Escaping cats; Newton vs. Einstein

[From Bill Powers (951121.1412 MST)]

Martin Taylor 951121 14:10 --

     Yes, that's the assumption on which the problem lies. The result is
     a cat-scientist who "knows" that lying on one's left facing the
     clock is the mechanism for opening the door.

     So I repeat:
        (1) How can the cat-scientist determine that the appropriate
     controlled perception is that of the stick moving?

He can't. Is that the answer you're looking for? The cat scientist can't
even tell if deliberately bumping the stick is "appropriate." All the
cat-scientist, or a human scientist, or anyone else can know is that
controlling one perception is sufficient to control another. You seem to
be asking how a scientist can know for certain that he understands what
is really, truly, going on.

     By hypothesis, the cat is NOT controlling for any perceptions that
     involve the actual release mechanism--in the example situation, the
     stick. But the experimenter, by naively applying the Test,
     determines that the cat IS controlling for a perception of the
     stick.

Then the experimenter should not apply the test naively. Every
experimenter-theorist I know treats all apparent answers with greatly
heightened skepticism: "Wait a minute, that was _too_ easy." Anyone with
physical sense should be able to investigate more closely to see how the
stick is being deflected, and to devise added tests that will eliminate
from consideration other variables that just happen to be correlated
with deflection of the stick (or show that they, not the stick, are
under control).

     Even so, the experimenter wants to know which perceptions are being
     controlled in support of the stick-move perception, and disturbs
     various candidate CEVs.

Are you thinking that the cat must _somehow_ include the movement of the
stick in the perceptions it is controlling? The stick movement _never_
needs to be a perception (except in the observer). Next you'll be saying
that the cat's perceptions have to include the linkage between the stick
and the latch on the door, or the internal mechanism of the latch, or
the friction in the hinges.

     Since the cat is really moving the stick only as a side-effect of
     some quite independent perceptual control, applying a disturbance
     to a controlled perception will change some kind of output, with
     resulting changes in side effects. If that change of side effect
     results in the cat not moving the stick, it will not escape, and
     will begin reorganizing, perhaps no longer controlling for the same
     things it had been controlling. The Test will then be failed,
     wrongly.

The Test is not going to reveal that the cat is controlling the position
of the stick. It is going to reveal what the cat is actually
controlling. Once we know that, we can explain (through physical
inspection) why controlling that perception entails moving the stick and
opening the door.

     Does the cat really "want" to lie on its left, and does it really
     want to face the clock? Yes, before the disturbance (such as moving
     the clock) that resulted in these behaviours failing to move the
     stick. But the higher reference perception is to be out of the box,
     and when lying on its left facing the clock stops working, it is
     likely to try something else, and the answer for whether these two
     perceptions are controlled becomes "No".

You are still thinking of disturbances that result in loss of control.
If you use that kind of disturbance, you are unlikely to succeed with
the Test.

     As I see it, the difficulty is in the fact that no matter what
     disturbances and changes in the situation (large or small) the
     experimenter introduces, the cat still has only one way to get out
     of the box. If the cat is controlling for a perception that is
     truly a part of getting out--moving the stick--then no matter what
     disturbances the experimenter introduces, the cat will resist them
     and move the stick. But if the cat is exiting as a side effect of
     some other control, it will still appear to be resisting the
     experimenter-induced disturbances, since it will in any case get
     out by moving the stick.

First, the experimenter makes sure that any disturbances that are
introduced don't keep the cat from escaping. I outlined a way to do this
by moving the fulcrum of the stick so that the stick was still pushed,
but with more force than before, while the cat went through its
stereotyped behavior. In this way we eliminate sensed force on the stick
as a controlled variable.

     I see better now what you had intended, but I'm not sure it
     resolves the problem. If the higher-level reference perception is
     to be out of the box, and it is the control of that perception
     whose output provides the (guessed) reference for, say, pressure on
     the stick, then the opening of the door would change the reference
     level for stick pressure, and the experimenter's disturbance would
     not be resisted in the same way.

Martin, this is just too confused for me to deal with. You're trying to
prove that "the environmental feedback function can allow side-effects
to mimic control." If you spent one tenth of the effort you have devoted
to proving this foregone conclusion on trying to find solutions to the
situations you're proposing, you would have provided your own answer.

If your way of applying the test results in
failure of control, you will be looking not at one control system but
a whole series of control systems of different organization, and will
learn nothing about any one of them.

     Yes, that's what would happen, and that's the problem, isn't it?
     You won't learn about any of them, but you will no longer conclude
     that the cat was controlling for the perception of the correct exit
     mechanism.

Why do you keep saying that? The cat _never_ controls for "the
perception of the correct exit mechanism" even if it walks up to the
stick and pushes on it. Do you think that the cat has to understand
_why_ controlling some variable results in opening the door? Maybe the
cat is controlling for the squeak generated by the mounting of the stick
when the stick is deflected. So what? Can't you think of a way of
telling what the cat is controlling for in such a case? I can, but you
don't seem to be putting much effort into finding such answers.

