Misc responses

[From Bill Powers (930103.1000)]

Dag Forssell (930103.0110) --

Your long discussion of prescriptive principles has a lot of
interesting points in it, but it makes me feel a bit the way I
did in grade school when people were always giving me
descriptions of how I should be. I was sometimes willing to
listen, but right at all the crucial points there were always
words that told me that the rules would mean anything the rule-
giver wanted them to mean. Consider:

An effective person develops a good mental map of the world.
An effective person develops a reasoned, legitimate set of
wants. An effective person balances control between self and
others.

How can I tell if my mental map of the world is a "good" one? Who
gets to decide whether it's good?

Who decides whether a set of wants is "reasoned" or "legitimate?"

How much control of me vs others is considered (and by whom?)
"balanced?" What should the balance be? 90-10? 10-90? 50-50? 45-
55?

An effective person perceives data accurately and chooses
perceptions with great care.

Who decides whether a perception of data is accurate or biased?
How does a person go about choosing perceptions? This implies
that there are lots of perceptions lying about, and that we sort
of look them over and decide carefully on the best one.

An effective person is attentive and sensitive to differences
between perceptions and corresponding wants.

Isn't this true of ineffective persons too? Does anyone work in a
different way than by being sensitive (to varying degrees) to the
difference between perceptions and wants? Is maximum error
sensitivity always what is wanted?

An effective person chooses from a range of options to select
one most appropriate for the situation.

What was the range of options for the wording of this sentence
from which you chose, and how did you discover that this was the
most appropriate one for the situation? What was the situation?
I think that people very seldom do this, or need to.

An effective person uses resources to function with adequate
effort.

How much effort is "adequate?" What are "resources?"
  >An effective person directs the action so that it has a strong

effect on the actual variable selected for attention.

I agree with this. But does this mean that if you're directing
your actions so they have only a weak effect on the variable
selected for attention, all you have to do is notice this and
redirect your actions? And if that's all it takes, why haven't
you already done it?

An effective person recognizes or anticipates disturbances and
adjusts minor wants as required in time to withstand the effect
without loss of control.

Anticipating disturbances is neither necessary nor sufficient for
effective control. In fact, control systems don't have to sense
or anticipate the causes of disturbances at all. Sometimes
anticipating perceivable disturbances can improve control. It can
never substitute for control.

···

-----------------------------------------------------
The basic problem I see with these prescriptions is that they
boil behavior down to a single and not very informative measure:
"effectiveness." This is similar to characterizing people in
terms of a single number called "intelligence." It also begs the
question, "Effectiveness for WHAT?" For achieving personal
satisfaction? For the achieving the goals of a manufacturing
concern? For being a psychotherapist? For winning a war? For
getting away with a bank robbery?

I spent of couple of years of my youth in pursuit of a state of
being called "clear." Being clear meant, in Dianetics, that you
were a perfect person, with no hangups, no inner conflict,
complete eidetic recall of your whole life, and knowledge of
exactly whom to blame for all your problems (other than
yourself).

It may be understandable that I draw back from positing a Perfect
Control System Hierarchy to which all good PCT advocates aspire,
and which, if they just follow all the right rules, they can
become. What follows that is a Committee which decides whether
you have met all the requirements and can be certified Perfect,
or else inform you of your remaining shortcomings and recommend
treatment for them (and perhaps require it). The members of the
Committee, naturally, have already certified each other.

And of course on the way to the shining goal, all the aspirants
will be criticising each other for not choosing adequate
perceptions, for not balancing their control properly, for
perceiving data inaccurately, for having illegitimate wants, and
so on. Somehow that sounds to me just like what is already going
on in most of the world.

Do I really have until 2030 (age 104) to live? Or is that just
when you predict that PCT will be overthrown by a better idea?
----------------------------------------------------------------
Gary Cziko (930103.0240 GMT) --
A very nice implementation of Rick's experiment. I'm sure that
demonstrating it is much simpler than explaining it in words.

Thinking about your description of what happens, and about Rick's
experiment, I first thought that the failure of the "cursor"
movement to predict the behavior must be due to the fact that
control is good enough to bring the error down to the noise level
of the system. This is the general explanation I've been
entertaining for some time, for the phenomenon of low correlation
between behavior and the variable that it controls.

There is, however, another possibility: chaos. Most of the
control systems we've investigated are modeled best by a system
that integrates the error signal to produce output. Integrals are
known for hypersensitivity to initial conditions, one brand of
chaos. When error is near zero, any slight perturbation will lead
to an output that drifts away from the optimal setting one way or
the other, which way depending on the sense of the perturbation.
The result is that the perturbations due to noise are greatly
amplified; the system disturbs itself and these disturbances
result in a continuous wandering of the controlled variable in
the vicinity of its reference level. So the wanderings are
actually much larger than one would predict from the basic
signal-to-noise level of the neural signals. They are large
enough, in fact, to be comparable to the amount of error required
to produce the output that opposes disturbances. I suspect that
this is a better explanation of the low correlation between
action and the controlled variable.
--------------------------------------------------------------
Rick Marken and Gary Cziko --

"On Purpose" is a nice title which all three of us have
independently thought of. Another, which Rick and I both thought
of, is "Controlling Man." Sexist, but nicely ambiguous.

