[Martin Taylor 951121 14:10]
Bill Powers (951120.1500 MST)
Martin Taylor 951120 14:10 --
(1) How can the cat-scientist determine that the appropriate
controlled perception is that of the stick moving?The cat-scientist does not need to determine this. All it needs to
determine is that when it controls a certain perception p, the door
opens. The causal linkages between controlling p to a certain level and
the opening of the door do not need to be perceived. All that matters is
that when the cat controls p, a reliable side-effect (of which the cat
need perceive nothing) of controlling p is that the stick gets moved.
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. Less frivolously, in human terms,
it is a scientist/witch doctor who knows that a certain chant together
with burning the right grasses cures a hangover by getting rid of the
nasty demons, or a scientist who knows that a moving body goes on until
it runs out of impetus, or who recognized the true nature of reinforcement
in grooving behaviours into the right track. All of these cat/scientist/
witch-doctors get the results they want, so long as the environmental
feedback functions available to them don't change.
So I repeat:
(1) How can the cat-scientist determine that the appropriate
controlled perception is that of the stick moving?
···
------------------
(2) How can the god-experimenter determine what perceptions the cat
is controlling when (by hypothesis) the cat is continnually
reorganizing until a side effect of some perceptual control results
in the stick being moved?Presumably, when the perceptual control is found that will reliably move
the stick, so that reorganization ceases, the cat will continue to
control that variable, whatever it is, in the same way. The experimenter
can then determine what that variable is in the usual way. I don't see
the problem.
Maybe the problem is only in my mind, but I'll try to restate it, nevertheless.
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. Even so, the experimenter wants
to know which perceptions are being controlled in support of the stick-move
perception, and disturbs various candidate CEVs. 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.
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".
-------------------
And now the new question (3) How can the experimenter determine
whether the cat is controlling any kind of perception of the stick,
or whether the success of the Test is a consequence of the
environmental constraints?This choice isn't clearly stated. If the cat is just blundering around
in the cage until it happens to hit the stick, the experimenter would
quickly conclude that it is still reorganizing. If the behavior settles
down to some stereotyped control process, the experimenter can look at
the connection between what the cat is controlling and the motions of
the stick, and gradually alter the situation until the cat resists what
the experimenter does. Again, I don't see where there's any difficulty.
Well, I thought you had seen the difficulty in your previous message in which
you introduced this problem. 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. The result will look the same
at all phases of the Test--except perhaps for the speed with which the
cat resists (or seems to resist) the disturbances.
You mistook my proposal to mean that the experimenter increased the
pressure of the stick against the cat. That was not the strategy; my
imagined experimenter moved the pivot of the stick between trials so
that if the cat insisted on putting itself in the same final position as
before, it would necessarily have to increase its pressure against the
stick in order to get there. If it did so, we could immediately conclude
that the cat was NOT controlling a perception of pressure against the
stick, and that it MIGHT be controlling for a specific final position.
Note that this experiment was done in such a way that control of the
door's opening would be successful in either case, so reorganization
would not be started up again.
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.
You then proposed a delay in the door opening, so as not to allow this
effect. But to delay the door opening is to sustain the error in the
"see onesself out of the box" control loop, and that also affects the
output of this higher ECU, and may lead to reorganization during the
delay period. Either way, it would only be by chance that the experimenter
induced disturbances would be resisted without a change in reference level
that would affect the observations.
This goes back to very early discussions in which you assumed applying a
sudden disturbance that caused the controlled variable to change by a
large amount away from its reference level.
I'm afraid I don't see the connection. Perhaps there is one, but it is
lost on me.
I don't think it applies to the following, which was an attempt, possibly
flawed, to get away from the problem that the singular nature of the
environmental feedback function can allow side-effects to mimic control.
Imagine, for example, that the cat could escape by walking onto a
particular floor switch, or by rubbing on the wall away from the
stick, as well as by moving the stick. The different means could
be designed so that frantic cats might be about as quick to hit one
as another, by chance. When one method is used to escape, change
it slightly, so that the same stereotyped actions will not work.
See if the cat uses the same method anyway, because if it escaped
by side-effect, the other methods should be as likely to be used.
Keep doing this, disturbing whatever escape method was used last
time.Your method uses disturbances that prevent escape from the cage, and
requires reorganization to find another method.
I don't think that's relevant. See below.
The first time the cat
is in the cage after you have made such a change, it will control the
same variable as before. It will continue to do this until failure to
get the door open leads to reorganization; then it will find some other
perception to control that will open the door, and settle into another
stereotyped behavior. 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. The Test
will fail, properly, in that the cat won't seem to resist the disturbance.
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
isturbances, 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.
How do you know that what you think you have found out about the
way the universe works is right, and that you haven't simply found
something that correlates with an environmental constraint?You don't. This is because the actual environmental constraints _never_
become visible.
Right. That's the point about the cat, science, and religion. It's the
answer to Question (1)--or at least I think it is. And it is the
central issue in your recently discovered series of letters to and by
Isaac Newton.
Martin