     If the cat had been actually controlling for a perception of stick
     movement, then it would have resisted the disturbances, the
     disturbances would not have prevented escape, and the other
     mechanisms would not be used. If the cat was moving the stick by
     side-effect, it would not resist those disturbances, and would have
     to reorganize, with a good chance that one of the other exit
     mechanisms would be used--the cat would convert to a different
     religion.

No, the cat would not have to reorganize if you were careful to use only
disturbances that did not prevent control. You're still thinking of
disturbances as a change in the controlled variable that the cat can't
easily counteract. Please wipe that assumption out of the discussion;
when the Test is done correctly, all disturbances are easily
counteracted. If you start seeing random changes in behavior, you'll
know that you violated that rule.

Haven't we about exhausted this topic?

···

-----------------------------------------------------------------------
Bruce Abbott (951121.1540 EST) --

     And lest we forget, Isaac was wrong. Albert said so.

Isaac was not wrong. Ask NASA whose equations they use in designing
spaceship trajectories. Albert's equations become identical to Isaac's,
within our normal limits of measurement, for all macroscopic effects we
can create by our own actions - relative velocities less than 100
kilometers per second.
-----------------------------------------------------------------------
Best to all,

Bill P.

[Martin Taylor 951121 17:05]

Bill Powers (951121.1412 MST)

the cat would not have to reorganize if you were careful to use only
disturbances that did not prevent control. You're still thinking of
disturbances as a change in the controlled variable that the cat can't
easily counteract. Please wipe that assumption out of the discussion;
when the Test is done correctly, all disturbances are easily
counteracted. If you start seeing random changes in behavior, you'll
know that you violated that rule.

Haven't we about exhausted this topic?

I wish we could sometime start to discuss the topic I am trying to address.

I thought on Friday we were beginning to do so, but since then, all we have
achieved is a set of cross-ed purpose messages in which you suggest that
I am trying to set up conditions I am not, and I propose conditions that
I (obviously wrongly) believe are instantiations of what you want.

We may have exhausted each other, but we haven't started to exhaust the
topic. We are engaging in the same random behaviour that the cat is
employing in trying to escape from the cage.

For example, in my discussion of allowing the cat multiple exit mechanisms,
I said:

    You won't learn about any of them, but you will no longer conclude
    that the cat was controlling for the perception of the correct exit
    mechanism.

To which you responded:

Why do you keep saying that? The cat _never_ controls for "the
perception of the correct exit mechanism" even if it walks up to the
stick and pushes on it. Do you think that the cat has to understand
_why_ controlling some variable results in opening the door? Maybe the
cat is controlling for the squeak generated by the mounting of the stick
when the stick is deflected. So what?

Do I have to take the long-winded wording EVERY time, and say "the perception
of the CEV that the experimenter knows that if properly controlled would
result in the exit mechanism being activated"? We'd get even more bogged
down in words. You really have to take some shorthand wordings for what
they (should) obviously mean. Making the point once should be enough, with
shorthand to follow.

You yopurself are not guiltless of saying things like "controlling
the angle of the steering wheel" in place of "controlling the perception
of the angle of the steering wheel" which could be even longer-winded if
the question was not of a "steering wheel" but of a "mechanism that the
mechanic knows to affect the orientation of the road-tracking wheels".
The situation here is analogous. The CEV is not a stick or pressure on a
stick, as such. It is whatever mechanism the experimenter has arranged
so as to open the door. I call that "the correct exit mechanism" to
distinguish it from "a spurious exit mechanism" that just happens to
activate the correct mechanism. "A spurious exit mechanism" might be
to "perceive oneself lying on one's left side facing the clock." The
experimenter knows that's wrong--in the "scientist" analogy, it's wrong
in the unobservable real world. The cat/scientist doesn't know it's wrong.
Yet.

You say "the cat would not have to reorganize if you were careful to use
only disturbances that did not prevent control" under conditions in which
it is taken as a foundational assumption that control is NOT PRESENT for
any CEV so far intuited by the experimenter, but in which every
sequential condition of the Test, as quoted by Chuck Tucker from you, is
fulfilled. We are obviously talking about quite different experiments
if you start by assuming that the cat is controlling some perception that
the experimenter already has intuited. Until we start talking about the
same experiment, we can't even make the subject leak a little, let alone
exhaust it.

There's a long trail ahead, in my imagination, in which the parallels
between the cat's problem and the good-or-bad scientist's problem are
to be investigated. But we can't start along the (possibly exhausting)
trail until we find where it leaves the parking lot. We can get very
tired racing in all directions across the parking lot under (or in my
case using) a hot sun, but we don't get very far along the trail that way.

The experimenter of whom we talk is just a cat of a different species.

Martin