[Frfom Bill Powers (2004.07.07.2140 MDT)]
Bruce Nevin (2004.07.07 21:22 EDT) –
It is the observer’s perception of
an aspect of the environment "at the
input" of the observed controller. Since the CV is often distant
from the
sensors of the observed controller, “at the input” must mean
something like
“as though from the point of view of the observed controller”?
I’m not
going to take the phrase too seriously, I just want to flag it as a
bit
puzzling.
Let me try to make it clearer with an example. In a tracking experiment,
the experimenter-observer sees a computer display with a person seated
before it, looking at the screen. The controlled variable as the observer
sees it is the separation of a red line (the target) and a green line
(the cursor) on the screen of the display, which is in the environment of
the person doing the controlling. That is how the observer would describe
what he sees. From our omniscient theoretical point of view, we claim
that this whole scene is a perception in the brain of the observer, but
he experiences it as something outside himself.
The CV is shown to resist
disturbances. Is shown to whom?
To the observer. The observer knows, and perhaps has himself arranged,
that there is a disturbance being applied to the target, moving it on the
screen. This disturbs the controlled variable, and the person doing the
controlling can resist that disturbance by moving the cursor, using a
mouse. The observer knows how much the target-cursor distance would be
made to change by the disturbance if the cursor did not move – if the
controlling person did nothing. He observes that the cursor follows the
target so this predicted effect does not occur.
Enter here the essentially
and necessarily social nature of science, in the person of explicit or
implicit co-observer(s), and the issue of shared
perceptions.
Perceptions can perhaps be duplicated but they can’t be shared. Each
person experiences a perceptual signal inside his own brain, which is not
the perceptual signal in anyone else’s brain.
Co-observers make their own observations. The problem is then one of
using communication to establish their agreement about what they
observed. From the point of view of any one of them, the others are black
boxes whose internal experiences have to be deduced or inferred. If the
observations involve low levels of perception it seems to be easier to
reach agreement than when higher-level variables are involved (as this
discussion illustrates so clearly).
The disturbances may be due to the
observer’s actions or may be
naturalistically observed.
“But only if the CV can be sensed by the controller.”
It is the observer who is proposing a definition of the CV and testing
it. The whole idea is to find out if the proposed CV is consistent with
the idea that the controller is controlling it…
Since we are being
careful here we must avoid equivocation. The CV was first defined as
a
perception in the brain of the observer, a perception of an aspect of
the
environment. We cannot say that the CV can be sensed by the
observed
controller if the CV is a perception in the brain of the
observer.
The observer makes a distinction between the experience of the CV in the
environment (which is, whether he knows it or not, a perception in his
brain) and his experience of himself as an observer. He sees the computer
screen and its display over there, and feels himself as viewing it from a
distance.
So this
should be "but only if the same aspect of the environment that the
observer
perceives as the CV can be sensed as well by the
controller."
No, that is what is being investigated. To test whether the controller is
also experiencing this variable, the experimenter must interrupt what he
believes to be the path by which the controller perceives the CV. He can
blindfold the controller or put up a shield between the controller’s eyes
and the screen. If his hypothesis is correct, this should cause loss of
control. If control continues, he must modify his definition of the CV
and against test to see if it is really being sensed by the controller.
Similarly, the experimenter must check that what he sees as the action of
the controller, the hand moving the mouse, is really what is causing the
cursor to move. If he removes the mouse and the cursor continues to track
the target, again he must look for an alternative definition – or for
some other system that is doing the controlling.
How can we know what can be sensed
by the observed controller? Well, in
practice the usual way is to disturb what we perceive as an
environmental
variable and perceive resistance to that disturbance which we perceive
as
being due to their control actions.
That only tests to see if there is apparent opposition to the
disturbance. We must also check the physical pathways by which the
proposed CV is sensed and affected, and experiment with them to make sure
we have that right (as near as we can determine from outside the other
person). All this has to be done in terms of the experimenter’s own
understanding of the physical relationships in the environment, and his
or her own concept of how the controller senses things in the environment
and acts upon them.
If they’re controlling it,
they must be able to perceive it. But that is circular, and only with
care and persistence can we “show” that it’s not a side effect
that they’re resisting, and that it’s not a different aspect of the
environment that they’re perceiving.
No, we test as directly as possible whether the variable we have defined
as the CV is in fact being detected by the other person. That has to be
established independently of the observation of resistance to
disturbance. The resistance alone does not prove that the experimenter’s
definition of the CV is what the controller is controlling. It just shows
that some system, somewhere, is somehow controlling something related to
the CV sufficiently to make it appear to be under control. The test is
not complete until the rest of the hypothesis is submitted to
scrutiny.
