Applying disturbances

[From Bill Powers (970311.1939 MST)]

I keep startling myself by realizing that this is still the same day on
which Mary and I got up at 3:00 AM in Holbrook, Arizona and drove the rest
of the way back to Durango. One of those 72-hour days.

···

----------------------------------------
In recent posts, someone (I forget who) raised the point that applying
disturbances to do the Test might be unethical. The point seems to be that
arbitrarily disrupting someone's control processes is not nice.

This is another in a series of misunderstandings about what it means to
apply a disturbance when doing the Test. A disturbance is NOT a change in
the controlled variable. It is a change in some other variable -- a
"disturbing variable" -- which affects the controlled variable additively
(adding to the person's own effects on the same variable). When you apply a
disturbance to a variable someone is controlling, there is not supposed to
be any significant change in the controlled variable. What is supposed to
happen instead is that the person's action immediately changes, cancelling
the effect of the disturbing variable. The whole point of the Test is to
apply a disturbance that does NOT succeed in disrupting control. If you use
such a large or sudden disturbance that there is a large effect on the
controlled variable, you're looking at a failed control system, not a
working one.

Consider the rubber band demo. Many people, when playing the role of the
experimenter, move the disturbing end of the rubber bands in jerks or in
large fast movements. Of course when they do this, the knot moves all over
the place and is hardly ever to be seen over the dot. The controller's hand
also moves around in large fast patterns, always too late to achieve
control. The experimenter is not doing the test; instead, the objective
seems to be to thwart all attempts at control. And none of the neat
relationships we can demonstrate can be seen, either.

If you did that to a person while trying to discover what's under control,
it might well be unethical. But worse than that, you would learn almost
nothing. You certainly wouldn't see the symmetry between action and
disturbance, and if the controller were slowly varying the reference
condition, you wouldn't be able to see that, either. It would be essentially
impossible to test hypotheses about what's being controlled.

If you were to apply a disturbance _as a change in the controlled variable_,
this would be equivalent to seizing the knot and moving it off the dot. If
you did this, the controller might make an attempt to pull the knot back
using that end of the rubber bands, but it would obviously be futile and the
controller would simply quit trying to control in that way. You wouldn't see
an "open-loop response." You'd see a higher-order control system saying "The
heck with that."

In the rubber-band experiment, the disturbing variable is not a change in
the knot's position, but simply the position of the experimenter's end of
the rubber bands, in laboratory space. If the controller were a perfect
control system, a disturbance could be applied with NO VISIBLE EFFECT ON THE
POSITION OF THE KNOT. Of course some people are better controllers than
others and nobody controls perfectly, but it is quite possible for the
varying disturbance to have so little effect on the knot that random
variations mask it. In that case the position of the controller's end will
closely mirror the position of the experimenter's end, and we can see all
the basic relationships we expect.

The basic point is this: the disturbance must be applied in such a way that
it could be resisted successfully without excessive effort on the part of
the controller. This means using just enough disturbance to be sure you're
seeing the action that prevents it from having any effect. It means keeping
the speed of variations well within the bandwidth of control, so the person
can easily keep up with the changes. It means keeping the amplitude of the
disturbance small enough so that the person can easily maintain control.
Think of applying the disturbance through a rubber band, not directly to the
knot. That's why rubber bands were used in this demo: to make control
possible even while disturbances are being applied.

I've told this one before, but it bears repeating here. One day, while Dag
Forssell was driving me on the freeway from his house to the airport, I
asked him if he wanted to see a really tight control system. Naturally, he
said yes. So I took hold of the side of steering wheel and started pulling
down on it (making sure not to pull too suddenly). By increasing the pull at
a modest rate, I actually raised it to a pretty substantial level before Dag
called a halt to the experiment. The car, traveling along at 60 miles per
hour in traffic, naturally stayed right in its lane, its path not deviating
more than a foot. Dag opposed my disturbance very successfully -- I made
sure he did, because as a passenger in the same car I had a personal
interest in NOT disrupting his control of the car. When you think about
applying disturbances in doing the Test, imagine that you could be killed if
the controller actually failed to control as well as usual.

