[From Bill Powers (940908.1420 MDT)]
Bruce Buchanan (940907.2010) --
In view of the fact that what we perceive is a function of the world as
well as of the perceptual apparatus, perhaps behavior may
alternatively and reasonably be described as Control of _What We
Perceive_.
_What We Perceive_ is undeniably a function of the perceiver, but it is
not _only_ that. It is also a function of what is perceived, and it
may both clarify understanding and and facilitate wider acceptance of
PCT to acknowledge that fact.
This is actually the view I am trying to get across, but without begging
the question of what is perceived. It's very easy to slip in an
unwarranted assumption here. We can say "Yes, my perception of an apple
is in part due to the construction of my perceptual systems, but it's
also due in part to the apple." Of course to speak of "the apple" apart
perceiving it, to know of the apple. And that is what I say we do not
have.
The claim I want to make about reality is much weaker than saying that
my perception of the apple is in part due to the apple. I want to claim
only that it is due in part to some regularity in the world. There is
evidence for that, because we can correlate taste, vision, audition, and
touch sensations all of which seem to have to do with one thing that we
call an apple. Moreover, in order to alter the experience of the apple
in any desired way, we must perform certain acts on the world outside.
We can't pick actions at random, because only some actions will have the
desired effects. We must discover what they are empirically. There is
strong evidence that there are regularities in the outside world
involved in our perception and control of apples. But this evidence
concerns only the _existence_ of the regularities; it does not give us a
unique description of them. It does not say that in the outside world,
there is an entity corresponding to the perception we call "apple." Of
course this argument does not say there is NOT such an entity. But the
basic problem here is how we could ever establish that there is one.
I'm perfectly willing to play Let's Pretend. I do it all the time in
modeling. Let's pretend that corresponding to the perception of the
position of a target, there is not only something called a target really
out there, but something called "position." If there is a perception of
the distance between a fingertip and the target, let's pretend that
there is not only a real fingertip, but something corresponding to
"distance." With this set of postulates accepted, I can proceed to merge
a physical model of the fingertip and target, where laws of geometry,
optics, and inertia are involved, with a functional model of a nervous
system in which there are signals representing fingertip, target,
positions, and distance. The two models fit together nicely; they
should, because they were both constructed to be self-consistent and
consistent with human observations of the world. I can easily pretend
that this is a true picture of how the organism relates to its
environment. But that doesn't solve the real problem.
The nature of the problem is perhaps best investigated by simulations.
We can set up a world with known objective properties, and a control
system that perceives some function of the variables in that world, and
show quite easily that there is no need for the controlled perception to
correspond to any entity in the external world. We can set up a world,
for example, in which there are two variables, x and y. We can set up a
control system to perceive ax - by, and to maintain the resulting
perception at any arbitrary reference level by converting the error
signal into a positive effect on x and a negative effect on y. This can
be done for any arbitrary choice of coefficients a and b other than
zero.
The control system will act exactly as if there were a weighted
difference between x and y in the world, which could be disturbed and
also kept under control against disturbances. Yet in the environment,
there would be nothing but x and y, which need have no relationship to
each other. The variable x could be the temperature of the air, and the
variable y the brightness of a picture on a television set. The value of
ax - by could be brought to any desired value and maintained there
despite disturbances acting on the air temperature and/or the brightness
of the picture, given only that we have actions which can affect
brightess and temperature (or either one alone) in the right direction.
I have run such models repeatedly, and Rick Marken has developed a
similar model with three levels of control and six environmental
variables. The environmental variables in such models don't correspond
to any physical variables; they're just arbitrary variables with no
meanings assigned.
So by looking at the perceptual signals in the control systems, what
could we deduce about the external world? Those signals appear to vary
in systematic ways, which the simulated system can control by systematic
action on its world. Yet the signals standing for the individual
variables x and y (or x1 through x6) could have any physical meaning
whatsoever; the physical meaning is irrelevant. The apparent variable p,
where p = ax - by, seems to represent an entity in the world outside.
But it was constructed in an arbitrary manner, and exists only because
there is a perceptual input function which receives a signal standing
for x and multiplies it by a, and receives a signal y and multiplies it
by b, and adds the two products together algebraically to determine the
momentary value of the perceptual signal.
