solipsism; inference; thermostats

[From Bill Powers (950301.0700 MST)]

Martin Taylor (950228.11:20)--

The solipsist has made a claim; let him support it.

     Why should he? YOU DON'T EXIST. And you know it, since you are
     inside the solipsist. Oh--sorry--I don't exist either. I should
     be quieter...

If we were ever to encounter a true proponent of that view (rather than
a non-solipsist attempting to imitate solipsism for the sake of
argument), we would see a person curled up in a foetal ball, oblivious
to us. When you try to imitate a solipsist, the first word in defense of
that position is a self-contradiction, because no sane solipsist would
attempt to persuade someone else whom he knows to be merely a figment of
his imagination. He would simply change that imagined person's mind.

Solipsism is a game of "let's pretend," not to be taken seriously by the
sane. To take it seriously is to start down a long dim corridor with
oblivion at the end of it.

     How could a solipsist assert that there could be a fact that is
     true independently of the observer, when there is no other
     observer?

He (let's pretend) is saying that solipsism is true independently of
himself -- that is, whether he wants to believe it or not. That
establishes the existence of at least one truth independent of at least
the observer in question. If the solipsist were ever to admit that the
fact of solipsism is being invented by himself, he would recognize that
belief in solipsism is a voluntary and arbitrary act which could be
undone as easily as it was done. This would cure him of his condition.

Solipsism is simply another form of paradox, one of those logical loops
one can get into like "The statement on the other side of this card is
true" and "The statement on the other side of this card is false." Any
real system organized in the implied way would necessarily buzz back and
forth between the two conditions. The way out is to step back, perceive
the loop, and stop it (or at least enjoy it as a phenomenon without
being caught up in it).

Pure logic always has this problem: it can become organized to create a
paradox, and there is no logical way out of a paradox. The only way out
is up a level -- as I believe Bertrand Russell finally realized.

     The solipsist position can't be defended--true--but neither can it
     be attacked. As I said, it's like creationism. Every possible
     observation you could make can be argued as "that's the way I (or
     God) imagined it."

It can't be attacked on logical grounds because it entails a paradox. If
God imagined that you believe in God, then God could make it otherwise,
in which case God would cease to exist, or would never have existed in
the first place.

But it can easily be attacked on other grounds, such as by recognizing
the principle that logic can be misleading, or by adopting a system
concept in which the universe was imagined by God so that we are once
again allowed independent existence inside that universe.

     I believe exactly what you said before, as well as now: "the case
     where we perceive one form, but it is a transformation of some
     quite different -- but still existent -- form." We cannot know
     that transformation, but by our successes and failures of control,
     we can put some constraints on it.

All right, let's just agree on that.

···

------------------------------------
     If I see an "apple" I am inferring in just that same way as if I
     see a "rotten" apple. To go out on a limb, here, I expect that
     much of our perceptual apparatus is the way it is because it allows
     us to see (hear, feel, smell) things that are good to eat,
     poisonous, dangerous to be near, and so forth.

When you are inferring a rotten apple from the perception of a brown
spot on the apple's surface, you are specifically imagining something
that you are NOT CURRENTLY PERCEIVING. You ARE perceiving the brown
spot; you ARE NOT perceiving the rottenness. To perceive the rottenness
you would have to cut the apple open and see that it is brown inside,
too, and perhaps taste it to find the perception you call "rotten."
There is a possibility that when you cut the apple open you will find
none of those signs of rottenness; the brown spot may simply be a
discoloration of the skin.

In other words, inferences are based on imagination. That is the sort of
inference you speak of above: whether things with a certain appearance
will be good to eat. What you're giving is an imagined reason why we
should imagine as we do. I don't believe that our perceptual apparatus
is as it is because making such inferences is useful to us. I believe
that we make such inferences AND that they are useful to us, sometimes.
You're putting the cart before the horse; evolution does not decide that
it would be good for us to be able to make indirect inferences about
reality, and then build an apparatus that will give us that ability.

But none of that is relevant to the point I was making, which is that we
can tell when we are perceiving and when we are inferring or imagining.
Even when we automatically turn away from an apple with a brown spot on
its skin, we can remind ourselves that we are making an inference, and
that we are not perceiving that which we infer. This allows us to
investigate further, and see whether the inference is correct by looking
inside the apple -- whether the perceptions we imagined would be there
are actually there.

This is not analogous to the hierarchy of real-time perceptions. The
higher perception does not depend on imagining that the lower
perceptions are there; it depends on their actual presence, because it
is a function of them.

     The word "inference" is slippery. Helmholz used "unconscious
     inference" to refer to all of perception.

That, again, is a metaphor. In terms of the PCT model of perception, it
is not an appropriate one. We distinguish between imagination, which is
a self-generated perception, and real-time perception, which depends
lawfully on the presence of signals in a lower-order world and
eventually, although in ways not presently known, on variables in the
external world.

The basic phenomenon of inference, as I see it, is that of experiencing
something which is not a present-time perception.

     It gave the moral authority a century later for people to use
     computer programs to "infer" from sensory data what was in a scene,
     and to argue that the "AI vision" might be the same as biological
     vision.

