More Control of Behavior

[R_Marken]

[From Rick Marken (980911.1220)]

For those of you with Vensim, here is another version of
Bruce Abbott's "Control of Behavior" model. The model
shows a CONTROLLER controlling a CONTROLLEE's output by
disturbing a variable (called "diagonal") that is controlled
by the CONTROLLEE. What is interesting (to me) about the
model is that the CONTROLLER is exerting control by disturbing
a dimension of the controlled variable that is not being
affected by the output (called "output height" in the model)
that is being controlled.

The controlled variable is the diagonal length of a right
triangle, as shown below:

height *
= * *
output * * < --- diagonal = qi
height (qo of * *
CONTROLLEE) + * *
disturbance * *
height (d) * *
                * *

···

************************
                        width = qo of CONTROLLER

The CONTROLLER's output ("o2" in the model) affects only the width
of the triangle; the CONTROLLEE's output ("output height") affects
only the height of the triangle. Moreover, the height of the
triangle is also affected by a random disturbance. Nevertheless,
the CONTROLLER is able to disturb the width of the triangle as the
means of keeping the CONTROLLEE's effect on the height of the
disturbance ("output height", which is the CONTROLLER's CV) under
control.

As you can see in the graphs of the CONTROLLER's and CONTROLLEE's
behavior, control is not perfect in this case; the trace of the
controlled variable for both CONTROLLER and CONTROLLEE is offset
somewhat from the trace of the reference specification. This
relatively poor control results from the fact that I had to reduce
the gain for both CONTROLLER and CONTROLLEE in order to stabilize
the behavior of the interacting control systems. I have still
been unable to remove an initial transient oscillation from the
behavior of both systems; I don't know if it's possible to remove
it using just a simple control model (it might take a couple
levels of simple control systems); those who are interested
can experiment with different combinations of gain and slowing
for each control system.

The main things to understand about this model, however, are
1) the CONTROLLEE is controlling a variable (diagonal = qi) that
is a function of _two_ independent environmental variables and
2) the CONTROLLER is able to control a one-dimensional variable
(the CONTROLLEE's output -- "output height") by disturbing an
aspect of the variable the CONTROLLEE is controlling -- even
though it is not and aspect of the controlled variable that is
directly affected by the CONTROLLER's disturbance.

The CONTROLLER (at least in this simple model) has no idea
what variable the CONTROLLLEE is controlling; that is, the
CONTROLLER doesn't know that the CONTROLLEE is controlling
the length of the diagonal of the right triangle. All the
CONTROLLER "knows" is that by varying his output (which
happens to influence the height of the triangle) he can
control the behavior ("output height") of the CONTROLLEE.
The CONTROLLER doesn't need to know the CONTROLLEE's CV
in order to control the CONTROLLEE's behavior.

Best

Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
disturbance height=
                        SMOOTH3i(random uniform(-100,100 , 0),4,-50)+110
        ~
        ~ |

diagonal=
                               (SQRT((output height+disturbance height) * (output height+disturbance height\
                ) + width*width ))/1.5
        ~
        ~ |

width=
                -o2
        ~
        ~ |

i2=
        output height
        ~
        ~ |

e2=
        reference for output height-p2
        ~
        ~ |

p2=
        i2
        ~
        ~ ~ :SUPPLEMENTARY
        >

reference for output height=
                20
        ~
        ~ |

o2= INTEG (
                                        ((20*e2 )-o2)*0.5,
                                                              0)
        ~
        ~ |

error signal=
        reference for diagonal-perceptual signal
        ~
        ~ |

output height= INTEG (
                                (20*error signal - output height)*0.5,
                                                               0)
        ~
        ~ |

perceptual signal=
        diagonal
        ~
        ~ |

reference for diagonal=
                         SMOOTH3i(random uniform(-100,100 , 10),3,0)+100
        ~
        ~ |

********************************************************
        .Control
********************************************************~
                                                                                Simulation \
                Control Paramaters
        >

FINAL TIME = 50
        ~ Second
        ~ The final time for the simulation.
        >

INITIAL TIME = 0
        ~ Second
        ~ The initial time for the simulation.
        >

SAVEPER = 0.1
        ~ Second
        ~ The frequency with which output is stored.
        >

TIME STEP = 0.01
        ~ Second
        ~ The time step for the simulation.
        >

