Bickhard on Gibson

[from Gary Cziko 930905.1300 UTC]

Bill Powers and Bruce Nevin:

If you'd like to read an interesting perspective on Gibson, I highly recommend:

Bickhard, M. H., & Richie, D. Ml. (1983). ON THE NATURE OF REPRESENTATION:
A CASE STUDY OF JAMES GIBSON'S THEORY OF PERCEPTION. New York: Praeger.

Bickhard and Richie contend that Gibson was moving in the right direction,
away from a mediated, "encodingist" view of perception to an interactive
one but didn't go far enough in that direction.

017.2 "Gibson's basic insight was that it is possible to derive information
about an environment from interactions with that environment without
encoding anything from that environment."

I found this book of particular interest in that it shows that PCT avoids
not just one but the TWO fatal mistakes of mainstream psychology. We all
know the first (that perception controls behavior). The second is that
perception involves an internal encoding of aspects of the environment.

Here are some other quotes from the book to give some of its flavor
(preceding numbers are page.paragraph).--Gary

···

==========================================================
011.1 Gibson quote: ""[It is] tempting to believe that the image on the
retina falls on a kind of screen and is itself something intended to be
looked at, that is, a picture. It leads to one of the most seductive
fallacies in the history of psychology--that the retinal image is something
to be seen. I call this the "little man in the brain" theory of the
retinal image (1966, p. 226), which conceives the eye as a camera at the
end of a nerve cable that transmits the image to the brain. Then there has
to be a little man, a homunculus, seated in the brain who looks at the
physiological image. The little man would have to have an eye to see it
with, of course, a little eye with a little retinal image connected to
a little brain, and so we have explained nothing by this theory. We are,
in fact worse off than before, since we are confronted with the paradox of
an infinite regress of little men, each within the other and each looking
at the brain of the next bigger man.""

016.4 "Suppose instead that the most direct focus of perception is the
functional nature of that which is perceived. Suppose that what are most
directly perceived are functional potentialities, potential usefulness.
The patterns of interactions that detect ambient light patterns, after all,
are not in any sense copies of those light patterns nor of the physical
surfaces and edges that yield them; they, rather, are simply interacting
outcomes that may indicate potentialities for further actions and
interactions. They are simply functional indicators.^2"

031.2 "Direct perception, then, in the sense of not involving mediating
encodings, is a valid metatheoretical position consistent with Gibson's
theory. Direct perception in the sense of direct encoding is an invalid
metatheoretical position inconsistent with Gibson's theory. The ease with
which the former is interpreted as the latter, both within and without the
Gibsonian camp, has obscured and retarded the basic task of understanding
perception."

051.3 "In both cases examined, Ullman (1980) and Fodor and Pylyshyn (1981),
the arguments are directed against a direct encoding interpretation of
Gibson from a mediated encoding perspective on perception. In both cases,
the general arguments are valid against a direct encoding interpretation,
through not against an interactive interpretation. Furthermore, in both
cases, the mediated encoding alternatives proposed are fatally vulnerable
to criticisms directed at their encoding foundations. In the Fodor and
Pylyshyn article, that mediated encoding alternative position is
particularly well elaborated; its vulnerabilities are thus exposed in
corresponding detail. In spite of Gibsons's own metatheory, then, a direct
encoding interpretation of Gibson is untenable. So also is the mediated
encoding alternative to Gibson. This leaves the interactive interpretation
of Gibson and the interactive approach in general."

057.4 "Encodings can only be explicated in terms of interactive
interpreters, yet such intentional interactive systems have been presumed
to be definable only in terms of (operations on) encodings. Strict
encoding, then, are circularly incoherent: They presuppose themselves."

058.1 "The incoherence of strict encodings, then, derives from their
presuming to be the grounds for explicating that which must in fact be the
grounds for their own explication. The impoverishment, insufficiency, and
lack of necessity of encodings all derive from encodings being
specializations of a broader representational function: the impoverishment
and insufficiency because that specialization does not capture all of the
broader interactive representational power, and the lack of necessity
because such specialization, however efficient it might be, is itself not
necessary."

058.2 "The claim that a level of interactive systems exists in this manner
between the material and the representational encoding levels actually has
four parts: 1) that such a nonmaterial, not necessarily representatinal,
level exists, 2) that representation can be explicated in terms of this
level, 3) that representation must be explicated in terms of this level,
and 4) that encodings can and must be explicated in terms of such
interactive representations. That such a level exists is easily
established: It is the domain of such areas as automata theory, control
structure theory, cybernetics, and so on. All of these areas study
processes, potentially interactive processes, in terms of their abstract
pattern of organization--that is, without essential reference to the
particulars of their material level instantiations--and without
presupposing or (necessarily) instantiating any representational
phenemena."