However, I am not going to write another book. You and Rick and
others are going to write it. When I try to write another book,
it turns out to be BCP all over again. The reason is that my
background is too limited to write anything else. _On Purpose_,
if it follows the outline of [animism, anti-animism, control]
should contain lots of references to real things that people have
believed or have been thought to have believed, with citations
and quotes and all that scholarly stuff that makes an interesting
story out of a bare recitation of principles. I can't do that.
But you and Rick can.

The responsibility for this revolution, whether you like it or
not, is passing into your hands. I am content to whisper in your
ears as I continue to work out implications of what I have
already done. The peak of the wave is yet to come, and the glory
for being the ones who ride its crest will be yours. All of
yours, speaking to all CSGers and NETters who are involved in
this project. The world will come to know of PCT through you and
your generation, not through me and mine.
If I may ask, what are you waiting for?
--------------------------------------------------------------
Oded Maler (921229) --

... you might find some inspiration in the books of the great
mathematician H. Weyl which are probably accessible. He was
concerned with the relation between the world and its
perception/representation and he conceived abstract algebra as
playing a fundamental role in this process of
'coordinatisation'.

Thats a job for a person who is a mathematician and also
understands PCT thoroughly. I'm not both. It will take a better
mathematician than me to link "coordinatisation" to the phenomena
of perception. Any volunteers?
--------------------------------------------------------------
Bill Silvert (930103) --

Roger on the "upload" directory. I guess I just missed seeing it.
My understanding now is that if I send something to "upload" you
will see it and put it in the proper place. And that Gary will
upload binary versions, which I seem unable to do.

And you have our continuing gratitude for your unselfish and very
valuable work on our behalf (yours too, I hope).
----------------------------------------------------------------
Best to all,

Bill P.

[From Rick Marken (920319)]

Martin Taylor (recently) says:

If there are two schools of thought, each with
good reason claiming that they have the truth and the other doesn't, they
are probably both right, except in that claim.

Well, I never claimed to have the truth. I claim that HPCT is the best current
explanation of a phenomenon that is currently not studied in psychology --
except obliquely -- control. So I claim that that model is better than
any other as an explanation of that phenomenon. I guess I am also saying
that most of what psychologists think of as the phenomenon called "behavior"
is actually control -- so I am saying that my model is the correct model
of behavior (as far as we can go towards correctness with current know-
ledge) and their models are, thus, wrong.

It is obvious to me that the basic ideas of PCT have to be right, just as
are those of (say) signal detection theory or information theory.

I think I don't agree -- maybe I don't understand.

my main point--that
most perceptual activity at any moment in time is passive and uncontrolled.
PCT has to acknowledge this

I don't know about "most perceptual activity at any moment in time"
being passive but has any PCT control theorist ever denied that a
great deal of our perceptual experience is not controlled? I certainly
don't deny it -- I'm perceiving all kinds of variables that I cannot
influence in any way except (if they are visual or auditory) to make them
go way (closing eyes, plugging ears). There are many that I'm not controlling
that I could control (eg, the color of my walls) and others that I will
never be able to control (the number of clouds in the sky). I love those
uncontrolled perceptions.

       --it is required by the informational arguments
and can't be wished away by faith or dogma.

You got me here? What are the informational arguments? And why should
they matter to me? (I sound like the disobediant child at passover).

There's plenty of bath water, but there are babies in it, so be careful what
you throw away.

I agree -- but I think it's more like dimes in the great salt lake. We
do try to keep an eye out for 'em (those of us who don't mind opening
our eyes in salt water; I don't mind doing it but I'd rather look for
my dimes at the bank).

I have been asked to give a 3-hour keynote address to a high-level "Autumn
Summer School" on human-computer interaction in Paris in early September.

You luck!

Naturally, I intend to take a PCT position as a base for looking at all the
different approaches to HCI that might come up, but this will probably be
a very general base, rather than dealing with specifics of the theory. I
mean that I will probably emphasize throughout that the user is doing whatever
actions he or she performs in order to be able to perceive some desired
thing. The computer output exists in order that the appropriate things
can be perceived and compared with the results of imagination as well as
with the reference signals.

Excellent! Don't forget to mention those disturbances; mistypings, different
results depending on context, etc!

My request is for CSG people to suggest themes that I might address, or
issues in HCI that have posed problems because they have not been addressed
from a PCT view, or, if it is feasible, more detailed discussions of aspects
of HCI that have been or might be addressed by PCT.

Well, it's not very detailed but I did write a little article for the
Human Factors Society Bulletin (the December 1986 issue, I believe) that
was about control theory and HCI. I thought it was great. Some others
did to -- but the conventional types spotted the heresy immediately and
suggested that I get with the "real" research program -- looking for the
effects of input variables on responses.