\\\—/// Sketch information - do not modify anything except names
V300 Do not put anything below this section - it will be ignored
*View 1
$0,0,Arial|12||0-0-0|0-0-0|0-0-0|-1–1--1|-1–1--1
10,1,reference for diagonal,177,389,62,8,0,3,0,0,0,0,0,0
10,2,perceptual signal,285,309,50,8,0,3,0,0,0,0,0,0
10,3,error signal,177,329,34,8,0,3,0,0,0,0,0,0
10,4,output height,116,242,38,8,0,3,0,0,0,0,0,0
10,5,diagonal,287,243,26,8,0,3,0,0,0,0,0,0
10,6,width,287,177,17,8,0,3,0,0,0,0,0,0
1,7,2,3,1,0,0,0,0,64,0,-1–1--1,1|(260,335)|
1,8,1,3,0,0,0,0,0,64,0,-1–1--1,1|(177,366)|
1,9,3,4,1,0,0,0,0,64,0,-1–1--1,1|(133,294)|
1,10,4,5,1,0,0,0,0,64,0,-1–1--1,1|(200,242)|
1,11,5,2,1,0,0,0,0,64,0,-1–1--1,1|(285,269)|
1,12,6,5,0,0,0,0,0,64,0,-1–1--1,1|(287,203)|
10,13,reference for output height,100,120,74,8,0,3,0,0,0,0,0,0
10,14,p2,35,176,9,8,0,3,0,0,0,0,0,0
10,15,e2,99,176,9,8,0,3,0,0,0,0,0,0
10,16,o2,208,177,9,8,0,3,0,0,0,0,0,0
10,17,i2,31,245,7,8,0,3,0,0,0,0,0,0
1,18,13,15,0,0,0,0,0,64,0,-1–1--1,1|(99,141)|
1,19,14,15,0,0,0,0,0,64,0,-1–1--1,1|(60,176)|
1,20,15,16,0,0,0,0,0,64,0,-1–1--1,1|(146,176)|
1,21,17,14,1,0,0,0,0,64,0,-1–1--1,1|(31,211)|
12,22,0,76,322,43,8,0,4,0,0,-1,0,0,0
CONTROLLEE
1,23,16,6,0,0,0,0,0,64,0,-1–1--1,1|(236,177)|
1,24,4,17,0,0,0,0,0,64,0,-1–1--1,1|(64,243)|
12,25,0,181,342,161,64,3,0,0,2,-1,0,0,0,0-0-0,0-0-0,|12||255-0-0
12,26,0,242,36,127,9,0,4,0,10,-1,0,0,0,0-0-0,0-0-0,|14||255-0-0
CONTROL OF BEHAVIOR SIMULATION
10,27,disturbance height,428,242,53,8,0,3,0,0,0,0,0,0
1,28,27,5,1,0,0,0,0,64,0,-1–1--1,1|(354,241)|
12,29,0,152,490,0,0,0,0,0,0,-1,0,0,0
12,30,0,179,157,164,54,3,4,0,0,0,0,0,0
CONTROLLER
///—\\\
:GRAPH CONTROLLER_BEHAVIOR
:TITLE Controller Behavior
:SCALE
:VAR reference for output height
:Y-MIN 0
:Y-MAX 50
:SCALE
:VAR output height
:Y-MIN 0
:Y-MAX 50
:GRAPH CONTROLLEE_BEHAVIOR
:TITLE Controllee's Behavior
:X-AXIS Time
:SCALE
:VAR reference for diagonal
:Y-MIN 80
:Y-MAX 120
:SCALE
:VAR diagonal
:Y-MIN 80
:Y-MAX 120
:L <%^E!@
1:Current.vdf
9:Current
15:0,0,0,0
19:100,0
5:output height

[Jeff_Vancouver2]

[from Jeff Vancouver 980911.1610]

[From Rick Marken (980911.1220)]

I ran your diagonal simulation in Vensim. I also ran the state trooper
simulation. Finally, I ran your control of behavior demos on the web.
This is all messing up a section of a paper I am writing about the control
of others (you all have created an error in a system of mine). I am
struggling to understand these simulations.

Let's focus on the diagonal simulation. One of the things that worries me
is the scale. Except for the initial oscillation, all the graphs (error,
outputs, inputs, though not disturbance or controllee reference) appear
fairly steady. However, if one looks at the controllee's changing
reference, it is has a variance relatively smaller than the variance
exhibited in the other graphs (i.e., the scale is +/- 5 units versus +/- 40
to 100 units for the outputs). Thus, how smooth is smooth?

Further, if I look at the error, I see that the controller has pretty
constant error. I cannot tell with the controllee as 0 is not a line on
the graph, so it is hard to tell. But the question is, are we seeing
control or compromise?

I tried to substitute the pulse function for the random and smoothing
functions, but got an overflow error. It seems it would be useful to see
what happens when something happens (a reference signal changes suddenly, a
disturbance is applied suddenly).

Finally, why is the diagonal divided by 1.5?

Sincerely,

Jeff

[R_Marken]

[From Rick Marken (980911.1530)]

Jeff Vancouver (980911.1610) --

Let's focus on the diagonal simulation. One of the things that
worries me is the scale. Except for the initial oscillation,
all the graphs...appear fairly steady. However, if one looks at
the controllee's changing reference, it is has a variance
relatively smaller than the variance exhibited in the other
graphs (i.e., the scale is +/- 5 units versus +/- 40
to 100 units for the outputs). Thus, how smooth is smooth?