059.4 "it can be argued . . . that skills and values can be understood
interactively, but not in terms of encodings. The representation by
indicator differentiation of a simple feedback system also provides an
example. An internal outcome indicator can be used to differentiate and
select ensuing courses of interaction, but it contains essentially no
information about what in the environment yielded that outcome; it contains
or constitutes no information about what it might be taken to encode and,
thus, no information that would allow it to be an encoding. We, as
observers of the system, might claim that it encodes such-and-such a
physical state of affairs that yielded that outcome, but the system itself
has no necessary knowledge of that sort, and it certainly need not infer
its next course of interaction on the basis of such encodings: Does a
thermostat infer its actions on the basis of encodings of temperature, or
does it simply discriminately respond to internal indicators? We confuse
our perspective as observers with the perspective of the system being
[060.1] observed when we take such indicators as encodings. In effect,
encodings presuppose and require too much knowledge to be able to explicate
such simple differentiating indicators."

083.2 "James Gibson's theory of perception, both internally and with
respect to the controversy surrounding it, exemplifies the conflict between
two fundamental approaches to the nature of representation: the encoding
approach and the interactive approach. We have argued that the encoding
approach forms an impoverished, asymptotically limiting case of the
interactive approach and that it is insufficient, unnecessary, and
incoherent when taken as a logically independent approach to the nature of
representation. The issues have been examined primarily with respect to
perception and with a primary focus on Gibson's theory, but those issues
are fundamental to all of psychology. It is hoped and expected that the
exploration of those issue will prove interesting and productive throughout
their domain."

[From Rick Marken (930905.1700)]

Gary Cziko (930905.1300 UTC)--

Bickhard and Richie contend that Gibson was moving in the right direction,
away from a mediated, "encodingist" view of perception to an interactive
one but didn't go far enough in that direction.

What is an "encodingist" view? What is an "interactive" one. I think
my view of perception could be called "encodingist" and I'm a card
carrying PCTer. I believe that perpcetions are signals resulting from
neural transformations of input sensory data: p = f(s). The function f()
determines what variations in s are represented by variations in p. The
magnitude of p can be considered a "code" -- it is simply an analog
neural representation of something about s. You can decode p if you are
given f(). That is, you can determine what sensory state any particular
value of p "coded" by solving s = f-1(p).

I found this book of particular interest in that it shows that PCT avoids
not just one but the TWO fatal mistakes of mainstream psychology. We all
know the first (that perception controls behavior). The second is that
perception involves an internal encoding of aspects of the environment.

Depends on what is meant be "encoding". PCT makes clear that there is
not necessarily a physically significant correlate of any particular
perception, p. For example, there is no "taste of lemonade" out there
(just a particular function of different intensities, sensations, etc).
But the p that IS the taste of lemonate exists because some neural
function derived it from other inputs. I always got the impression that
Gibson was saying that "invariants" in the environment are just
"detected" -- as though that somehow solved a problem. It seems to me
like the ULTIMATE in non-model based Behavioristic explanation because
it encourged people to look for explanations of a psychological phenomenon
(like our perception of a chair) in the environment. Gibson is saying
that there are these "invariants" out there that "cause" perceptions;
this is exactly like Skinner saying that there are"reinforcements"
out there that cause (select) behavior. Gibson is as far from the
PCT view of perception (it seems to me) as Skinner is from the PCT
view of control -- for similar reasons.

The quotes from Bickard & Richie suggest that they are as confused as
Gibson; but maybe I'm confused because I have no idea what they mean by
"mediated, encodingist" (which seems like an OK description of my view
of perception) or "interactivist".

Help me out here.

Best

Rick (the invariant -- but willing to vary if necessary) Marken

[From Bill Powers (930905.1845 MDT)]

Gary Cziko (930905.1300) --

RE: Bickhard & Richie on the nature of representation.

You (they) cite Gibson as saying

... one of the most seductive fallacies in the history of
psychology--that the retinal image is something to be seen. I
call this the "little man in the brain" theory of the retinal
image (1966, p. 226), which conceives the eye as a camera at
the end of a nerve cable that transmits the image to the brain.
Then there has to be a little man, a homunculus, seated in the
brain who looks at the physiological image. The little man
would have to have an eye to see it with, of course, a little
eye with a little retinal image connected to a little
brain, and so we have explained nothing by this theory. We
are, in fact worse off than before, since we are confronted
with the paradox of an infinite regress of little men, each
within the other and each looking at the brain of the next
bigger man.""

This quotation shows only how limited Gibson was as a modeler.
For neurons to represent an optical image, it is necessary only
that each distinct neural signal represent the magnitude of one
of the constructable attributes of the retinal stimulation --
color, edge, orientation, distance, size, shape, motion, and so
forth. Each independent attribute can be represented as a single
one-dimensional signal. There is no necessity for the
representing signal to have any physical resemblance to that
which is represented. Representation by analogs in a hierarchical
structure does not imply any infinite regress.

So Gibson was simply confessing that he didn't understand how
there could be representation unless it was duplication. He
should have consulted an engineer.

The infinite regress bogeyman has been invoked over and over, but
it's a straw bogeyman. I don't know of anyone who has ever
advocated this view of perception: it was invented by those who
needed some obviously silly argument to oppose, presumably
because they didn't know how to oppose more serious ones. The
same method was used to argue against the idea of purpose: the
future can't affect the present; purpose requires such effects to
occur; ergo, purpose doesn't exist. The straw bogeyman is the "of
course" assertion that purpose requires an effect of the future
on the present. The whole problem was that the opponents of
purpose couldn't think of a way for purpose to exist without that
assumption. If they had tried a little harder they might have
avoided embarrassment.

