Free Will, In the blind , MCT

CSGnet Archive CSG1997
1

[From Rick Marken (970129.0930)]

Rupert Young (970129.1100 UT)

I feel it is more positive to forget about free will and personal
responsibilty and consider that the the behaviour of many people
is shaped by their experiences in a negative way.

I agree that it's a good idea to forget about free will. However, I
think it is also a good idea to understand what personal responsibility
means from a PCT perspective. It simply means that we are responsible
for the perceptions we control; we are not responsible for uncontrolled
side effects of our controlling.

The idea that people are "shaped" by their experiences is also brought
into question by PCT. People are not really shaped by their
experiences; people spend most of their time _shaping_ their
experiences to match their autonomously selected specifications for
those experiences. People do this shaping within the constraints of a
real world that includes other controllers who are often assholes.
But the assholes of the world (like drug prohibitionists and wealth
accumulators) are not shaping the criminals; they are just making it
hard for people to shape their own experiences in ways that these
assholes find acceptable;-)

Hans Blom (970129b) -

what is "appropriate gain"?

In an integral controller o = o + k+(r-p) * dt. k is the gain.
"Appropriate gain" is any value of k that allows the controller
to operate properly (eg., without oscillation).

You also invert what I have said about control "in the blind"...
It is an observation _that_ our eyes are sometimes (briefly)
closed while walking...That is a phenomenon that needs an
explanation.

This phenomenon has nothing to do with control "in the blind".
We don't know that a person is controlling "in the blind" when
he walks with his eyes closed.

I translate this problem into the question "how can a controller best
remain in control if its feedback information is occasionally
briefly missing?"

But we don't know that the controller's "feedback information" is
occasionally missing. All we know is that the person is walking with
his eyes closed. This may be irrelevant to the controlling that was
going on before the eyes were closed. The "feedback information"
that was controlled before the eyes were closed may have been all
kinesthetic.

You are making up a fact (control without feedback information) that
you have not observed. I think you are doing this because such a
fact would justify the application of MCT to behavior. If you were
interested in what is _really_ going on when people "control without
feedback" you would be working on Bill's experiment that tests what
_actually_ happens to control of a variable when perception of that
variable is briefly lost. But facts seems to be of no interest to you.
What you call facts are really "just so" stories. It's getting pretty
boring, Hans.

Bill Powers (970129.0430 MST) --

My, you do carry on:-)

It doesn't do the reputation of PCT any good to make silly
statements

I hasn't hurt the reputation of MCT or Artificial Life;-)

Rick, I think your impression that the nature of the output function
doesn't matter, beyond its gain and sign

I didn't mean to imply that the nature of the output function doesn't
matter. And I never said that _only_ the sign and gain of the output
function mattered. But I'm happy to take the heat.

In mitigation of your impetuous overstatement :-),

Still crazy after all these years;-)

I can see that these MCT types are going to say that a PCT (ie. normal)
controller contains an implicit "model of the environment" no
matter what. For example, our simple integral controllers that
contain only gain and sign parameters can still control in a rather
wide range of environmental situations (different feedback functions).
Nevertheless, MCT types will say that the _constant_ settings of
the gain and sign parameters are an implicit model of all these
different environmental functions. There's no stopping them;-)

If it's "silly" to say that the gain and sign of the output function
must be set to appropriate values to allow the system to be stable,
what is it when MCT types say that these "appropriate settings"
represent an implicit model of every environment in which
they work? I have a suggestion;-)

Best

Rick

2

[Hans Blom, 970130]

(Rick Marken (970129.0930))

what is "appropriate gain"?

In an integral controller o = o + k+(r-p) * dt. k is the gain.
"Appropriate gain" is any value of k that allows the controller to
operate properly (eg., without oscillation).

Your term "properly" doesn't help much, but "without oscillation" is
far more descriptive than "appropriate"; it can have only two values.
So it will be easy to give the gain a value such that no oscillation
is observed. But that is not enough: when the controller has a gain
of zero, I'm pretty sure that there will be no oscillations. Yet it
will not control properly. So the question remains: can you be more
explicit about the value of the gain that results in proper operation
of the controller?

I hope I'm not asking too much...