The controller's reference is constant at 20. The controlled
input (output height) is nearly constant also, though it's off
by about 3 units. I don't know the RMS error or stability but
I imagine they are quite good.

The variance in the controllee's reference is only about 5 units;
I didn;t know how to make it bigger. If I did it wouldn't have
changed anything; the controlled variable (diagonal) tracks this
reference rather precisely (about 2 units constant deviation).

All graphed variables are measured the same units, by the way:
pixels.

Further, if I look at the error, I see that the controller has
pretty constant error. I cannot tell with the controllee as 0
is not a line on the graph, so it is hard to tell.

The constant error is the (nearly) constant distance between
reference plot and controlled variable plot; it's the same
constant (and _small_) deviation in both cases. Both the
controller and controllee are controlling quite well -- better
than I do in the real (Java) situation.

But the question is, are we seeing control or compromise?

Control.

Finally, why is the diagonal divided by 1.5?

To keep from getting that overflow error you got. There is
a strongly non-linear relationship between the controllee's
output (output height) and the controlled variable (diagonal)
because the controlled variable is the square root of the
product of the squares of two variables (height and width).
That's why it was so hard to tune the model up and eliminate
the initial oscillation.

Keep working on it and feel free to keep the question coming; I'm
glad it's relevant to something you are writing about.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[Marc_Abrams5]

From [ Marc Abrams (980912.1016) ]

[From Rick Marken (980911.1220)]

Rick, I am not trying to be contentious but why couldn't you
use a real example for this model? This is a perfect example
of a theoretic PCT model. It works well and shows what you
want it to show. Problems arise when you try to use this
model as an analogy or metaphor ( i.e. "state trooper" ) for
a real world example. It breaks down. Can you make up a
story about a real life example that fits the model? That
might make for better discussion.

Marc

For those of you with Vensim, here is another version of
Bruce Abbott's "Control of Behavior" model. The model
shows a CONTROLLER controlling a CONTROLLEE's output by
disturbing a variable (called "diagonal") that is

controlled

by the CONTROLLEE. What is interesting (to me) about the
model is that the CONTROLLER is exerting control by

disturbing

a dimension of the controlled variable that is not being
affected by the output (called "output height" in the

model)

that is being controlled.

The controlled variable is the diagonal length of a right
triangle, as shown below:

height *
= * *
output * * < --- diagonal = qi
height (qo of * *
CONTROLLEE) + * *
disturbance * *
height (d) * *
               * *
               ************************
                       width = qo of CONTROLLER

The CONTROLLER's output ("o2" in the model) affects only

the width

of the triangle; the CONTROLLEE's output ("output height")

affects

only the height of the triangle. Moreover, the height of

the

triangle is also affected by a random disturbance.

Nevertheless,

the CONTROLLER is able to disturb the width of the triangle

as the

means of keeping the CONTROLLEE's effect on the height of

the

disturbance ("output height", which is the CONTROLLER's CV)

under

control.

As you can see in the graphs of the CONTROLLER's and

CONTROLLEE's

behavior, control is not perfect in this case; the trace of

the

controlled variable for both CONTROLLER and CONTROLLEE is

offset

somewhat from the trace of the reference specification.

This

relatively poor control results from the fact that I had to

reduce

the gain for both CONTROLLER and CONTROLLEE in order to

stabilize

the behavior of the interacting control systems. I have

still

been unable to remove an initial transient oscillation from

the

behavior of both systems; I don't know if it's possible to

remove

it using just a simple control model (it might take a

couple

levels of simple control systems); those who are interested
can experiment with different combinations of gain and

slowing

for each control system.

The main things to understand about this model, however,

are

1) the CONTROLLEE is controlling a variable (diagonal = qi)

that

is a function of _two_ independent environmental variables

and

2) the CONTROLLER is able to control a one-dimensional

variable

(the CONTROLLEE's output -- "output height") by disturbing

an

aspect of the variable the CONTROLLEE is controlling --

even

though it is not and aspect of the controlled variable that

is

directly affected by the CONTROLLER's disturbance.

The CONTROLLER (at least in this simple model) has no idea
what variable the CONTROLLLEE is controlling; that is, the
CONTROLLER doesn't know that the CONTROLLEE is controlling
the length of the diagonal of the right triangle. All the
CONTROLLER "knows" is that by varying his output (which
happens to influence the height of the triangle) he can
control the behavior ("output height") of the CONTROLLEE.
The CONTROLLER doesn't need to know the CONTROLLEE's CV
in order to control the CONTROLLEE's behavior.