Other than that, it must be
“shown” that it is at least possible for the
controller to perceive the CV.
No, it’s the other way around. The experimenter must show – demonstrate
– that the controller is actually perceiving the CV, by doing something
that is guaranteed to prevent that perception and showing that control is
lost. If you have played the coin game, you will remember how easy it is
to postulate a CV, show that it is resistant to disturbance, and be
wrong.
How is this shown without
“showing” that the
controller actually does perceive the CV, or has done at some
time?
It’s not done that way. One must demonstrate that interrupting the
ability to perceive the environment where proposed CV is prevents control
of it by that controller. That’s part of the Test.
Perhaps it is only a
statement of plausibility because (the observer perceives that) the
observed controller has sense organs that someone sometime has
“shown” to take this sort of environmental input (as perceived
by that other observer) at one end and “responds” with a neural
signal at the other end.
No,that’s not good enough. Actual demonstration is required,
independently of whether the CV is resistant to disturbance. You could
blindfold the controller and say “Please move the cursor to the edge
of the screen.” If the cursor moves to the edge of the screen, that
controller is not controlling it. You could say, “Tell me where the
target is.” If the person says “What target?” you have
shown that your definition of the CV is wrong – or your idea of who or
what is controlling it is wrong.
Maybe it’s written down in a
book on anatomy and physiology. On the
other hand, perhaps the statement of plausibility is based on no more
than
homomorphism and the assumption that others construct perceptions
from
input to e.g. their eyes in the same way as oneself. None of this
is
possible without assuming or “showing” shared perceptions at
various
junctures-- shared with the investigator whose work is reported in
the
book, or with the person or organism that I perceive as having sense
organs
homomorphic with what I perceive to be my own sense organs.
(Possibly
supporting that assumption on the strength of books on anatomy and
physiology.)
Nope, that’s not the idea at all. Plausibility is insufficient. What the
experimenter wants is to be backed into a corner from which the only
escape is to admit that this definition of the CV is the ONLY one that
remains despite all attempts to prove otherwise.
This
means it is strictly a phenomenon in the perceptions
of the experimenter-observer-analyst who is observing the control
system
from outside and trying to guess what it is
controlling.
Except for all this “showing” and the knowledge or assumption
or perception
that the CV (a perception in the brain of the observer") can be
sensed by
the controller.
As I keep saying, this is not an assumption or a mere
“perception” in the sense you use. It is an inescapable
demonstration of a fact. As long as grounds for doubting the
demonstration can be found, the Test is not completed.
Let’s
approach this through a familiar example.
The experimenter (for short) sees what looks like a cursor and a
target
moving up and down on a computer screen. After observing tracking
behavior
Wait a minute. All I see is what looks like a cursor and a target moving
up
and down on a computer screen. That doesn’t constitute tracking
behavior
unless I perceive or impute that a controller is controlling perceptions
of
what I perceive as a cursor and a target moving up and down on a
computer
screen, and that this controller is capable of perceiving the
environmental
variables that I perceive as a cursor and mark moving up and down on
a
computer screen.
No, that’s jumping way ahead of what you can “show”. First you
have to show that the CV – in this case, the distance between the cursor
and the target – is being controlled by something. You do that by
applying a disturbance that should change the distance between target and
cursor, and observing that something makes the cursor move so as to
prevent that distance from being affected. You use random, unpredictable
patterns of disturbance; you use means of application that are known only
to you and are invisible to anyone else. You make sure that there is
nothing present that can move the cursor without the mouse moving. And
you observe that the cursor never moves unless the mouse does (or else
that the cursor movements correspond to known disturbances only, if
you’re using a second disturbance). In short you use the best of your
knowledge to show that the ONLY way in which the observed behavior could
take place is for some controller to exist somewhere, detecting the CV as
you have defined it and acting by using the only means there is for
moving the cursor. None of this requires knowing anything about how
controllers work inside.
To propose a process entirely going
on in the perceptions in the brain of
the experimenter-observer-analyst you have to start with
naturalistic
observation. But even there you narrow it down with knowledge and
assumptions that are essentially social in origin. No
experimenter-observer-analyst can possibly start with a tabula rasa
brain.
Nor is it necessary to assume that. You are at much too early a point in
your argument to be asserting that knowledge and assumptions are
essentially social in origin. That’s a conclusion you have to prove; you
can’t use it as a premise.
We are talking about a theory here, not “naturalistic
observations.” We do not observe naturalistically that there are
perceptions inside anyone’s head, our own or other people’s. What we
observe is the world as it appears to us, including the outside
appearance of other people.