Greetings,

Hans

3

[From
[From Bruce Gregory (970130.1000 EST)]

Rick Marken (970129.0930)

I agree that it's a good idea to forget about free will. However, I
think it is also a good idea to understand what personal responsibility
means from a PCT perspective. It simply means that we are responsible
for the perceptions we control; we are not responsible for uncontrolled
side effects of our controlling.

I still find this hard to understand. Someone cuts me off in
traffic because she is controlling for getting to her
appointment on time. I am unable to avoid hitting her. Since
the accident is the result of uncontrolled side effects of her
controlling, she is not responsible for it. Enlighten me, Oh
Great Marken, I beseech thee.

Bruce Gregory

4

[From Rick Marken (970130.1830)]

Me:

"Appropriate gain" is any value of k that allows the controller to
operate properly (eg., without oscillation).

Hans Blom (970130) --

can you be more explicit about the value of the gain that
results in proper operation of the controller?

I don't see what your problem is. From my experience playing
with control models I have found that output amplification
(that's what k really is) matters. A simple, integral control
system will not keep it's perception accurately tracking its
reference input if k is not set appropriately. The "appropriateness" of
the value of k depends on 1) other amplifications that occur in the loop
2) the nature of the functions in the loop and 3) the speed with which
loop variables change over time. I'm sure there are equations that make
it possible for an engineer to compute the "optimal" value of k and
other control parameters (values that makes it possible to meet some
criterion of control) given knowledge of these various aspects of the
control loop. Put I can get a control system to show "pretty darn good
control" of its perceptual input by just fiddling with k until the
controller works. It's really not hard to do. And control systems are
pretty robust little puppies; they control
"pretty darn well" with a fairly wide range of values of k.

I think the reorganization system makes control systems work in
essentially the same way I do it. It keeps fiddling with the
parameters of the control system until the system is controlling "pretty
darn well" at which point it stops messing around. If it
becomes evident (via increases in average error over time?) that
control is deteriorating, the reorganizing system starts fiddling with
parameters again, just as I would.

Best

Rick

5

[Hans Blom, 970131b]

(Rick Marken (970130.1830))

can you be more explicit about the value of the gain that
results in proper operation of the controller?

I don't see what your problem is.

My professional problem is to design controllers that operate
properly, and to give them that gain value (amongst other things) so
that they operate properly. So I need a theory/tool/algorithm to find
explicit values.

From my experience playing with control models I have found that
output amplification (that's what k really is) matters.

From my experience I have found that much more matters.

A simple, integral control system will not keep it's perception
accurately tracking its reference input if k is not set
appropriately. The "appropriateness" of the value of k depends on 1)
other amplifications that occur in the loop 2) the nature of the
functions in the loop and 3) the speed with which loop variables
change over time.

Hey, you're addressing my problem!

I'm sure there are equations that make it possible for an engineer
to compute the "optimal" value of k and other control parameters
(values that makes it possible to meet some criterion of control)
given knowledge of these various aspects of the control loop.

Yes, there are. The simplest ones do it for cases where the world in
which the controller lives does not change. But often it does. Then
we do not need explicit proper parameter values, but an _algorithm_
that computes and adjusts those parameter values on-line, real-time,
during the controller's actual operation. Those "equations" are far
more intricate, and your approach doesn't work anymore:

But I can get a control system to show "pretty darn good control" of
its perceptual input by just fiddling with k until the controller
works. It's really not hard to do. And control systems are pretty
robust little puppies; they control "pretty darn well" with a fairly
wide range of values of k.

But what if the range of values of k (and other parameters) is far
wider than "fairly wide"?

I think the reorganization system makes control systems work in
essentially the same way I do it. It keeps fiddling with the
parameters of the control system until the system is controlling
"pretty darn well" at which point it stops messing around. If it
becomes evident (via increases in average error over time?) that
control is deteriorating, the reorganizing system starts fiddling
with parameters again, just as I would.

Rick, you would have solved my problem if you could describe
explicitly, in an algorithm or in formulas, just _how_ you fiddle
those knobs, depending on _what_ exactly. The recipe "just do it
until it works" is far too fuzzy to be a basis for an implementation.

But otherwise I think you _do_ see exactly what my problem is!

Greetings,

Hans