# 4-D control; Milgram Expt

[From Bill Powers (931123.0830 MST)]

Tom Bourbon (931122.1711) --

I thought (perhaps incorrectly) that Bill's point in the
demonstration would be that the person can *control* his or
her perceptions by *manipulating* (using, not controlling)
dimensions that are unperceived, perhaps unperceivable. ... Did

No, you read me correctly. What the person will control is at
best a 3-D projection of the 4-D object. Any component of a
disturbance in the unseeable dimension will not be resistible or
resisted except by accident. Any effect of action in the 4th
direction will be unperceived unless there is an interaction with
one of the 3 "real" axes.

I wondered, in a science-fiction mood, whether I might not
accidentally learn to perceive in an extra dimension. If that
happened, I could control in the extra dimension, and might
inadvertently twist myself out of this continuum. But I hope not;
I'm not quite finished here.

A simpler demonstration would be to give the environment some
complex invisible linkages between the handle and the effect on
the cursor. These linkages could change during the experiment,
but in a way that leaves the net effect of the handle on the
cursor the same. Obviously, the person could control the cursor
but would have no awareness of the nature of the connecting
linkages, or of changes in them. Here's an example:

-----x A^2---->
/ \
handle ---- + -----> cursor
\ /
-----x B^2---->

Now the cursor position is h*A^2 + h*B^2, mediated by two
different environmental paths. If B and A vary as a function of
theta, where A = sin(theta) and B = cosine(theta), A^2 + B^2 =
constant, and the actor will be able to control the cursor as if
the actual environmental situation were simply cursor = constant
* handle. There will be no indication that A and B are varying or
that the handle is affecting the cursor through two paths. The
variable "theta" and its effects will be invisible.

The same point can be made using any environment that contains
more degrees of freedom than the effector's outputs or the
sensor's inputs have.

This is a sort of Goedel's theorem of PCT: There is no way to
determine the "objective" correlate of a controlled perception.
We human observers/actors can only construct a plausible
consensus as to what is being controlled, in terms of human
perceptions of types we all share.

ยทยทยท

--------------------------------------------------------------
Chuck Tucker (931123/2) --

While the Milgram experiment is important in revealing something
about human behavior, what it reveals is somewhat obscure. It
would take a long period of psythotherapy, I should think, to
explore the goal-structure that permitted or encouraged this sort
of behavior to happen in the subjects who went along with it.

If the only conclusion is that people can be persuaded to do some
pretty awful things, nothing much new has been found. Why do we
think they are awful? Did the subjects really think they are
awful? And if the subjects did think they were awful, what other
goal was there that overrode the goal of not doing something
awful? Were there subjects who enjoyed their role in this
experiment and enjoyed the apparent suffering of the subject, and
therefore experienced no real conflict? Was there an element of
enjoyment in ALL the subjects? Did all the subjects believe in
the reality of the experiment?

Were any questions like these explored, and if so, what was
found?
--------------------------------------------------------------
Best to all,

Bill P.

From Tom Bourbon [931123.1307]

[From Bill Powers (931123.0830 MST)]

Tom Bourbon (931122.1711) --

I thought (perhaps incorrectly) that Bill's point in the
demonstration would be that the person can *control* his or
her perceptions by *manipulating* (using, not controlling)
dimensions that are unperceived, perhaps unperceivable. ... Did

No, you read me correctly. What the person will control is at
best a 3-D projection of the 4-D object. Any component of a
disturbance in the unseeable dimension will not be resistible or
resisted except by accident. Any effect of action in the 4th
direction will be unperceived unless there is an interaction with
one of the 3 "real" axes.

Glad to know I was reading you correctly. Actually, I didn't see any way
you could have been implying that the person would control the fourth
dimension -- at least not at the start.

I wondered, in a science-fiction mood, whether I might not
accidentally learn to perceive in an extra dimension. If that
happened, I could control in the extra dimension, and might
inadvertently twist myself out of this continuum. But I hope not;
I'm not quite finished here.

That would be the ultimate experience in stick wiggling!

A simpler demonstration would be to give the environment some
complex invisible linkages between the handle and the effect on
the cursor. These linkages could change during the experiment,
but in a way that leaves the net effect of the handle on the
cursor the same. Obviously, the person could control the cursor
but would have no awareness of the nature of the connecting
linkages, or of changes in them. Here's an example:

-----x A^2---->
/ \
handle ---- + -----> cursor
\ /
-----x B^2---->

Now the cursor position is h*A^2 + h*B^2, mediated by two
different environmental paths. If B and A vary as a function of
theta, where A = sin(theta) and B = cosine(theta), A^2 + B^2 =
constant, and the actor will be able to control the cursor as if
the actual environmental situation were simply cursor = constant
* handle. There will be no indication that A and B are varying or
that the handle is affecting the cursor through two paths. The
variable "theta" and its effects will be invisible.

The same point can be made using any environment that contains
more degrees of freedom than the effector's outputs or the
sensor's inputs have.

And I thought this was what you might have in mind, beyond the 4-D task.
Perhaps one reason I "caught your drift" is that I have been playing with a
somewhat similar line of control tasks and modeling. In my attempt, I have a
person or a PCT model perform a tracking task, acting on a control handle
that affects a cursor. In the simplest version, the handle affects the
cursor directly, along with a disturbance that can be a constant zero. In
other versions, the person's (or first model's) handle affects a cursor
controlled by another model and the handle of the second model affects the
cursor the person intends to control. The person or the first model do not
perceive the second model, or its cursor. The person or model of the person
must act to disturb the cursor controlled by the second model, which affects
the cursor that the person or the model of the person intends to control.
The person or first model achieves control by way of disturbing a variable
controlled by a model whose existence is unknown, and the handle of that
unknown model directly affects the cursor the person or first model intends
to control. (I probably repeat myself, but I hope at least a few people
resading this will have some idea of what I mean.)

During various runs, I give the second model different reference signals,
so it controls different relationships between its cursor and target. (The
first model has no perception of the target of the second.)

All the person (or the model of the person) perceives are the target and
cursor on the screen. No matter which connection exists between handle and
cursor, the person (or first model) controls just as well. I have not yet
made the selection of handle-cursor paths random, from one run to the next,
but when I do, that will not matter to the person or to the first model:
either will control just as well in any of the conditions. That would also
be the case were the person's handle to disturb a variable that is part of
an interaction among n models whose presence was not known to the person.

This line of work came about as a result of assertions, during the
discussions of "feedforward," that a person or a control model of a person
should contain a model of the environment, or that a person controls the
"real" CEV. Bill's next paragraph expresses my thoughts on that matter
better than anything else I might write:

This is a sort of Goedel's theorem of PCT: There is no way to
determine the "objective" correlate of a controlled perception.
We human observers/actors can only construct a plausible
consensus as to what is being controlled, in terms of human
perceptions of types we all share.

Bill, I am not quite playing with the fourth dimension, but I might be at
risk for vanishing into some pretty strange social arrangements with a bunch
of unseen PCT models! See you in the twilight zone.

Until later,

Tom