[From Bill Powers (940914.1445 MDT)]
Bruce Buchanan (940913.2200) --
... the perception _is_ a _function of_ reality. We are not aware of
_the perception itself_. We are aware of reality _through_ (albeit
conditioned by) the perception.
Suppose we are looking at a surface which physical measurement shows to
reflect two wavelengths out of all those in an illuminating white light:
a red wavelength and a blue wavelength. What we perceive in such a case
is a single color: purple. What we are aware of is purpleness; but there
is no purpleness reflected from the surface, only red and blue. I don't
see how, in such a case, we could say that we are aware of purpleness
_through_ our sensory perceptions. There is no purpleness there,
according to physics, to be aware of.
The basic problem is that single unitary perceptions are functions of
multiple environmental variables; this is the situation at all levels,
even at the lowest levels of sensory perception. The physics model
therefore disagrees with direct experience.
The only way I know of to resolve this disagreement is to assume that
the nervous system combines two signals, one standing for red and one
for blue, adds them together with appropriate weightings, and generates
an output signal as a result. And further, we apparently must assume
that the subjective perception of the color purple is associated with
the output signal from this neural computer, and not with either of the
input signals. If we make those two simple assumptions, the physics
model, the neural model, and subjective experience all come into
agreement.
There is great reluctance among some scientists to accept this simple
resolution of the difficulty. I think the reason is obvious: it has the
same sort of effect on our concept of the world that the Copernican
revolution had: it says that the appearances of the world are a matter
of our point of view. It says that no scientist, however good an
observer, can observe the world as it really is. We are all constrained
to see the world that is presented to us by human perceptual functions,
which are apparently, and fortunately, similar enough within our species
to allow us to have similar experiences of the world.
When Dr. Johnson kicked the stone and said "I refute it _thus_" he was
not refuting the subjectivity of the world of experience; he was
confirming it. The only way he had of verifying his visual impression of
the stone was to provide himself with a tactile impression of it which
he felt in his toe. However, he proved something that he did not know he
was proving; that there is in fact an objective world that we can't see
or feel. He knew that by acting in a certain way, he could alter his own
sensory world, and he did so. But if you had asked him to explain how it
comes to be that generating a certain action affects both a tactile and
a visual part of the world, he could not have told you; he wasn't a
physicist, and knew nothing of physical models of motion,
compressibility, mass, inertia,and so forth. If you had asked him how
intending to kick the stone results in actually kicking it, he couldn't
have told you that, either; he was not a control theorist.
What the epistomology of control theory says is yes, there is a physical
environment of which perceptions are a function, and no, the perceptions
do not necessarily have counterparts in the environment. The perception
of purple is a function of the intensities of red and blue light in the
environment, as we model that environment in physics. It depends on the
wavelengths of electromagnetic energy that are present, or the energy
contained in light quanta (pick your theory). But what we experience has
no simple connection with wavelengths or quanta, just as concepts of
wavelengths and quanta have no simple connection to whatever is actually
going on out there. There is no thing called purple that exists in the
objective environment. Purple is a way that our perceptual signals sum
up a set of environmental variables, a way peculiar to human beings.
···
-------------------------------
----, (940912.0649)--
My understanding is that any specific control system has a reaction to
disturbances which depends upon the deviations introduced as the
disturbance, not on its origins and nature. But the nature of the
problem, while outside the purview of the individual control system can
affect the system in other ways that [it] has no inkling of. This might
be important to the life of the system but not register with it as a
control mechanism.
You are re-inventing the PCT model, which is reassuring to me. This is
part of the model we are trying to build, to show how physical effects
of actions (assuming the physics model) may affect the organism in ways
other than through the senses. These effects are not conditioned by the
way we perceive the world and our bodies, but by the actual
relationships that exist, of which the behaving system has no direct
sensory knowledge and which our physical models can only approximate.
The reorganizing system is an attempt to take effects like these into
account. The basis for action by the reorganizing system is a set of
variables inside the organism which are affected physically by external
and internal events, without sensory systems intervening. Of course even
the reorganizing system must detect these variables before they can be
compared with reference signals, so there is still no direct knowledge
of the real world involved. But evolution has seen to it that certain
variables in the body are sensed, and that a reference signal is
provided for each one specifying its design-center state. Speaking
metaphorically, these sensor signals and their reference signals amount
to hypotheses made by the species, to the effect that these variables
are critical to survival, and that they must be maintained at specific
reference levels. Since these variables are directly affected by the
real world, they represent actual effects of behaviors and disturbances
on the organism, not hypothetical effects of variables created by
perceptual organizations.
The outputs of the reorganizing system alter the organization of the
learned hierarchy, including its perceptual functions, in ways not
reducible to any existing algorithm -- that is, at random. This is the
only way in which changes appropriate to the actual environment can be
made in the absence of direct knowledge about the actual environment.
Fortunately, this process or one akin to it can be quite efficient; that
is, it can actually produce changes in organization which result in
altering the environment (as a side-effect of controlling perceived
variables) so that actual effects on the critical variables bring those
variables closer to their reference states.
The very randomness of the output process bypasses the need for "true"
knowledge of the external world. Only some very permissive constraints
on the real world are needed to allow this process to work without
knowledge of actual external relationships. The main one is continuity;
that small changes have small effects. Where this condition doesn't
hold, reorganization may become difficult or impossible, and some other
process may be needed (if there is any process that can work). But
reorganization, as outlined here, can work in most ordinary situations,
or so it seems now.
The concept of reorganization and of the kind of system needed to
implement it bring us very close to phenomena like those responsible for
evolution. This is why I keep searching for a mechanism behind
evolution, and a criterion for "mutation" other than simple success or
failure. Reorganization would not work if the only criteria were "error"
and "no error." It would not work efficiently enough. There must be some
way to detect the difference between the existing state of affairs and
some desired state, and to tell whether this difference is increasing or
decreasing even if a substantial difference still exists. Otherwise the
process would be like a game of "hot and cold" in which nothing was said
to the player until she had actually found the hidden object.
All evolutionists tacitly recognize this problem. Look at the proposed
simulation that Martin Taylor came up with: the survival rate and the
reproduction rate did not depend on reaching the goal position, but only
on improving the position of the organism in a continuum of effects on
survival and reproduction. While the computation of improvement or
worsening of the situation was not explicitly attributed to the
simulated organisms, it was there in the program, and had to be there
for this approach to work. The model is _everything_ in the program that
is needed for its operation, whether or not the ancillary computations
are part of one's conception of the model.
Sorry that I keep veering back onto the evolutionary thread. It seems to
be much on my mind these days.
------------------------------
I am beginning to think many (all?) PCTers see perception in the
PCT framework as in the overall driver's seat for controlling all
behavior.
Ah-ah-ah! Perception does not control behavior. Behavior controls
perception (via actions on the external world). When this begins to seem
natural to you, everything will fall into place
------------------------------
So, as I see it, logical relations are the sine qua non of any science.
Logic is fine in its place, but it must be used consistently with
principles, which are of a higher level. "Things which are equal to the
same thing are equal to each other" is not a logical deduction; it is a
principle to which all logical developments must conform. "A straight
line is the shortest distance between two points" is a principle which
dictates the kind of geometric logic you will use; change that
principle, and you will have to use a different logic (geometry on a
sphere, for example). Logic is a tool for reaching goals you want to
reach. The Nazis were very logical in their approach to genocide. Higher
levels than logic are at least as important as doing correct logic.
----------------------------------------------------------------------
Best,
Bill P.