The solution offered by Bickhard & Richie to the supposed
"encoding" problem doesn't seem any more coherent to me than what
it replaces:

"Suppose instead that the most direct focus of perception is
the functional nature of that which is perceived. Suppose that
what are most directly perceived are functional potentialities,
potential usefulness. The patterns of interactions that detect
ambient light patterns, after all, are not in any sense copies
of those light patterns nor of the physical surfaces and edges
that yield them; they, rather, are simply interacting outcomes
that may indicate potentialities for further actions and
interactions. They are simply functional indicators.^2"

Copies of WHAT light patterns? WHAT physical surfaces, WHAT
edges? WHAT "functional nature"? By what magical extrasensory
means do B&S know there are such things? The attempt at an
argument is destroyed by the very terms it uses, which assume
exactly the kind of direct knowledge that is supposed to be part
of the conclusions.

I don't see any improvement here: it is still assumed that the
perceiver somehow knows directly of the environment, only now
what it known consists of abstractions and concepts instead of
things and events. B&R assume that they know what is really there
in the universe, as opposed to what we perceive of it. There is
still the problem of showing that the abstractions and concepts
known to the speaker of the above words correspond to anything
outside the speaker. The epistemological problem has been
distilled into vapors, but has not been solved.

I think that the basic problem is the attempt to anchor perceived
reality outside the organism by hook or by crook, mostly by
crook. There is a much simpler solution that I have proposed
before. It is to say that perceptions take place inside the
brain. This is the only view I know of that is consistent with
all our models as well as with private experience. There is no
encoding problem, so we don't need elaborate refutations of it.
The world we experience, as I have said before, has already been
"encoded" or "represented" by the time we become aware of it --
that is, this world that we experience is a neural world, not the
external world at all. It may be, according to physical and
neurological models, derived from the impingement of external
energies on sensory endings, but that is only an hypothesis, to
exactly the same extent that physical and neurological models are
hypothetical -- neither more nor less hypothetical.

If we propose that the world we experience is identically a world
of neural activities in a brain, we achieve complete harmony
among all our basic models of reality. The model that says there
is a brain and nervous system in the first place becomes
consistent with models about brain and nervous system
biochemistry, with models about the relation of neural signals to
physical processes via sensory endings, with models of the
physical environment and the actions of organisms on it. And
harmony is achieved between these models and private experience
of the world, because to say that the world that is experienced
is a system of neural signals in a brain does not change the
experience of that world one bit. It simply changes what we say
about it.

If you believe that ANY ASPECT of the world you experience is
outside your brain, then you DO have an encoding problem, a
problem of representation, a problem with infinite regress, a
peculiar sort of epistemological problem. You're required to
select certain perceptions as having a special ontological status
(the very thing, of course, that you're questioning). You're
required to propose a purely magical way of knowing about the
world. You're led to see concepts and other abstractions as
having a different sort of existence than sensations, objects,
events, and other low-order perceptions. As long as you cling to
the idea that ANY part of experience is something other than a
neural signal in a brain, you're forced into all kinds of
intellectual contortions and unwitting self-contradictions.

I've heard my view dismissed as a return to Bishop Berkeley. All
this shows is inattention. How could I speak of a "brain" if I
were rejecting an objective reality? How could I speak of
"sensory endings?" The fact that I consider physics, neurology,
chemistry, and so on to deal with an hypothetical universe
doesn't mean that I know these hypotheses to be wrong. It's just
that, being hypotheses, they can never be proven in a positive
sense. The goal of the old-time epistemologists, to identify
justified true knowledge, is a will-of-the-wisp. That is simply a
wish that such knowledge existed. I see no reason whatsoever to
suppose that it does exist, or even wish that it did.

Under my proposal, the problem of perception is not that of
figuring out its end-product. We already experience its end-
product, directly. We call it "the world." The problem is to work
out models that explain how this end-product could be derived
from the world that is described by our other models -- physics
and chemistry and neurology and so on. Ernst von Glasersfeld once
said that the brain is not the black box: the universe is. He
later sort of disclaimed this view, but if he doesn't want it any
more, I'll take it. The world of experience is laid out before us
in plain view. That is how some of the neural signals in the
brain look to human awareness. It's the cause of that world that
remains mysterious and hypothetical, that calls for a series of
interrelated models that attempt to infer what is going on in
that unexperienced world beyond experience.

There is much in the B&R writings that suggest vague parallels
with PCT. But the level of abstraction is so great that any
actual parallels are simply conjectural. When an opportunity
comes to make a more direct comparison, we find things like

" The representation by indicator differentiation of a simple
feedback system also provides an example."

That sounds more like a smow job than an example. What the hell
is the indicator differentiation of a simple feedback system? It
sounds like Geordi on New Generation explaining how a warp-drive
generator works.

···

----------------------------------------------------------------
Best,

Bill P.