It seems to me that this sort of demonstration very clearly shows the
relationship that exists between perception and the world. Yes,
perceptions depend on the world; no, they do not necessarily have
correlates in the world. Until we fully accept this apparent fact, there
is no way we can make a systematic attack on the epistemology of
perception. If we keep trying to make perception into something that
magically corresponds to real things in the world, we will never find
out how it is that perceptions come to be the way they are. I think we
can come up with a convincing picture, but we can't do it if we become
alarmed at what the implications of the first steps seem to be, and shy
away from simple reasoning because we're afraid of where it might lead.
Once we accept the simple picture that we can get from straightforward
reasoning, we can begin to see how progress could be made toward a non-
solipsistic epistemology. Donald T. Campbell pointed out one direction
we could take, in his concept of "triangulation."
Suppose we have four variables, x,y,s, and t. As it happens, let us say,
there is a law in the world that says ax - by = cs + dt. Let us suppose,
furthermore, that we have two control systems, one perceiving p1 = ax -
by and the other perceiving p2 = cs + dt. There are now two perceptual
signals that covary if the values of the weightings a,b,c, and d are
exactly right; p1 will equal p2 at all times. If the values are
multiples of the right values, p1 will be a multiple of p2 but strictly
proportional to it. So the behaviors of p1 and p2 are no longer
arbitrary; there is a relationship between them which can be found if
the weightings of the first level of perception are correct, or related
to the correct values in a simple way. To control s and t in a certain
way is to control a related function of x and y. That relationship is
set by the property of the world which says ax - by = cs + dt. A
perception made of p1 and p2 would therefore reflect a regularity in the
world, although it would not be a direct representation of the nature of
that regularity. Why such a correlation between perceptions would be
sought is a matter for further thought. But the fact is that we have
here one circumstance in which a law in the environment would lead to a
discoverable relationship between otherwise arbitrary perceptions.
The other route is the one on which Martin Taylor has been working (I've
been considering it, too, for some years). It might be that the
difference ax - by is closely related to some physiological effect of
the world on the organism containing the control system. The variable x
might correlate with a positive effect on some critical variable, while
y correlates with a negative effect on the same variable, in the
proportions a:b. Given a reorganizing system that monitors the critical
variable and alters the organization of the control systems until that
variable is maintained at some particular level, we could understand why
a control system might come into being that controls ax - by. The
construction of the perceptual signal from the inputs x and y is still
completely arbitrary, but now another constraint has been added: the
control system must ultimately come to perceive and control a function
of x and y such that the physiological effects of x and y on the
critical variable keep that variable near its genetically-specified
reference level.
Now we have supplied a missing ingredient in the epistemology with which
we began: a way for the organism to know something objective about the
correlates of perceptual signals. This knowledge is highly indirect, but
it involves a route by which information about the environment can get
into the organism in a way other than through the senses. This
information comes in through direct physical effects of the environment
on the body of the organism.
At the very least, the requirement that these direct physical effects
leave critical variables near their reference states puts a constraint
on what variables the organism will sense and control with its
behavioral systems.
So we now have two approaches to the problem, which beconme apparent
only when we accept the arbitrariness of perception. We can look for
principles of reorganization based on relationships among arbitrary
perceptions, and we can look for principles based on direct
physiological effects of variables associated with perception. Both of
these approaches entail considering properties of the outside world,
independently of how they are perceived. Both approaches give the actual
properties of the real world an influence on how the hierarchy of
perception and control will develop.
This is just a bare start toward a real solution, toward answering the
question of whether perceptions correspond to meaningful entities in the
outside world. It has the great advantage over a leap of faith that it
can accomodate cases where perceptions clearly do not have any external
counterparts, or where they even misrepresent the external situation (as
in illusions). Those mistakes of perception are unimportant if it can be
shown that some perceptions, for logical and demonstrable reasons,
necessarily correspond to meaningful external entities. The way they
come to do so will, I believe, be undertood ultimately as the outcome of
a continuing process of reorganization that converges toward
veridicality, at least in critical dimensions.
It may be that even when this development is finished (and it is a long
way from finished), we will still find a basic ambiguity between
perception and reality. But we will have moved a step closer to knowing
how experience relates to That Which Is.
I find this approach to epistemology a lot more interesting than playing
Let's Pretend.
ยทยทยท
from my perception of it is to claim to have another way, beside
-----------------------------------------------------------------------
Best,
Bill P.