Best

Rick
--
Richard S. Marken Phone or Fax: 310

474-0313

Life Learning Associates e-mail:

rmarken@earthlink.net

···

http://home.earthlink.net/~rmarken

[Bruce_Gregory]

[From Bruce Gregory (980912.1140 EDT)]

Rick Marken (980911.1530)

Until I see a model of a trooper controlling the speed of my car without
blocking the road ahead of me in the Marken-Abbott fashion I will assume
that it is not possible and will do my best to ignore claims to the
contrary.

Bruce Gregory

[R_Marken]

[From Rick Marken (980912.1020)]

Marc Abrams (980912.1016)

Rick, I am not trying to be contentious but why couldn't you
use a real example for this model? This is a perfect example
of a theoretic PCT model. It works well and shows what you
want it to show. Problems arise when you try to use this
model as an analogy or metaphor ( i.e. "state trooper" ) for
a real world example. It breaks down. Can you make up a
story about a real life example that fits the model? That
might make for better discussion.

I have already applied the model to a real world example;
I wrote a Java program (not posted to the Web) where the
subject (me) is able to control the output of another control
system that happens to be controlling a perception of the
length of the diagonal of a right triangle. So the Vensim
model CBCV.mdl accurately describes the behavior of a real
controller (me) controlling the output of a "diagonal
controlling" controllee (computer). The model doesn't "break down"
when applied to a "real world example"; it works like a charm.

But perhaps what you mean by "real world example" is more like
"familiar example". I agree that this might help people understand
the implications of the model. So here goes.

What the CBCV.mdl shows is that it is possible to control a
person's behavior even if you don't know exactly what perceptual
variable that person is controlling; all you have to know is that
something you do (equivalent to the controller varying "width"
in the model) has a reasonably regular effect on something someone
else does (like the controllee varying his output effect on the
height of the triangle).

A familiar example would be controlling behavior by the use of
rewards. It has been found that variations in the size of a
reward (such as money) can have a reasonably regular effect on
certain actions (work output). From a PCT perspective, this must
be happening because the reward is a disturbance to some
perception that is _also_ influenced by work. For example, a
person who is controlling for money will do whatever works to
bring his perception of money to its reference. What my demo
shows is simply that you don't have to know exactly what this
person is controlling in order to be able to control his work
output with money. The person might be controlling for money;
but he might be controlling for something that is only partly
influenced by money (just as the controllee in my demo is
controlling a variable -- diagonal -- that is only partly
influenced by the controller's output --height). For example,
the person might be controlling for being respected, a perception
that might depend not only how much money the person makes but,
also, on the treatment he gets from his management (this
variable is analogous to the height disturbance in the model).

Let's say that the person is controlling for respect and that
respect is a complex function of money and "treatment by
management". What my model shows is that it's still possible to
control this person's work output using money, regardless of how
he is treated by management. Of course, you might have to pay
him more to make up for poor management treatment. But, like
the controller's control of the controllee's output in CBCV.mdl,
it can be done. This shows that you don't need to know what
variable the person is actually controlling in order to be able
to control the person's behavior.

Some people have worried that PCT might actually provide the
tools to help people control people even more effectively.
What my demo shows is that PCT won't help them; people who
want to control other people (an unfortunate but apparently
unavoidable want) can't do any better than they have been
doing for the past thousands of years using the tried and
true methods of reward and punishment (see MSB, p. 109).

Best

Rick

···

---
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

[R_Marken]

[From Rick Marken (980912.1050)]

Bruce Gregory (980912.1140 EDT)--

Until I see a model of a trooper controlling the speed of my
car without blocking the road ahead of me in the Marken-Abbott
fashion I will assume that it is not possible and will do my
best to ignore claims to the contrary.

                               Reference for
Trooper speed of
                                 your car
                                     >
                                     v
                              p---->|C|------>e
                              > >
                             >s> >o>
Environment----------------------------------------
                  radar speed i o angle of dial
                   measure | |
                              > >
                           speed of radio signals
                           your car |
                              ^ |
                              > >
                     foot pressure --> <----lever---
                                                                    pressure

The trooper has a radio controlled lever installed in your car;
he controls your speed by turning a dial that sends raio signals
to the lever. The lever can exert far more pressure on the
accelerator pedal than you can. So the trooper can control the speed
of your car without blocking the road. This is an example of
coercive control; the trooper is controlling a variable that you
are also controlling (unsuccessfully).

It would be a good exercise to implement this model in Vensim.

Best

Rick

···

----
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

[Abbott_Bruce]

[From Bruce Abbott (980912.1250 EST)]

Bruce Gregory (980912.1140 EDT)

Until I see a model of a trooper controlling the speed of my car without
blocking the road ahead of me in the Marken-Abbott fashion I will assume
that it is not possible and will do my best to ignore claims to the
contrary.