If you can't get a hold of a copy (call the HF Society in Santa Monica,CA
maybe) I can send you a reprint if you're interested.

On the random-walk of reorganization: there is a real degrees-of-freedom
problem here. In the 2-D demo, there is a good probability that the walk
is within (say) 60 degrees of the direction to the target. In 1-D, the
probability is 0.5, in 2-D 0.33, and goes to zero for very large
dimensionality.

Actually, we have never looked at efficiency as a function of the dimension-
ality of the space in which the value of the one dimensional controlled
is computed. Might be interesting. Maybe Bill Powers has some data

To continue the discussion of moths & S-R systems, what would be
the story about a dog that was hardwired to salivate when it
smelled food?

  in the moth case, the presumed result of the effector is to
make the sonar signal go away, while in the salivation case, the effector
does NOT make the smell of food go away. But as far as what's inside
the critter itself, the setups look pretty much the same.

Well, you are assuming your conclusion. If smell caused salivation as
you postulate and salivation has no effect on smell, then we've got an
SR system and it doesn't look like anything is controlled. I'm not
sure it actually works that way, though. First, I'm not sure that
there is a hard wired connection between smell and salivation. I
believe that the "natural stimulus" for salivation is stuff in the
mouth that tends to absorb liquid. So I'd say sensed "wetness" is the
"stimulus" for salivation -- that is my assumption; I believe that you
might test this by seeing if a dog salivates to smell when it's mouth is
full of water or if it salivates when an orderless dry substance is placed
in the mouth. I predict no to the first and yes to the second.

If smell is involved in salivation (without training) then I would
have to be convinced that salivation really has no effect on the
smell signals -- it's plausible that it does have such an effect though;
salivation probabaly deceases the amount of "smell chemicals" that go
back up into the nasal passages from food in the mouth.

And furthermore, is it so obvious that the moth's system is actually
making the signal go away?

No. But falling doesn't have to make the signal go away --
it just has to affect it -- which it obviously does.

What it might be doing is making the moth
undetectable by getting it on the ground as fast as possible

Of course it DOES. But that's not what the moth is doing. That is an
Irrelevant Side effect (ISE -- there, got it right this time) that happens
to be VERY relevant to the moth's survival. The moth just knows nothing
about it. If the falling put the moth into the hand of a friendly bug
researcher (like Bill Powers and his pet beetle) that would be fine with
the moth too -- just get that damn signal outta here, sayeth the moth's
control system.

So it seems to me that exactly the same hunk of circuitry might be
appropriately regarded as an ECS or an S-R system, depending on
whether there is an R-S connection that is an important aspect of
the functioning of the system.

You betcha!! Now think about how much you can "do" without influencing
one or another of your sensory inputs. When you realize that that amount
is almost precisely ZERO then you have entered the PCT zone.

Regards

Rick

···

on that? Avery Andrews (920319) says:

     **************************************************************

Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
Internet:marken@aerospace.aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)

[Martin Taylor 920320 14:30]
(Rick Marken 920319)
(I don't know why I am doing this, with so many deadlines so close...but
this is more fun. Conflict...resolution)

Martin Taylor (recently) says:

If there are two schools of thought, each with
good reason claiming that they have the truth and the other doesn't, they
are probably both right, except in that claim.

Well, I never claimed to have the truth. I claim that HPCT is the best current
explanation of a phenomenon that is currently not studied in psychology --
except obliquely -- control. So I claim that that model is better than
any other as an explanation of that phenomenon. I guess I am also saying
that most of what psychologists think of as the phenomenon called "behavior"
is actually control -- so I am saying that my model is the correct model
of behavior (as far as we can go towards correctness with current know-
ledge) and their models are, thus, wrong.

Yep, that's the attitude I was taught about in grad school. Think about
how much can be properly treated by "classical" methods even if their
model is "wrong." Think about the fact (about which we are believed to
agree) that a control system with a fixed reference works like an S-R
system. Think about the many reference levels that can be assumed to be fixed
by the agreement of the subject to participate in an experiment. And then
think carefully about whether their "wrong" models might not be a subset
of your model, even if neither they nor you find it intuitively to be so.

I'm not claiming you are wrong, nor that "they" are. But I have learned
to be wary about the fragility of structures that fit together in lots of
ways. Even if the foundations are insecure, the superstructure often holds
together when they are replaced by better foundations. Newtonian mechanics
does not fail because we now understand that kinematics is all a question
of geometry, not mystical "forces."

It is obvious to me that the basic ideas of PCT have to be right, just as
are those of (say) signal detection theory or information theory.

I think I don't agree -- maybe I don't understand.

All I'm saying is that you can't legislate pi to be 3. To disagree that
the basic ideas of signal detection theory or information theory are
correct is equivalent. It is only an opinion, though, that the basic
ideas of PCT are right. I hold that opinion, but it could be changed
by experimental evidence. Pi cannot be set to 3 by experimental evidence.

Martin