It also perpetuated the assumption that perception is simply a matter of
recognizing what is really there. What "is" in a scene is the
orderliness that our perceptual apparatus puts into the scene. As you no
doubt realize, the moral authority you mention allowed us simply to beg
the basic questions about perception we have been talking about. True
"AI" vision might come a lot closer to biological vision if we let the
perceiving system organize itself in any way it found convenient,
without regard to how we happen to see "the scene." We bypass this
problem by building "self-organizing" networks, and then telling them
what they _should_ be seeing -- which, of course, is what we ourselves
see.

     For me, "inference" has the connotation of sifting through a number
     of possibilities, probably by iterating some choice procedure.
     Perception, at least below the program level, doesn't do that.
     What you see is what you get, so's to speak. To apply "inference"
     to everyday perception, as Helmholz did, seems perhaps to extend
     the range of meaning of "inference" a bit far.

I agree, and that's not what I mean by inference, as should now be
apparent. I specifically wish to distinguish (real-time) perception from
inference.
-----------------------------------------------------------------------
Bruce Buchanan (950228.1200 EST) --

     An example (since I do not know how to render a diagram with my
     email program!): The thermostat measures the difference between the
     room temperature and its set point. Among the causal factors that
     affect the room temperature (e.g. outdoor ambient temperature, open
     doors and windows, use of lights and oven, number of people, etc.)
     is also the furnace burner and its time on and periodicity. The
     feedback signal goes from the thermostat comparator to the burner
     switch, where it is applied with negative sign, which is simply one
     of the causal input factors.

Here is a ASCII diagram, the only way, however tedious, to send a
diagram on the net:

                               > ref signal
                               >
                     -->- comparator ->-- error signal
                    > >
              percep>tion |
                    > v
                [sensor] [furnace] (output function)
                    > >
                  Temp | heat output quantity
                    ^ |
                    > thermal |
                    <--- properties -<---
                    ^ of room and air
                    > (feedback link)
                    >
                    disturbing heat losses and gains

You will note that the signal going from the comparator to the output
function (the furnace) is not the feedback signal, but the error signal.
The sign of the feedback is the product of signs all the way around the
loop: the negative sign is usually found at the comparator, where the
temperature perception is _subtracted_ from the reference signal. All
the other signs in the loop above are positive. There can be any mixture
of signs in the loop, as long as the number of negative signs is odd.

     Because the feedback links levels, the operation of one component
     (the furnace) can be integrated into a functional outcome of the
     whole system (room temperature).

The above is a one-level control system.

     This is one specific example of the model I had in mind. It is also
     an example of the way in which a quantitative feature of a
     superordinate system is precisely related to a quantitative aspect
     of a component subsystem, i.e. relation of whole to part, which I
     recognize is always an ongoing cyclical process if functional
     integration is to be realized.

Yes, the whole behaves in terms of the relationships among its parts. A
thermostat is cyclic in terms of time, but while the furnace is active
the room temperature is also changing and the perceptual signal is
changing. These components are always active; they do not take turns
acting. Only the fact that the comparator is an on-off contact closure
makes the behavior of this system show temporal cycles. In a continuous
control system where sensing is proportional, comparison is done using
continuous electronic subtraction, and the output of the furnace is
throttled in proportion to the error signal, there need be no temporal
cycles.

The system shown above is a whole; the controlling part is not
superordinate to the thermal properties of the room. The thermal
properties of the room are a component of the same loop. To get a
superordinate system, you would need, for example, a human being
controlling sensed comfort by varying the reference setting of the
depicted thermostat. Sensed comfort depends not only on room
temperature, but on humidity, activity, and clothing, so room
temperature is just one contributing variable in the control of comfort.
That is what makes sensed comfort superordinate to sensed temperature.
-------------------------------
     I am trying to clarify a possible mechanism by which events which
     occur in different times can be recorded by the nervous system so
     that they can be considered together (in fact by locating those
     records in different places in the CNS - e.g. in different memory
     traces and perhaps regions). In a sense time is translated into
     space and then managed in spatial terms, always, as Bill Powers
     notes, in the Present.

But you have proposed no mechanism: you have only said that these
phenomena occur. If you want to see some proposed mechanisms, read BCP,
chapter 15. I don't say that these proposals are correct, but they do
propose mechanisms rather than describe outcomes.
---------------------------------
     O.K. but, speaking more specifically, values in the large or at
     higher levels must be related to, or perhaps identical to,
     controlled variables.

Values are not controlled variables; they are reference levels against
which variables are compared in the process of making those variables
into controlled variables. There are controlled variables at every
level, not just at the lower levels. If I see you consistently being
polite to old ladies, I can reasonably conclude that you are adjusting
your relationships with old ladies so as to maintain a variable called
politeness, a perception at the principle level, near some particular
reference-state. That is just as observable a controlled variable as the
positions of your limbs as you usher the old lady through a door, and it
can be tested in the same way: by applying disturbances and seeing if
they are resisted.

     As understood by PCT these are perhaps not a problem. What is a
     problem to my mind is whether they may cast useful light on the
     methods and problems of value analysis addressed by social
     scientists, opinion researchers, politicians, etc.

Well, try using the PCT ideas that way and see what you come up with.
But make sure you are using PCT ideas, and not making up your own
nomenclature (at least not without announcing that you're doing so).
.

Have fun skiing with the family.
--------------------------------------------------------------------
Best to all,

Bill P.