The state trooper controlling the speed of passing cars by issuing tickets
to motorists who exceed the speed limit would require a different model.
Consider the following scenario:

    The trooper monitors the speeds of passing cars via radar. When the
    radar-indicated speed of a car exceeds the trooper's criterion (which
    may be, for example, speed limit + 8 mph), the trooper pursues the
    offending motorist with lights flashing. The motorist pulls over to
    the side of the road and the trooper issues a speeding ticket. The
    motorist, not wishing to receive more tickets, goes on his way but
    now is careful not to exceed the speed limit.

We could model this scenario in a variety of ways, depending on what
assumptions we wish to make about the trooper and the motorist. Each
resulting model would constitute an hypothesis, and would require testing to
determine its validity as a description of the actual mechanism.

The trooper is probably controlling for giving tickets to speeders (and for
not giving tickets to nonspeeders). This is a logical relationship: if
speeding then ticket else don't ticket. This may be all there is, from the
trooper's point of view. Alternatively, this control may constitute the
means by which the trooper hopes to control the motorist's speed, by making
the relationship between speeding and receiving a ticket (with associated
delays, fine, points penalty, etc.) credible to the motorist. If this
works, then the motorist will be thrown into a conflict between continuing
to speed (with its associated perceived gains in reduced travel time etc.)
and risking further tickets. The trooper may hope that this conflict will
be resolved by the motorist readjusting his reference for speed to the level
of the speed limit, thereby minimizing the risk of further tickets. If this
is the mechanism, then it would represent a second way by which the
motorist's behavior can be controlled (at least potentially), by creating a
conflict in the controllee which the controllee will resolve (if the
controller's efforts are successful) by changing his or her (the
controllee's) reference to the value desired by the controller.

Even if the trooper is only concerned about seeing that speeders get
tickets, one can reasonably assume that lawmakers, who created the
speed-limit law, did so with the intention of limiting the speeds of the
motor vehicles. The trooper's actions may tend to have this effect (even
though not every motorist will comply with the law, and some with continue
to ignore it even after being ticketed) whether the trooper herself is
controlling for it or not, as a side-effect of issuing tickets to speeders.

Of course, my verbal descriptions are poor substitutes for actual models.
It is evident that an adequate model to account for the behaviors of both
trooper and motorist in this scenario would have to be fairly complex, and
would embody a number of assumptions as to what is going on in the heads of
these two living control systems. If anyone would care to take a stab at
creating such a model in Vensim, be my guest, as I do not intend to do so.
(I'm busy enough right now, thank you.) I think it would be an interesting
and worthwhile exercise, although of course, the model would only represent
one person's guesses. Even so, if successful it would demonstrate that a
model having the "right" behavioral properties can actually be constructed
from basic control-system elements. That would be a great start on the problem.

Regards,

Bruce

[Bruce_Gregory]

[From Bruce Gregory (980912.1445 EDT)]

Bruce Abbott (980912.1250 EST)

Of course, my verbal descriptions are poor substitutes for actual models.
It is evident that an adequate model to account for the behaviors of both
trooper and motorist in this scenario would have to be fairly complex, and
would embody a number of assumptions as to what is going on in
the heads of
these two living control systems. If anyone would care to take a stab at
creating such a model in Vensim, be my guest, as I do not intend to do so.
(I'm busy enough right now, thank you.) I think it would be an
interesting
and worthwhile exercise, although of course, the model would only
represent
one person's guesses. Even so, if successful it would demonstrate that a
model having the "right" behavioral properties can actually be constructed
from basic control-system elements. That would be a great start
on the problem.

I think your verbal descriptions are perfectly reasonable. Let's imagine for
the moment that they are translated into a working model with the "right"
behavioral properties. My question would be, does the model represent a
situation in which the trooper is controlling my speed? If I keep my speed
below 75 mph because I want to avoid a ticket, is the trooper hidden over
the crest of the next hill controlling my speed? If I confess my infidelity
because I fear that if I don't God will send me to hell, is God controlling
my behavior? There _may_ be some virtue in drawing a distinction between
controlling my speed by blocking my path and controlling my speed by
encouraging me to adopt a limit to the reference speed I will adopt when
driving on the turnpike.

Bruce Gregory

[Bruce_Gregory]

[From Bruce Gregory (980912.1420 EDT)]

Rick Marken (980912.1050)

The trooper has a radio controlled lever installed in your car;
he controls your speed by turning a dial that sends radio signals
to the lever. The lever can exert far more pressure on the
accelerator pedal than you can. So the trooper can control the speed
of your car without blocking the road. This is an example of
coercive control; the trooper is controlling a variable that you
are also controlling (unsuccessfully).

Is this the same mechanism that extraterrestrials use to control your
thoughts?

