Hypothetical Controlled Variables

[From Oded Maler (970429)]

[From Rick Marken (970428.1555)]

Martin:

> The experimenter has to deduce the effect of the source variations
> on the supposed controlled environmental variable.

Me:

>Wrongo! The experimenter has to _observe_ the effect of these
>disturbances on the supposed controlled variable.

Martin:

> How on Earth is the experimenter supposed to _observe_ the effect
> of the disturbances on something that is a function of the
> _subject's_ perceptual input function?

Try my "Test for the Controlled Variable" demo and see how easy it
is. Remember, the experimenter can perceive the hypothetical controlled
variable too -- though not necessarily as the subject perceives it. An
experimeter can perceive the echo-pattern controlled by a bat even
though she can't perceive it as bat does.

> The experimenter can _observe_ only the effect on his/her OWN
> perceptions. This is an EXTREMELY crucial point to understand, Rick.

Yes, indeed. Do you understand it?

> But I think you should ask Bill whether he said that the
> experimenter can or can't see the influence on the subject's
> controlled environmental variable. The fact is that the
> experimenter can't, whatever Bill said.

Take the Three Squares (Test for the Controlled Variable) demo and call
me in the morning.

I completely agree with Martin. What do you really mean by:

"the experimenter can perceive the hypothetical controlled variable
too -- though not necessarily as the subject perceives it" ?

Either you assume that there is some Platonic object P such that
perception P1 of the subject (as a function the subjects' low-level
perceptions) and the perception P2 of the experimenter (as a function
the experimenters' low-level perception) are both approximations of,
or you are a true PCTer and know that "it's all perception". How can
the experimenter discover that the subject controlled perception is
whether the sum of squares of the pixles coordinates is divisible by
17, or maybe whether the line between its elbow and the mouse is
pointing toward Mecca? Or whether the voltage of neuron B7689X plots a
curve of certain form? While such hypothetical perceptions might seem
absurd in the context of controlling for "simple" "self-evident"
geometrical forms, I believe this is the case whenever you try to work
with non-trivial perceptions.

This is not to say that the test for the controlled variable is a stupid
idea, on the contrary. But if you admit its *inherent* limitations, and
stop pretending that it can *ever* be equivalent to experiments in
Newtonian physics, it will make your arguments more credible. Applying the same
level of criticism to one's favorite methodology and to the "Other"'s
(S-R mathematicians, that is :wink: is a good evidence for controlling for
Truth herself and not just some personal perception.

Truely,

--Oded

[From Bill Powers (970429.0330 MST)]

I guess that decaffeinated tea wasn't so decaf.

Oded Maler (970429)--
Rick:

Take the Three Squares (Test for the Controlled Variable) demo and call
me in the morning.

I completely agree with Martin. What do you really mean by:

"the experimenter can perceive the hypothetical controlled variable
too -- though not necessarily as the subject perceives it" ?

Either you assume that there is some Platonic object P such that
perception P1 of the subject (as a function the subjects' low-level
perceptions) and the perception P2 of the experimenter (as a function
the experimenters' low-level perception) are both approximations of,
or you are a true PCTer and know that "it's all perception".

You don't understand the Test. Let me elaborate.

If you hypothesize that some function of the visible environmental variables
is under control (or simply one of the variables), and you can "perceive" it
-- i.e., with your natural equipment or by calculation -- then you can use
the Test on it. The Test will rule it out if it is NOT under control. That
is what the Test is for: to rule out wrong hypotheses about what is controlled.

If the Test is passed, this means that you have done the following steps:

1. Show that the hypothesized perception resists all disturbances within
some range, when you know that the disturbances alone (with no other system
acting) would change the hypothesized perception.

2. Trace the reason for the resistance to an action by some other system,
and identify the physical connections.

3. Verify that if the ability of that same system to sense the pertinent
parts of the environment is cut off, control is lost or observably worsened
(the disturbance becomes effective again).

All of these steps are necessary to conclude that the hypothesized
controlled perception is not ruled out. The first step, which is all that
most people mention when discussing the Test, is not sufficient by itself.

So now we have a proposed controlled perception, and a control system, that
has passed the Test. Does this mean that we have proven that this perception
is in fact controlled? No, we have not.

In fact, some other function of the environmental variables might actually
be under control by the system we are investigating. All that is required
for a proposed perception to pass the Test is that it be closely related to
the actual controlled perception. For example, in the Coin Game a person
might be maintaining the pattern of an N in any orientation (a coin at each
vertex or line end). If I hypothesize that the person is controlling for the
perception of a Z in any orientation, this hypothesis will pass all phases
of the Test. The reason it will pass is that an N is operationally identical
to a Z, in terms of all three phases of the Test. The only difference is in
how you imagine lines connecting the coins. The figure could also be
perceived as an X or a rectangle.