Of course, as Martin Taylor will no doubt comment again, one can learn
things about a control system, such as its dynamic properties and limits of
action, by applying disturbances that tax its abilities. But even in that
application, it is not necessary to use large enough disturbances to prevent
successful control. By applying small oscillating disturbances at various
frequencies and watching the relationship between the action and the
disturbance, one can map the frequency response of the control system
without seriously affecting the control that's going on. It's not so easy to
test the limits of control without disrupting things, but brief enough
disturbances can minimize the disruptive effects.

This is particularly important in dealing with living hierarchical control
systems, because as I said, people will simply not continue trying to
control when they aren't succeeding. A tracking task using a disturbance
large enough that the person's behavior cancels only 90 percent of it is
experienced as an extremely difficult task. That's with an error of only ten
percent of the variations in the disturbance. When the error gets to 15
percent or so, most people simply give up, saying they can't do it. This is
barely enough error to give an indication of the "corner frequency" where
the control accuracy is beginning to fall off. You'd have to go all the way
to 30 percent error to get to the customary corner frequency, and no human
controller I have seen would continue trying to control with that much error.

In dealing with artificial systems, we don't have this problem: they never
give up, even when the disturbance is varying so rapidly that the errors are
90 percent of what they would be with no control. So we can map the
frequency response of the artificial system from zero to infinity, if we
like, even with big disturbances. This can't be done with living systems:
you end up with a different system from the one you started with -- unless
the disturbance is so small that the person can just average it out,
perceptually.

So no, the Test is not unethical if it is done right.

Best,

Bill P.

[Martin Taylor 970413 01:10]

Bill Powers (970311.1939 MST)]

I keep startling myself by realizing that this is still the same day on
which Mary and I got up at 3:00 AM in Holbrook, Arizona and drove the rest
of the way back to Durango. One of those 72-hour days.
----------------------------------------
In recent posts, someone (I forget who) raised the point that applying
disturbances to do the Test might be unethical. The point seems to be that
arbitrarily disrupting someone's control processes is not nice.

These two paragraphs may have something to do with one another. It's not
nice working after such a long drive.

As I remember it, the ethical problems was not with disrupting someone's
control processes in a way that allows them to sustain control. It was
applying disturbanced to perceptions that may not be well controlled but
that are vital (as would have been a failure of Dag's steering output
for lane control). The problem was that one could not know whether the
disturbance would disrupt control and induce a reorganization that might
lead to dangerous actions.

My memory may be wrong. Lots of water under the (Red River) bridge since
then.

Martin

[Martin Taylor 970414 11:30]

Bill Powers (970311.1939 MST)]

In recent posts, someone (I forget who) raised the point that applying
disturbances to do the Test might be unethical. The point seems to be that
arbitrarily disrupting someone's control processes is not nice.

This is another in a series of misunderstandings about what it means to
apply a disturbance when doing the Test. A disturbance is NOT a change in
the controlled variable. It is a change in some other variable -- a
"disturbing variable" -- which affects the controlled variable additively
(adding to the person's own effects on the same variable). When you apply a
disturbance to a variable someone is controlling, there is not supposed to
be any significant change in the controlled variable.

I commented on an aspect of this yesterday. But there's another point I
didn't make (hadn't noticed), and that is the fact that it is highly
unlikely that the disturber (person) knows exactly what variables the
disturbee (person) is controlling when performing the Test. And even
more unlikely that the disturber is able to influence precisely those
variables without affecting other perceptions available to the disturbee,
whether controlled or not.