Bruce Gregory

[Marc_Abrams5]

From [ Marc Abrams (980912.1728) ]

[From Bruce Abbott (980912.1250 EST)]

It is evident that an adequate model to account for the
behaviors of both trooper and motorist in this scenario

would >have to be fairly complex, and would embody a number
of >assumptions as to what is going on in the heads of

these two living control systems. If anyone would care to
take a stab at creating such a model in Vensim, be my
guest, as I do not intend to do so. (I'm busy enough right
now, thank you.) I think it would be an interesting
and worthwhile exercise, although of course, the model
would only represent one person's guesses. Even so, if
successful it would demonstrate that a model having the
"right" behavioral properties can actually be constructed
from basic control-system elements. That would be a great

start on the problem.

Great idea Bruce, and since I seem to be the one haranguing
poor Rick about a "real" problem. I'll do one. But it won't
be on the state trooper. Give me a couple of days. I want to
give this some thought. :

Marc

[R_Marken]

[From Rick Marken (980912.2230)]

Bruce Gregory (980912.1445 EDT)

Let's imagine for the moment that they are translated into
a working model with the "right" behavioral properties. My
question would be, does the model represent a situation in
which the trooper is controlling my speed? If I keep my speed
below 75 mph because I want to avoid a ticket, is the trooper
hidden over the crest of the next hill controlling my speed?

Why not build the model and answer this question for yourself?
What you want to build is a model of a control system (the
trooper) that is controlling your speed only when you go over
the speed limit. Bruce Abbott has provided the basic building
blocks in his Vensim "Control of behavior" simulation. So why
not go for it! I'd like to see it myself.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

[Bruce_Gregory]

[From Bruce Gregory (980913.0935 EDT)]

Rick Marken (980912.2230)

Why not build the model and answer this question for yourself?
What you want to build is a model of a control system (the
trooper) that is controlling your speed only when you go over
the speed limit. Bruce Abbott has provided the basic building
blocks in his Vensim "Control of behavior" simulation. So why
not go for it! I'd like to see it myself.

Fair enough. We start with a model in which I am controlling my speed and
the trooper is controlling the program SPEED LESS THAN 75? LOOK AT NEXT CAR;
SPEED MORE THAN 75? PURSUE AND ISSUE TICKET. If I am controlling with a
reference speed less than 75 mph, nothing happens, i.e., I continue to
control the speed of my car and the trooper moves his attention to the car
behind me. The two control loops are effectively uncoupled. If, on the other
hand, I am controlling with a reference speed greater than 75 mph, the
trooper controls a program TURN ON LIGHT AND SIREN, PULL UP BEHIND CAR IF
CAR SLOWS DOWN AND PULLS OVER, SLOW DOWN, STOP AND WRITE TICKET, IF CAR DOES
NOT PULL OVER, CONTINUE PURSUIT AND FORCE CAR TO PULL OVER. When I hear the
siren and see the flashing lights in my rear view mirror, a higher level
system sets my reference level for speed to lower and lower levels and I
pull over to the side of the road.

At the moment I lack the modeling skills needed to convert this story into a
working model, but conceptually I think it is quite clear. At _no_ time does
the trooper control the speed of my car. I, however, control the speed of
his car as I slow down ala Abbott assuming that he does not want to hit my
car. If my car is faster than his and I have reason to flee, I might even
_increase_ the speed of my car when I hear his siren. If he wants to catch
me, I am again am controlling the speed of his car.

Bruce Gregory

[Bill_Powers1]

[From Bill Powers (980913.0732 MDT)]

Rick Marken (980912.2230)]

Bruce Gregory (980912.1445 EDT)

Bruce Abbott .... et. al.

One thing that would help in the "state trooper" type of discussion would
be to state first what the control system is supposed to be controlling,
and THEN work out a model for it. If you change your mind about what's
being controlled, then you have to change the model, too. If you change the
situation you're talking about and the other person is still thinking about
the old situation, not much communication will happen.

To define a variable that someone is controlling, you have to define an
input quantity and an output quantity. The input quantity is the physical
situation that is being perceived, and the output quantity is the means by
which the control system has an effect on the input quantity. Disturbances
are independent variables in the environment that also influence the input
quantity.

If the person does perceive the input quantity, and does have an effect on
it by acting on the environment, and if the signs are all right and the
loop gain is sufficient, then the person is controlling the input quantity.
There's no need to ask what the environment thinks about being controlled,
or whether it wants to be controlled or likes being controlled or is only
pretending to be controlled or would have behaved that way anyhow. That's
irrelevant in defining a control process. If the loop is closed and the
loop gain is high enough (which you define to suit your purposes), control
is taking place. If the environment changes so that one minute later the
loop is broken, control is no longer taking place.

Also, it is good to remember, now and then, that people do not control
things, they control variables. So one person does not control another
person, any more a person controls a car. A person controls some _variable
aspect_ of another person, such as where he is standing or which way he's
looking or how much he's eaten or the size of his salary. When a kid says
"There's a spider on your arm," the other kid (or adult) looks, the first
kid says "Made ya look, made ya look!", and that is perfectly true. The
first kid controlled where the other was looking. It may never work again,
but it worked that time, and the relationship was that of control.

