[From Bill Powers (920911.0900)]
A goody to pass on from a murder mystery. A commercial announcer is
warming up and doing a sound level check. Over the speaker system in
the studio, his important voice says "Yes, friends, try Preparation H
and kiss your hemorrhoids goodbye!"
Greg Williams (920911) --
Any single manipulative act of any kind
could be judged in all four of these ways by different people.
Yes, but I think judgement by the MANIPULEE (sometime after a
successful manipulation, or after being apprised of all relevant
facts >about a failed manipulation) is most telling. PCTers like to
take the >point of view of the organism, don't they, for "true"
explanation of >behavior?
That's what I meant -- "by different manipulees."
For short, this is "manipulation of action
through disturbance of uncontrolled perceptions" or just
"manipulation through uncontrolled perceptions" (OK?).
I think so. You've spooked me about agreeing to anything with
"control" >in it because I still don't know exactly how you mean it
all the time. >Manipulation IS control in the sense that the
manipulator is >controlling some of his/her perceptions, but I'm not
so sure any more >whether saying it is "control of something else"
makes anything less >muddy.
I always mean control in only one way: the controller acts on the
environment to make a perception in the controller match a reference
level determined by the controller.
A manipulator works in two stages (maybe this will make it clearer).
One stage is direct control of perception of an environmental
variable. This is, shall we say, the _instrumental_ variable -- it
will be instrumental in effecting manipulation, if manipulation
succeeds. The manipulator does this in order to affect a second
variable, which is the aspect of the action of the manipulee that the
manipulator perceives, hopes to affect, and desires to control.
In the case of control by disturbance of perceptions, the manipulator
controls (which is invariably short for "controls a perception of...
") the instrumental variable as a means of controlling the action of
the manipulee. As long as the manipulee maintains a constant reference
level and succeeds in keeping the controlled variable at that level,
the manipulator will have true control over the manipulee's action,
and the manipulee will have NO control over that action (or the lower-
level reference signal that specifies the action). The action is
ENTIRELY under the control of the manipulator, not the manipulee -- as
long as the manipulee is using this action to stabilize the
manipulee's controlled variable. This is only to say that disturbances
must be opposed by the action if control is to continue, which is true
for all parties concerned.
Mostly for the sake of onlookers, I'll make this concrete by using the
Rubber Band Game. The following analysis gets a little complex in
spots, so it may be a good idea to demonstrate each point with actual
rubber bands and players.
In the rubber band game, the instrumental variable is the position of
the manipulator's end of a pair of linked rubber bands. The
manipulee's controlled variable is the position of the knot. The
manipulee's action that the manipulator perceives, affects, and hopes
to control is the position of the manipulee's hand at the other end of
the linked rubber bands. The manipulator controls the instrumental
variable, the position of his/her own end of the rubber bands.
Changing the reference position of the instrumental variable, and thus
changing the actual position (because control is involved) tends to
disturb the knot, so the manipulee moves the hand holding the other
end of the rubber bands to counteract that effect and keep the knot
over the selected spot on the table.
Thus the manipulator can affect the manipulee's hand by varying the
intended (and actual) position of the instrumental variable. By
watching the manipulee's hand, the manipulator can thus CONTROL the
position of the manipulee's hand, placing it anywhere feasible that
does not cause injury, cause the hand to dip into the soup, or create
other problems for the manipulee.
If the manipulee changes the reference position of the knot and moves
it by moving the hand position, the manipulator can compensate by
altering the instrumental variable, and return the manipulee's hand to
the original position. If a third party connects another rubber band
to the knot position, and pulls on it, the manipulee will move his/her
hand to counteract the disturbance of the knot. But the manipulator,
seeing the movement of the hand, will change the instrumental variable
so as to return the manipulee's hand to the original position intended
by the manipulator.
From the manipulee's point of view, it seems that resisting a
disturbance of the position of the knot now requires far less effort
than it does when manipulator is not manipulating. Just beginning to
move to counteract a disturbance seems to be enough to eliminate it --
in fact, the hand scarcely moves at all, and returns immediately to
its original position. When the manipulee intends the knot to be in a
new position, the mere thought, and just a suggestion of a movement of
the hand, seems enough to get the knot to the new position. Again, the
hand ends up where it began.