To put it more plainly, (almost) all instances of the Test involve side-
effects that influence other perceptions of the person whose control is being
tested. These side effects, if they influence controlled perceptions, may
or may not have dangerous consequences. If they influence perceptions at
that time uncontrolled, they may influence the disturbee to bring those
perceptions under control at the cost of losing control over others. The
disturber does not know where bifurcations exist in the system composed of
the disturbee and the environment--i.e. where slight changes in the
environment may make consequential changes in the results of actions
(think of the standard comedy trick of moving a chair slightly away from
someone in the process of sitting down, for example).

If the Test were to be applied always and only to the CEVs of perceptions
correctly assumed to be controlled, and the Tester could guarantee that the
influences permitted the controlled perception to remain under control,
there would be no ethical issue in using it, no matter how high a perceptual
level was being tested. But these conditions almost never hold, and
specifically do not hold when the Tester is trying to find out which
perceptions are controlled out of a set of candidates. In other words,
"What is supposed to happen" in the following may well happen. But if
it does not, and if the Tester is controlling for perceiving no damage
to the disturbee, the Tester may find out too late that this perception
has not been successfully controlled. And therein lies the ethical
issue, at least in part.

What is supposed to
happen instead is that the person's action immediately changes, cancelling
the effect of the disturbing variable. The whole point of the Test is to
apply a disturbance that does NOT succeed in disrupting control. If you use
such a large or sudden disturbance that there is a large effect on the
controlled variable, you're looking at a failed control system, not a
working one.

Yes, but the Tester may not know whether even the perception under Test
is being controlled well, being controlled on the brink of failure, or
not being controlled. Still less does the Tester know these things
about the perceptions influenced by side-effects of the Test.

I've told this one before, but it bears repeating here. One day, while Dag
Forssell was driving me on the freeway from his house to the airport, I
asked him if he wanted to see a really tight control system. Naturally, he
said yes. So I took hold of the side of steering wheel and started pulling
down on it (making sure not to pull too suddenly). By increasing the pull at
a modest rate, I actually raised it to a pretty substantial level before Dag
called a halt to the experiment. The car, traveling along at 60 miles per
hour in traffic, naturally stayed right in its lane, its path not deviating
more than a foot.

What might have happened if, in resisting the disturbance, Dag had suddenly
and painfully pulled a shoulder muscle?

Martin

[From Bill Powers (970414.1318 MST)]

Martin Taylor 970414 11:30--

I commented on an aspect of this yesterday. But there's another point I
didn't make (hadn't noticed), and that is the fact that it is highly
unlikely that the disturber (person) knows exactly what variables the
disturbee (person) is controlling when performing the Test. And even
more unlikely that the disturber is able to influence precisely those
variables without affecting other perceptions available to the disturbee,
whether controlled or not.

To put it more plainly, (almost) all instances of the Test involve side-
effects that influence other perceptions of the person whose control is
being tested. These side effects, if they influence controlled
perceptions, may or may not have dangerous consequences.

This is true, but it's true of every interaction of one person with another.
There is a very common city-dweller's way of getting across a busy street on
foot: the basic trick is to start across the street while carefully avoiding
eye contact with any approaching drivers. You rely on the drivers to respond
to this disturbance by slowing down rather than hitting a pedestrian who
obviously isn't looking at the traffic. It works fine. Dangerous? Probably,
but probably not significantly more dangerous than being on foot in a city
in the first place.

The alternative to not disturbing other people's controlled variables is not
to interact with them at all.

What might have happened if, in resisting the disturbance, Dag had
suddenly and painfully pulled a shoulder muscle?

I'm not a suicidal idiot. I applied the disturbance slowly, and with a
_very_ firm grip on the wheel, and using position, not force, control. If
Dag had let go of the wheel entirely, its rim would have moved perhaps an
inch before I had control of it. It was, in fact, quite a safe experiment.