Best,

Bill P.

[Marc_Abrams5]

From [ Marc Abrams (980913.1024) ]

[From Bill Powers (980913.0732 MDT)]

Bill, this should be Modeling session 1b. :-). Thanks, you
helped clarify a few things for me. Now lets see if I can
use these new found insights and incorporate them into my
model :-).

Marc

[Abbott_Bruce]

I think your verbal descriptions are perfectly reasonable. Let's imagine for
the moment that they are translated into a working model with the "right"
behavioral properties. My question would be, does the model represent a
situation in which the trooper is controlling my speed? If I keep my speed
below 75 mph because I want to avoid a ticket, is the trooper hidden over
the crest of the next hill controlling my speed? If I confess my infidelity
because I fear that if I don't God will send me to hell, is God controlling
my behavior? There _may_ be some virtue in drawing a distinction between
controlling my speed by blocking my path and controlling my speed by
encouraging me to adopt a limit to the reference speed I will adopt when
driving on the turnpike.

That is an interesting question, but I think the answer depends more on what
you mean by "the trooper is controlling my speed" than on the specifics of
the model. Certainly, you are controlling your (your vehicle's) speed, by
varying the accelerator position so as to maintain some selected (reference)
value. And certainly, you are setting that reference speed in order to meet
other goals, which may include getting to Point B, minimizing driving time,
keeping vehicle speed within the limits of one's perceived ability to
maintain control, keeping the cost of the trip to a minimum, and others.

Just as certainly, the trooper is acting to control your speed when he (a)
monitors your speed with his radar and (b) pursues you with lights flashing
and siren going when you are detected speeding. If this action fails to
bring about your stopping, he will take other actions, if necessary calling
ahead to set up road blocks, shooting out your tires, or forcing you off the
road. He is acting as a control system to bring your speed (as he perceives
it) to zero; whether these actions succeed in doing so is another matter.

But is the state controlling your speed when it sets up speed laws and
enforces them via the state troopers? By so doing it changes the
environment in which drivers operate their vehicles. In this changed
environment, there is a probabilistic relationship between exceeding the
posted speed limit and receiving a ticket. Those for whom the prospect of a
ticket arouses anxiety can reduce this anxiety to zero by driving no faster
than the posted limit. Yet driving at this speed may conflict with other
goals, such as not being late for an appointment, and in such cases drivers
may resolve the conflict by maintaining the speed limit or by speeding and
risking the ticket. In one sense your (the motorist's) behavior is
completely under your control -- you set your own references, after all.
Yet by setting up an artificial contingency between your motoring speed
(relative to a posted speed) and the probability of receiving a ticket with
its associated delays and fine, the state sets up an internal conflict in
the motorists which in many cases will be resolved by keeping speed within
the posted limit. To the extent that this contingency works, the state is
able to bring motorists' speeds down to values near the posted (reference)
values. When it doesn't, the state, through its agents, the state troopers,
takes action by stopping the offenders and issuing tickets. To the extent
that this action convinces motorists to slow down, we have a nice little
control system, complete with reference values (posted speed limits),
sensors (radar-equipped state troopers), and output systems (state troopers)
whose actions tend to bring the controlled variable (peceived speed of
motorists) down toward the reference values. Yet the motorists are
completely in control of their own motoring speeds. Neat paradox, don't you
think?

Is the trooper, hidden over the next hill, controlling your speed? Yes and
no. You are controlling your speed. If you are speeding, the trooper is
about to detect, and then act with the intention of controlling your speed,
both in the present (as he tries to stop you) and in the future (as you
reset your speed to avoid further tickets). But until you crest the hill,
he doesn't even know you exist, and is not controlling your motoring speed.
However, to the extent that your knowledge of the contingency between
speeding and ticketing has convinced you to set your speed reference to the
posted limit as a means of avoiding a ticket, the state, by setting up the
contingency and demonstrating its ability to enforce it (you've seen other
motorists getting tickets for speeding, or have gotten one yourself), is
controlling your speed even before your vehicle crests the hill and
encounters the beam of the trooper's radar. This trooper in particular is
not controlling your speed, but the possibility that he will be there, over
the next hill, certainly is.