Aside from the simple fact that
there is no such thing, there are just too
damned many variables. You have to narrow it down. Which is of course
why
you contract with people to do simplified tracking experiments with
a
cursor and a target mark moving around on a computer
screen.
Tell me about it. My models are currently handling 500 variables at once.
Admittedly, this line of work is at an early stage, but
The
experimenter finds out, or himself arranges, that a disturbance of
known behavior is moving the target.
“A disturbance of known behavior”: a perception (referred to by
the word
“known”) about a disturbance, that is, about a perception of a
relationship
between the CV-perception and a perception of another aspect of the
environment. The relationship is such that it “should” change
the CV in the
“known” way.
Yes, and what is the basis of this knowledge and expectation? It’s the
whole discipline of physics, so to raise doubts about this knowledge you
would have to supply a workable alternative to the world-view of physics.
I’m not talking about hunches and guesses here, but about the application
of the most reliable of human models which fail so seldom that they are
held up as the standard of knowledge, of what it means to know something.
Having
shown the likelihood that the proposed CV or something closely
related to it is being controlled, the experimenter then satisfies the
two
auxiliary requirements that have to be met. First, simply from the
physical
setup, it is clear that the cursor moves only because the controller’s
hand
moves.
You’re measuring the movement of the mouse or joystick, but your
perception
of the physical setup is that this cannot move unless the controller’s
hand
moves. Would that in situations of more interest to us it were so easy
to
rule out extraneous variables!
But it is not hard to rule out hypotheses if you really try. The problem
is that people are more interested in proving that they are right than in
thinking up challenges to what they know.
While
there might be a second disturbance between the hand and the
cursor, the experimenter would know how this second disturbance is
changing
and could establish that the cursor position is what it should be at
all
times, given the hand position and the disturbance magnitude at each
moment.
You’re talking about a second disturbance introduced and measured by
the
observer? The observer would not know a disturbance such as a hand
tremor.
Yes, a second pattern of disturbance added to the measure of handle
position to determine the cursor position. This is often done in my
tracking experiments; it’s a way of showing that what is learned is not a
particular hand movement. Hand tremors do appear as small variations in
cursor position that can’t be accounted for by the model, but they are no
more than 4 or 5% of the range of the movements, at most.
Even if you measure hand movement
in addition to mouse/handle movement, how
could you distinguish which hand movements were intended control
actions
and which were unintended side effects of muscle tension, fatigue,
or
whatever causes the tremor?
The only measure of hand movement is mouse movement. But we can account
for all but a few percent of the hand movements, so that’s not a
problem.
The
other condition is that for control to continue, both the cursor
and
the target must be visible to the controller, under the hypothesis
that
there is control of a visual perception of the distance between target
and
cursor. A simple experiment shows that that control is lost if the
visual
pathway is interrupted for any reason from sudden blindness to blanking
or
covering of portions of the screen.
This BTW demonstrates that the controller actually perceiving the aspect
of
the environment that the observer perceives as the CV, and a fortiori
that
the controller is capable of perceiving it.
No, it doesn’t. It only shows that the controller is perceiving some
aspect of the environment that is being sensed visually. It doesn’t
establish that the particular CV as defined is being perceived. We have
no direct way to measure what a person is perceiving. That has to be done
by using varied disturbances and measuring the effects on the controlled
variable. This test is failed if control continues with the sensory
pathway is interrupted…
Just imagine how a control system works. All the Test does is to see if
all the conditions represented by the diagram (that are observable from
outside the system) are actually present. It’s not a ritual that we go
through for form’s sake. It’s mainly common sense.
Note
that so far nothing at all has been said about the perceptual input
function of the controller.
You have “shown” earlier that the controller can perceive the
aspect of the
environment that you perceive as the CV.
No, I have shown only that the CV as I defined it must affect the senses
of the controller, and that when it is prevented from doing so, control
is lost. The Test never proves that the controller perceives and controls
the same CV I perceive being controlled. It provides a reasonable basis
for proposing a model of what is inside the controller. There are always
alternative models that would produce the same observed effects; an
infinity of them. The best we can do is propose the simplest model that
will do the job, and say that whatever is going on inside, it
accomplishes what this model accomplishes.
The
Test is conducted strictly in terms of
variables and situations observable by the experimenter using his
own
senses alone.
There’s a lot of dependence on the senses of others.
There is none whatsoever. It is impossible for one person to depend on
the senses of others. At best, we can depend to some extent on what other
people say they sense. That is not the same thing.
I can see that you are very motivated to make your point, since there
appears to be no end to this argument. I think we have stalled.
Best,
Bill P.