Of course, Dag might have been on the very brink of a psychotic episode, and
my tug at the wheel might have been the trigger that made him suddenly turn
the wheel the same way I was pulling and send the car careening across the
traffic lanes. In that case, we could both have been dead or worse as a
result of this demo. For all I know, simply asking Dag if he wanted to see a
really tight control system might have triggered the same episode. Maybe he
was already seething with suppressed rage at having to interrupt his life to
take me to the airport, instead of my declining his offer and taking a taxi.

On the other hand, if we gave equal weight to every imaginable side-effect
of our actions, how could we bring ourselves to get out of bed in the morning?

Best,

Bill P.

X-MIME-Autoconverted: from 8bit to quoted-printable by beasley.cisco.com id OAA16036

[From Rick Marken (970414.1430)]

Martin Taylor (970414 11:30) --

To put it more plainly, (almost) all instances of the Test involve

side-

effects that influence other perceptions of the person whose control is

being

tested.

So? This is true even when we are not Testing for controlled variables.

These side effects, if they influence controlled perceptions, may
or may not have dangerous consequences.

The side effect would have to be an insuperable disturbance to a
controlled variable.
You would probably find out very quickly (via audible complaints from
the Testee) that this was the case.

Have you actually tried Testing for controlled variables, Martin? It's
really quite safe;-)

If the Test were to be applied always and only to the CEVs of

perceptions

correctly assumed to be controlled, and the Tester could guarantee

that the

influences permitted the controlled perception to remain under

control,

there would be no ethical issue in using it, no matter how high a

perceptual

level was being tested.

I think you must be making this stuff up so that you can feel positively
self-righteous about not doing PCT research.

Best

Rick

��

[Martin Taylor 970415 10:55]

From Rick Marken (970414.1430)]

Martin Taylor (970414 11:30) --

> To put it more plainly, (almost) all instances of the Test involve
side-
>effects that influence other perceptions of the person whose control is
being
>tested.

So? This is true even when we are not Testing for controlled variables.

>These side effects, if they influence controlled perceptions, may
>or may not have dangerous consequences.

The side effect would have to be an insuperable disturbance to a
controlled variable.
You would probably find out very quickly (via audible complaints from
the Testee) that this was the case.

Have you actually tried Testing for controlled variables, Martin? It's
really quite safe;-)

I'd be a lot happier with this if you substituted "usually" or "almost
always" in place of "really."

> If the Test were to be applied always and only to the CEVs of
perceptions
> correctly assumed to be controlled, and the Tester could guarantee
that the
> influences permitted the controlled perception to remain under
control,
> there would be no ethical issue in using it, no matter how high a
perceptual
> level was being tested.

I think you must be making this stuff up so that you can feel positively
self-righteous about not doing PCT research.

Am I mistaken, or is this the same Rick Marken who never comments on
people's motives?

Martin

···

[Martin Taylor 970415 11:00]

Bill Powers (970414.1318 MST)]

Martin Taylor 970414 11:30--

>To put it more plainly, (almost) all instances of the Test involve side-
>effects that influence other perceptions of the person whose control is
>being tested. These side effects, if they influence controlled
>perceptions, may or may not have dangerous consequences.

This is true, but it's true of every interaction of one person with another.
...

The alternative to not disturbing other people's controlled variables is not
to interact with them at all.

The situation I thought we were discussion was more related to therapy, or
perhaps education or the like, where the disturbances were deliberately
introduced to Test for controlled variables in situations where (a) there
is considerable question as to what is being controlled, (b) there is the
appearance of conflicted control, and (c) there may be the appearance of
ineffective and (from the external observer's viewpoint) counterproductive
attempts to control. Under these conditions, danger is more likely and the
conseqences more likely to be long-lasting than in the examples of low-level
control you use to suggest that ethical issues don't arise.

All I'm saying is "Be careful." Rick says "No need." I'm not sure what
you are saying, but it reads a bit like the person who argues that there
is always risk if you step out of the house (or if not), and therefore
he's quite happy climbing cliffs without ropes or other equipment in a
thunderstorm.