However, this does not mean that all closely-related perceptions are
indistinguishable under the Test. Suppose I hypothesize that you are
controlling the position of one end of a simple lever, but in fact you are
controlling the position of the other end. The first phase of the Test with
disturbances will be passed by either perception, the second phase of
tracing the cause of the resistance to your actions will be passed, but the
third phase will be failed: If I block your view of one end of the lever,
and that end is not the end under control, control will be NOT be lost (or
observably worsened). I would have to block your view of the end of the
lever that you're watching in order to cause control to be lost. Blocking
your view of the other end would have no effect. Since it had no effect, I
could leave the view of that end blocked, and block the view of the other
end, too: then control would be lost and the Test would be passed.

This is specifically a test for a PCT-style controller. Note that the third
phase of the test would rule out a PCT-style controller if the controller
were really an MCT controller. The MCT controller would apparently continue
to control after the perception was blocked. One would have to keep it
blocked for a long time to see that eventually control would be lost. Since
an MCT controller that relies primarily on real-time perception IS a PCT
controller, the Test would remain one for PCT control.

The Test isn't just a game with fixed rules. It's a logical procedure for
identifying a PCT control system. The experimenter can change hypotheses and
introduce any new conditions needed to refine the Test, whenever that will
help to narrow down exactly what is controlled. In the case of the coin
game, the experimenter (if experienced) would probably realize that the same
pattern of four coins could be a Z, and N, a parallelogram, or an arrow, and
will use disturbances that can eliminate as many hypotheses as possible. If
the choices boil down to Z or N, the experimenter can ask the controller
which is being controlled, a Z or an N. Why not? That, too, provides
information about a perception, although one would prefer objective evidence.

When every perception you can think of as a possible controlled variable has
been ruled out, except one, the one that is left is your best guess as to
what is being controlled. If you are left with more than one, then you must
refine the definitions and apply disturbances more cleverly until you manage
to rule out all but one. If you can't eliminate all but one, then you have
to say that as far as you can see, the remaining definitions are
operationally equivalent. Then, if the controller can talk, you can ask
which one is really being controlled. By that time you will probably know
more about what is actually being controlled than the conscious controller
does, so in fact the answer could be wrong!

Rick's "squares" demo works because there is only one action available, and
all its effects on objects on the screen are known (in fact they are the
same for all three of the squares). Likewise, there is nothing to be
controlled but the three squares seen on the screen. So steps 2 and 3 of the
Test can reasonably be bypassed. All that is left, then, is step 1:
determining which of the squares is most resistant to disturbances. Since
the reference level is varying (the person can produce any desired movements
of the selected square), the identification has to be made statistically.
The squares showing the highest correlations of position with the
disturbances (which are different for the three squares) are ruled out,
leaving the square showing the lowest correlation as the controlled
variable. Alternatively, the program could look for the highest negative
correlation between a disturbance and the position of the mouse. This, being
a statistical test, is not infallible. The program makes mistakes, but not
very often. If the person moved the mouse without watching the screen, or
didn't actually try to move a square in a particular pattern, the program
could be fooled: it is set up to pick the most probable controlled variable,
and there must always be one even if no square is being controlled. But the
program works sufficiently well to be impressive as an example of "mind
reading."

Of course, this being science, nothing is completely certain. However, when
all three phases of the Test are applied carefully, and all disturbances are
explored, and the proposed perception passes the test, one must resort to
guessing about highly improbable alternatives even to suggest that the
proposal is wrong -- and those suggestions, of course, could also be
submitted to the Test.

Best,

Bill P.

How can

ยทยทยท

the experimenter discover that the subject controlled perception is
whether the sum of squares of the pixles coordinates is divisible by
17, or maybe whether the line between its elbow and the mouse is
pointing toward Mecca? Or whether the voltage of neuron B7689X plots a
curve of certain form? While such hypothetical perceptions might seem
absurd in the context of controlling for "simple" "self-evident"
geometrical forms, I believe this is the case whenever you try to work
with non-trivial perceptions.

This is not to say that the test for the controlled variable is a stupid
idea, on the contrary. But if you admit its *inherent* limitations, and
stop pretending that it can *ever* be equivalent to experiments in
Newtonian physics, it will make your arguments more credible. Applying the same
level of criticism to one's favorite methodology and to the "Other"'s
(S-R mathematicians, that is :wink: is a good evidence for controlling for
Truth herself and not just some personal perception.

Truely,

--Oded