[from Wayne Hershberger, 921208]
Re: catching up on a month's E-mail.
First things first:
Bill and Greg, I second Oded's recommendation that you
consider _Nature_, if Mary's ploy doesn't get _Science_ to wake up
and smell the coffee.
Gary, congratulations on your two _Educational Researcher_
articles. You have positioned yourself perfectly to augment (and
take advantage of) the paradigm shift we all are anticipating!
Now, to address a couple threads:
Controlling Error:
(Audra Wenzlow 92.12.02) The notion of "controlling
perceptions" has always disturbed me precisely because it
does not include the reference signal in its description....
Without a reference setting, controlling a perception would
have no meaning....For instance, in the rubber band
experiment, I don't really care where the knot is, only how
far I perceive it to be from the reference point. In other
words, I am not controlling my perception, but the
difference between my perception and my reference level.
The fact that controlling occurs relative to some reference
setting is precisely what seems to be disregarded when we
say that we "control our perceptions."
You are right, Audra, to be concerned about the reference signal!
However, the fact that controlling occurs relative to some
reference value (of a PARTICULAR perceptual dimension or scalar)
is precisely what tends to be disregarded if we claim that we
merely "control our error." Imagine yourself at the chalkboard
engaged in the rubber band experiment, and all you can see is a
meter registering how much error is present: the absolute, not the
algebraic amount--an algebraic amount presupposes a perceptual
dimension (e.g., right/left) and that is what you are supposing is
irrelevant. What will happen? A lot of blind variation. The
knot will move around like an E-Coli bacterium. Controlling error
without controlling particular perceptual variables amounts to
poor control of error!
The perceptual dimensions (or scalars) comprising the spatial
coordinates of our psychophysical world is what allows us to
control the position of a manipulandum (e.g., the knot) with a
minimum of false starts--the error in the control loop is kept low
precisely because the reference value is specified in such
perceptual terms as "intending to keep the knot on the dot." An
error signal's perceptual origin is what gives it the requisite
dimensionality for efficient control (feedback must be negative);
for example, as compared to the reference value, the controlled
perceptual variable may be too far right, too far left, too high,
too low, too bright, too dim, too loud, too soft, too near, too
far, too red, too white, etc. etc. In contrast, an error which is
simply "too much," is virtually a joke; such control is possible
but it is a very limited (special) case.
Because reorganization actually requires blind variation, it is
reasonable to assume that reorganization involves the control of
intrinsic error (the Hullian behaviorists called it drive
reduction and trial and error learning; Freud spoke of libido and
the pleasure principle), but that is a very special case (which
Mark Olson and Bruce Nevin seem to recognize).
Perception: Realization vs. representation
(Rick Marken (921020) What is your model of perception,
Wayne? Here is my model: EV --->S--->PF--->PS, where EV is
an environmental variable, S is a sensor, PF is a perceptual
function and PS is a perceptual signal....In your example of
"diagonal movement", PS IS the perception of diagonal
movement -- constructed from the sensory inputs that are
ultimately caused by the horizontal and vertical EV
movements. What else is needed here -- other than the
delineation of how PS results from EV
Rick, you are missing the point! The problem with this schematic
(I hesitate to call it a model or to say it's yours; I recognize
it as Thomas Reid's idea) is not so much what it lacks as what it
presupposes, namely that there is some realization process (EV)
that is distinct from and CAUSALLY a priori to the process of
perception. In short, you've got the PF too localized. What you
are calling PF must permeate the whole process--so you beg the
question by characterizing EV as horizontal and vertical
movements. Let me repeat myself, particularly the part in capital
letters concerning EV:
(Wayne Hershberger 921018) The perceptual process involves
an ecological dipole. Perception is NOT simply a process of
transporting a representation of some putative conceptual
reality comprising one end of the dipole (the environmental
pole) into the other end (the organism pole). Such a
conceptualization (representationalism) begs the fundamental
question of perception, which is the realization of the
perceptual world in the first place. For example, consider
two light emitting diodes (LEDs) moving in phase in the
dark, one vertically, the other horizontally, like this
(i.e., conceive it thus):
^
>
>
>
v <-------->
Doing this, Gunnar Johansson found that one perceives
diagonal motions. The two lights are seen as separating
diagonally as the PAIR moves along the orthogonal diagonal,
from lower left to upper right; and then as the PAIR moves
diagonally back to the lower left the two lights are seen to
approach each other, diagonally. Describing the LED's
motion as vertical and horizontal is a conceptual
convenience. And it is realistic. But this conceptual
reality doesn't account for the diagonal motions that are
perceived. And for the same reason, it doesn't account for
the perception of vertical and horizontal motion (e.g., when
the room lights are on), either. Why should it?