Martin

···

>What might have happened if, in resisting the disturbance, Dag had
>suddenly and painfully pulled a shoulder muscle?

I'm not a suicidal idiot. I applied the disturbance slowly, and with a
_very_ firm grip on the wheel, and using position, not force, control. If
Dag had let go of the wheel entirely, its rim would have moved perhaps an
inch before I had control of it. It was, in fact, quite a safe experiment.

Of course, Dag might have been on the very brink of a psychotic episode, and
my tug at the wheel might have been the trigger that made him suddenly turn
the wheel the same way I was pulling and send the car careening across the
traffic lanes. In that case, we could both have been dead or worse as a
result of this demo. For all I know, simply asking Dag if he wanted to see a
really tight control system might have triggered the same episode. Maybe he
was already seething with suppressed rage at having to interrupt his life to
take me to the airport, instead of my declining his offer and taking a taxi.

On the other hand, if we gave equal weight to every imaginable side-effect
of our actions, how could we bring ourselves to get out of bed in the morning?

Best,

Bill P.

[From Bill Powers (970415.1018 MST)]

Martin Taylor 970415 11:00]

All I'm saying is "Be careful."

And all I'm saying is "be gentle." There's no need to cause any significant
perturbation in a person's controlled variables; all you need is to see
whether a gentle push results in a counterpush. You don't want the
controlled variable to change enough to bother the other person.

Rick says "No need." I'm not sure what
you are saying, but it reads a bit like the person who argues that there
is always risk if you step out of the house (or if not), and therefore
he's quite happy climbing cliffs without ropes or other equipment in a
thunderstorm.

I'm saying that if you treat people like bombs that will go off at the
slightest jiggle, you're not going to deal with them at all. There must be
some space between that attitude and just blundering around without regard
to consequences. It's possible to have practical dealings with others
without spending all your time worrying about subtle bad effects you might
be having on them. Most people can take care of themselves just as well as I
can -- even patients in therapy. Of course one judges as one goes -- we're
not talking about a policy one adopts and then sticks to regardless of the
situation.

Best,

Bill P.

[Martin Taylor 970415 13:30]

Bill Powers (970415.1018 MST)]

Martin Taylor 970415 11:00]
>All I'm saying is "Be careful."

And all I'm saying is "be gentle."

It's possible to have practical dealings with others
without spending all your time worrying about subtle bad effects you might
be having on them.

I imagine we all know people who "err" on either side of the happy medium--
the super-diffident who can annoy by trying hard not to annoy, and the
boors and psychpaths who don't even try to concern themselves with how
others are affected.

Most people can take care of themselves just as well as I
can -- even patients in therapy. Of course one judges as one goes -- we're
not talking about a policy one adopts and then sticks to regardless of the
situation.

I guess we see the same glass, half-full or half-empty. I'm glad it does
seem to be the same glass. Certainly I'd agree wholeheartedly with this
wording. If it were spoken, I'd probably think about the emphases, but
as written, I would happily have written the same.

Martin

[From Rick Marken (970417.0750 PDT)]

Bill and Mary -- Just flew over Durango last night on the way home from
a business trip. Looks like you've still got plenty of snow
down there. It looks gorgeous from 35,000 ft.

Me to Martin:

I think you must be making this stuff up so that you can feel
positively self-righteous about not doing PCT research.

Martin Taylor (970415 10:55) --

Am I mistaken, or is this the same Rick Marken who never comments on
people's motives?

I don't think I ever said that I never comment on people's motives.
Indeed, I think I _always_ comment on people's motives, (where, to me,
"motives" are controlled perceptions). What I said was that I don't (or,
at least, I try not to) judge people's ideas in terms of their motives
(what they are controlling for). I don't reject information theory and
MCT because I don't like your (or Hans') motives. I reject these ideas
because they seem to contribute nothing (actually, less than nothing
becuase I think they serve to obfuscate the basic principles of control)
to our understanding of the behavior of living systems.

Best

Rick