This uncanny ease of control is actually thanks to the manipulator,
who is doing most of the work.
The manipulator, it should be pointed out, cannot freely choose the
reference level for the instrumental variable. As long as the
manipulator's goal is to keep the manipulee's hand in a specific
position, the requirement for doing this is that the instrumental
variable be in the position physically required to produce this end.
Thus the manipulator's action (positioning the instrumental variable)
must depend on the manipulee's choice of reference position for the
knot, and on the amount and direction of any third-party disturbances
of the knot. So in the final analysis, the manipulator's hand position
(the instrumental variable) is not under the control of the
manipulator, and the manipulee's hand position is not under the
control of the manipulee.
The manipulee can control the manipulator's hand position by varying
the reference position of the knot; the manipulator can control the
manipulee's hand position by varying the reference position of the
hand. Each one controls the other's means of achieving the same
other's goal.
ยทยทยท
----------------------------------------------------------------------
The King has arrived in the Queen's boudoir. The Captain of the Guards
has hastily concealed himself behind the window drapes.
"My dear," says the King, "you seem a little flushed. Perhaps I should
open the window..."
The Queen leaps to her feet, rushes to the window, reaches between the
drapes, and opens it wide. "You're right, dear Highness. I feel much
better now that the breeze is coming in."
The King says "You seem troubled. I cannot see your face in this
light. Perhaps if I just pull the drapes back..."
The Queen hastily turns up the dressing-table lamp. "No, Highness, I
am quite untroubled; my brow is smooth as you can see."
There is a clink and a scuffling sound. The Queen relaxes. "Of course,
Highness, if you will that the drapes be open, so they shall be." And
she flings back the drapes, revealing to all but the King that the
Captain has escaped through the window.
-------------------------------------------------------------------
Just yesterday morning, Pat and I were discussing the possibility
that >"rubber-banding" is just a special case of the broader class of
manipulation which I've been trying to convince you is ubiquitous
(and >not necessarily "statistical"/"unimportant"/always "bad"), lo,
these >many days.
If we can enumerate all the methods of manipulation that exist, we
will see what broader classes there are. I believe that the underlying
mechanisms are far simpler than their numerous manifestations.
----------------------------------------------------------------------
-
There is another type of manipulation that you've brought up
previously. We could call it "manipulation by altering the properties
of the environment." This is what Skinner referred to as altering a
"schedule of reinforcement."
Here you are being too specific. Skinner would have said that ANY
(actual) manipulation of ANY kind would HAVE to be done by altering
the >properties of the other's environment.
By altering "properties" of the environment I do not mean altering
instrumental variables in the environment. I mean altering
relationships among the variables, and thus altering the effect that
any action by the manipulee will have on the manipulee's controlled
perception.
This kind of manipulation can be done without the act of manipulation
itself disturbing the manipulee's controlled variable at all. Skinner
could, by resetting the apparatus, change the number of bar-presses
needed to get each reward. Doing so would not, in itself, immediately
alter the rate at which rewards are obtained. But it would alter the
effects of pressing the bar at any given rate. If the required number
of presses per reward were decreased, the rat would find that the next
reward appeared sooner. The hourly or daily rate of obtaining food
would rise if the rat continued pressing at the old rate, but the rat
would soon lower its pressing rate, bringing the obtained amount of
food per unit time back toward the reference level. All this assumes
that the schedule is not in the range of extreme deprivation, but in
the range where normal control relationships hold: more food, less
effort.
So by changing the schedule, it is possible to affect the rat's
behavior in a known direction. The effect of the change in the
properties of the apparatus is made visible only through the effect of
the rat's actions on the outcome, the rate at which food is delivered.
There is no direct disturbance of the reinforcement rate or the
behavior rate. The only effect on reinforcement rate is that of the
rat's behavior. It is still true that zero behavior produces zero
reinforcement rate.
Skinner seemed to recognize ONLY this means of manipulating behavior.
I don't think he ever used direct disturbances of the reinforcement
rate.