The phenomenon described above illustrates the
difference. One can conceptually represent (i.e., re-
present) a perceptual reality as I did above in describing
the LED's motion as vertical and horizontal--that is, I have
described the motion (i.e., conceptually realized it) as it
looks (i.e., is perceptually realized) in the light. But
this equivalence (between the perceptual and conceptual
realizations) is an accident of the room lights being on;
when the room lights happen to be off, the equivalence
vanishes. However, this is not to say that the perceptual
realizations (vertical/horizontal motion or diagonal motion)
are themselves an accident of the room lights being either
on or off. THE PERCEPTUAL REALIZATION DEPENDS LAWFULLY UPON
THE STATUS OF THE ROOM LIGHTS--SAME BRAIN, DIFFERENT INPUT.
THE INPUT, OF COURSE, IS NEITHER "DIAGONAL MOTION" NOR
"VERTICAL/HORIZONTAL MOTION," BUT RATHER SOMETHING FROM
WHICH THESE ALTERNATE REALIZATIONS MAY BE ACHIEVED.
PERCEPTUAL REALIZATIONS ARE NOT CAUSED, THEY ARE ACHIEVED.
It seems to me that Bill captured this perspective perfectly in a
recent post--providing one recognizes that the environment is just
as much an analog "computer" as the nervous system:
(Bill Powers (921102) When you put signals into an analog
computing network, you get signal variations everywhere in
the network, not just at the nominal outputs. All of these
variations have meaning, because they stand in continuous
relationship to the inputs, according to rules that are the
properties of the intervening devices. There isn't a symbol
anywhere in the network: just voltages and currents behaving
through time. Of course the user of the computer assigns
symbolic meanings to the various signals: this is a muscle
tension, that is an acceleration, the other is a sum of
forces. But the computer itself operates without any such
interpretations.
Grandmother Level:
Bill, it seems to me that Martin and Oded's points are well
taken. A particular constellation of features or attributes (e.g,
Grandmother) is not necessarily realized as a particular scalar
value at some unique level (grandmother level?) of perceptual
processing. After all, we don't control grandmother as such; we
only control her attributes or features, such as her proximity,
her disposition, etc., etc..
When Aristotle distinguished five separate senses he also
identified three principles of mental association by which the
various isolated sensations, hypothetically, could coalesce to
form unitary percepts. Neurophysiologist have been looking for
these synthetic engrams (cells) ever since, to no avail. Although
contemporary neurophysiologist are still looking, others have
begun to recognize that the quest is a wild goose chase. Parallel
processing is not necessarily isolated parallel processing. As
you say, "When you put signals into an analog computing network,
you get signal variations everywhere in the network, not just at
the nominal outputs. All of these variations have meaning, because
they stand in continuous relationship to the inputs, according to
rules that are the properties of the intervening devices."
New Software:
Bill, the 5.25" floppy you sent me recently with the latest
version of Demo2 appears to be defective. Could you please send
another. Could you send copies of you manuscrits as well--the
current ones with Greg and Tom?
Warm regards, Wayne
Wayne A. Hershberger Work: (815) 753-7097
Professor of Psychology
Department of Psychology Home: (815) 758-3747
Northern Illinois University
DeKalb IL 60115 Bitnet: tj0wah1@niu