programmed into"
the architecture of the perceptual functions of the nervous system. The
apparent consistency of certain perceptions (across people) -- such as
perceptions of straightness, distance, whatever -- may result from the
fact that these kinds of perceptual functions "must be" built in order to
be able to control perceptions within the constraints of the outside world.

Maybe I should read JGT after all. He's starting to seem a LOT less silly
than Freud.

Best regards

Rick

···

**************************************************************

Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
E-mail: marken@aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)

4

[Martin Taylor 920610 18:30]
(Rick Marken 920610 12:30)

philosophically, I suppose, JGT is an
empiricist -- he's saying that we LEARN to perceive (perceptual functions
are not built-in at birth.)

Yes, he was an extremist in this. In my view, he went overboard in that
direction. There's no reason I can see why evolution should not have built
in lots of perceptual functions, especially at low levels. Many of the same
functions must be useful to cats, birds, fish, and people. So why should
each individual have to learn them? But JG would have none of that. From the
retinal cones to the muscle fibres, everything was assembled by the individual.

I don't know if JGT pointed this out, but Bill P. noted that this control
based approach to perceptual learning does not lead to "arbitrary"
perceptual functions. The process should converge on certain types of
perceptual functions. This convergence is a result of the fact that
control takes place in a "real world" -- the place that scientists are
trying to model-- which is both the the input to our perceptual functions
and the constraint on how we can influence perception.

I don't know whether he pointed it out, but I assume it to be true. How could
it be tested, though? We are unlikely to have the means to propose "sensible"
types of perceptual function that do not, in fact, occur.

It is a property of the
real world that "straightness" is invariant with respect to translation.
So, if a control system needs to perceive a constant signal as the eye
is moved left to right, it will "have to" construct a perceptual function
that is sensitive only to excitation of receptors that fall on a straight
line (in the plane of eye movement).

Unfortunately, straightness is a property of the translation of objects in
a special abstract world, not of the projection of the movement of objects
across the retina. The special abstract world has no friction and no
gravity. In a real world, there are tests for the straightness of edges,
but except in special cases, straight things or motions in the world don't
project onto straight lines or even great circles on the retina. Bill was
finding this out in trying to cross-map the Little Man's kinaesthetic and
visual sense of the world. JG found that with distorting spectacles, the
perceptions of the straightness of straight things was regained when he acted
to control them, but not otherwise. The perception of straightness is
indeed a subtle and non-obvious problem.

Maybe I should read JGT after all. He's starting to seem a LOT less silly
than Freud.

If you try to read JGT, two warnings: the writing is very dense and dry, so
it is tempting to give up before reading very far, and it is not cast in the
HPCT framework, so you may be put off by all the references to Clark Hull
and other S-R theorists. It is their influence that JG transcended, but
they provided the background for his thought.

Martin