A simpler example would be that of a person operating a lever to
control the position of a spot of light in relation to a moving
target. The experimenter can arrange for the ratio of spot movement to
lever movement to change, effecting the change just as the lever
crosses through the zero position. The change itself will cause no
deviation of the spot, but as the subject's action continues it will
be found that the spot now responds far more to the same amount of
lever movement. As a result, the participant will decrease the lever
movements, to keep the spot tracking the target.
This is clearly a different way of affecting behavior. It does not
involve applying a disturbing force either directly or indirectly to a
controlled variable. It entails altering THE FORM OF THE ENVIRONMENTAL
FEEDBACK FUNCTION.
We can call this "manipulation by altering properties of the
environment."
I call it a broader view of how ANY manipulation could be
accomplished.
The only thing this method shares with the method of disturbing
controlled perceptions is that it allows the manipulator to control
his/her own perception of the manipulee's actions. The actual method
is entirely different.
In the rubber band game, an equivalent method would be to substitute
two parallel rubber bands for the single one on the manipulee's end.
Now the manipulee would respond half as much to disturbances at the
manipulator's end. In operant conditioning, a second way of
introducing this kind of manipulation would be to substitute a new
lever that moves sideways rather than up and down: by this means you
cause the rat (eventually) to switch from a vertical to a horizontal
action.
You can make an inexperienced public speaker speak less loudly into
the public-address system by moving one of the loudpeakers closer to
the orator, or obtain exactly the wanted loudness by giving the orator
earphones with the volume-setting adjustable by you. You now control
the loudness of the orator's speech as perceived by you. The orator
continues to control the loudness of his/her own perception of the
same speech at the same reference level, but the orator no longer
controls the loudness as heard by others.
The environmental feedback function expresses the way the manipulee's
perception depends on the manipulee's actions. Such a function can be
expressed as a polynomial in ascending powers of the action variable,
or as a nonlinear differential equation. If a constant term is
included, then the most general means of purposefully manipulating the
action of another control system could be described as "control by
changing coefficients in the environmental feedback function." An
externally-adjustable constant term is, of course, a direct
disturbance of the controlled variable. So this last mode of
manipulation covers all cases so far.
The method of "giving new information" still remains as an
alternative.
OK?
---------------------------------------------------------------------
Eric Harnden (920911) --
3) (here it is...) structural loops cannot be built that would result
in simultaneously dependent equations.
I hope that you mean "linearly dependent." If Stella rules out cases
that amount to analog solution of simultanous equations, it's not
going to be much good for HPCT. It could be used to simulate a single
control system, but not two control systems controlling different
aspects of the same environment.
Suppose you have two "levels", x and y. You want one control system
that controls the perception s = 2x + 3y, and a second one that
simultaneously controls the perception t = 3x - 2y. It is possible for
reference signals s* and t* to be set and varied arbitrarily, so that,
independently, s = s* and t = t*. Rick Marken's spread-sheet program
does a similar thing with SIX ways of perceiving six environmental
variables, and at three hierarchical levels.
s = 3x + 2y; es = s* - s; x = x + gs*es; y = y + gs*es;
t = 2x - 3y; et = t* - t; x = x + gt*et; y = y - gt*et;
If you set gs and gt to small values, this system will converge to the
condition s = s* and t = t* for any settings of s* and t*. I think
that x and y can be set up as "levels" with the "flows" being gs*es
and gt*et. Could you try this with Stella and see if it works? This is
sort of crucial for any advanced uses of Stella.
----------------------------------------------------------------------
-
Gary Cziko (920910.1558) --
Looks as if Forrester did understand a lot about control theory.
George Richardson, author of _Feedback thought in social science_, is
a system dynamicist and a follower of Forrester's methods. George is
friendly to the CSG (he got an article of mine published in the System
Dynamics Review) but he has never joined. I'm pretty sure that
Forrester never saw the whole picture; his concept of "levels" and
"flows" is only a special case of possible relationships that will
work in closed loops. But we, too, have found that integral output
functions are pretty universally useful. He was mostly concerned with
simulating social systems, I think.
----------------------------------------------------------------------
-
Best to all,
Bill P.