Regards,

Bruce

[Bruce_Gregory]

[From Bruce Gregory (980913.1105 EDT)]

Bruce Abbott September 13, 1998 9:45 AM

Is the trooper, hidden over the next hill, controlling your
speed? Yes and
no. You are controlling your speed. If you are speeding, the trooper is
about to detect, and then act with the intention of controlling
your speed,
both in the present (as he tries to stop you) and in the future (as you
reset your speed to avoid further tickets). But until you crest the hill,
he doesn't even know you exist, and is not controlling your
motoring speed.
However, to the extent that your knowledge of the contingency between
speeding and ticketing has convinced you to set your speed
reference to the
posted limit as a means of avoiding a ticket, the state, by setting up the
contingency and demonstrating its ability to enforce it (you've seen other
motorists getting tickets for speeding, or have gotten one yourself), is
controlling your speed even before your vehicle crests the hill and
encounters the beam of the trooper's radar. This trooper in particular is
not controlling your speed, but the possibility that he will be
there, over
the next hill, certainly is.

I think I'll heed Bill's admonition. When it comes to the speed of my car,
only I am in a position to control it. The state may have sanctions designed
to influence the way I establish the reference levels I choose for my speed,
but these sanctions do not constitute control in the sense that PCT
describes control.

Bruce Gregory

[Bruce_Gregory]

[From Bruce Gregory (980913.1108 EDT)

Bill Powers (980913.0732 MDT)

If the person does perceive the input quantity, and does have an effect on
it by acting on the environment, and if the signs are all right and the
loop gain is sufficient, then the person is controlling the input
quantity.

Thanks, Bill.

Bruce Gregory

[R_Marken]

[From Rick Marken (980913.1000)]

Bill Powers (980913.0732 MDT)--

If the person does perceive the input quantity, and does have
an effect on it by acting on the environment, and if the signs
are all right and the loop gain is sufficient, then the person
is controlling the input quantity.

Bruce Gregory (980913.1108 EDT)--

Thanks, Bill.

I'm glad you liked it. If you subtitute "trooper" for "person"
and "speed of your car" for "input quantity" you will see that
Bill is saying just what Bruce Abbott and I have been saying: that,
if the signs are all right and the loop gain is sufficient, then
the trooper is controlling the speed of your car when the speed
of your car is the controlled input variable.

If you had bothered to quote the rest of Bill's comment, you (and
everyone else) could have seen that Bill was also saying that
the trooper is doing this controlling whether you keep your speed
below the speed limit or not. The part of Bill's comment that you
didn't quote went as follows:

There's no need to ask what the environment thinks about
being controlled, or whether it wants to be controlled or
likes being controlled or is only pretending to be controlled
or would have behaved that way anyhow. That's irrelevant in
defining a control process. If the loop is closed and the
loop gain is high enough (which you define to suit your purposes),
control is taking place. If the environment changes so that one
minute later the loop is broken, control is no longer taking place.

Since you (the driver) are the main cause of variations in the
environmental variable ("speed of your car") that is being
controlled by the trooper, you can see that what Bill is saying
here is a polite (if thorough) contradiction of all your claims
about the control going on in the "trooper" example. Bill is
saying that it doesn't matter whether or not you want the speed of
your car to be controlled by the trooper, whether or not you are
only pretending that the speed of the car is being controlled by
the trooper, whether ot not you would have maintained the speed of
the car where the trooper wanted it even if the trooper were not
around; all that is irrelevant to defining the control process
(in this case, the process by which the trooper controls the speed
of your car). If the trooper is controlling the speed of your car
then he is controlling the speed of your car; what you (the
environment, in this case) thinks about it is irrelevant to
defining the trooper's control process.

If the trooper is organized as a control system with respect to
a particular environmental variable (like the speed of your
car) and if the loop is closed and the loop gain is high enough
then control is taking place -- even if the speed of your car
matches the trooper's reference for your speed so that the
trooper need do no more than sit quietly on his mototrcycle to
keep the speed of your car at his reference level.

Bruce Gregory (980913.0935 EDT) --

At the moment I lack the modeling skills needed to convert
this story into a working model

This seems like a serious shortcoming in a person who says that
we can only understand what we can model.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

[Bruce_Gregory]

[From Bruce Gregory (980913.1345 EDT)]

Rick Marken (980913.1000)

If the trooper is organized as a control system with respect to
a particular environmental variable (like the speed of your
car) and if the loop is closed and the loop gain is high enough
then control is taking place -- even if the speed of your car
matches the trooper's reference for your speed so that the
trooper need do no more than sit quietly on his motorcycle to
keep the speed of your car at his reference level.

O.K. If you want to believe in magic, go right ahead. The trooper is _not_
organized like a control system with respect to the speed of my car. _I_ am
organized like a control system with respect to the speed of my car. There
is no way that the trooper can alter the speed of my car except by forcing
me off the road. Unless you believe in magic... My car's speed is controlled
by (among other things) the throttle setting. Only I have access to the
throttle setting (except in your wild flights of fancy).

Bruce Gregory (980913.0935 EDT) --

> At the moment I lack the modeling skills needed to convert
> this story into a working model

This seems like a serious shortcoming in a person who says that
we can only understand what we can model.

I have many serious shortcomings. Fortunately magical thinking is not one of
them.

Bruce Gregory