defending PCT, Program Control

CSGnet Archive CSG1998
1

[From Rick Marken (980722.2220)]

Bruce Nevin (980722.1123 EDT)--

Your answer to "how can we do it better" is "give up."

Not quite. My answer ia "lower the gain". I am still controlling
for PCT accuracy (I'm not giving up); I'm just trying to
control less "tightly".

I don't think this is a useful answer.

I think this is one of the most useful answers PCT has to give
regarding social interactions. But you are certainly entitled
to your opinion (see how low gain I am now;-))

That question is the most important in this post: Can you go up
a level to what is behind the desire to correct people's
misstatements?

I suppose so. Actually, I think I did. That's why I don't feel
any conflict between my desire to "control for accuracy of PCT"
and to "control for not hurting others". I went up and saw that
I can control for both; I just can't do this _and_ control for
_either_ with particularly high gain.

My going up a level, by the way, doesn't solve the far more
interesting conflict that exists between me and those who make
what I see as inaccurate statements about PCT. That conflict
results from the fact that those other people and me are
controlling the same perceptual variable (nature of PCT) relative
to somewhat different references. There is really no solution to
_that_ conflict (other than one of us giving up). I hope that
lowering the gain of my control of PCT accuracy will keep these
conflicts from getting too far out of control.

Part of the problem may be simply jumping to conclusions as to
the intentions of the other person

I don't see how this is relevant to the problem of interpersonal
conflicts over PCT. If someone says (as someone once did) "control
of perception can be described as 'perception drawing forth
behavior'" then what is the relevance of the utterer's intentions?
I can't conjure any intention behind such a statement that would
keep it from being a disturbance to my perception of how control
systems work. The best I can do is try to lighten up on the
corrective action -- maybe explain what's wrong in a paragraph
or two and then drop it. That's what would happen if I could
lower my PCT accuracy control gain.

Bruce Abbott (980722.1455 EST) --

Actually, I've understood the concept of controlled variable for
a long time.

Perhaps. I guess I get confused about this when you talk about
all the controlled variables that have been discovered by
conventional psychologists. I cannot find any (not even the
LOT variable discovered by McBeath et al in their baseball
study; they just didn't do the appropriate tests to determine
whether or not this variable was actually under control -- and
it's not).

The question was, how do you set a reference for a program
perception if you do not know what the program is going to be
at the outset?

I don't understand this question. According to the PCT model
you set a reference for a program perception in the same way that
you set a reference for any perception. When you set a reference
for cursor position, for example, you don't have to know what
the cursor position is going to be when you set that reference
to a particular value. The point of setting a reference is
to specify the perception (position, program) you _will_ have
if you can control that perceptual variable.

I didn't ask whether you think all behavior is the control of
perception, I asked whether you thought that we normally create
detailed specifications for a program (program references)
prior to carrying them out.

I think that what we "normally" do when we control a program is
set a single specification (the value of the reference signal) for
the level of the program variable we want. I think the program
perception is represented by the time varying value of a perceptual
signal -- the output of a neural program perception function.

Me:

If we control programs (and I have shown experimentally that at
least one person -- me -- can control programs)

Bruce:

Please describe said experiment.

I did it in the context of the experiments I did for the
"Hierarchical Behavior of Perception" paper, which is at

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

I was testing to see whether a program had to occur more slowly
than a sequence in order to be controlled; the rationale for
the experiment is the same as that used in the "Hierarchy of
Perception and Control" demo at my demo page:

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

A sequence of numbers (N) alternately occured at two different
positions on the screen. The program was something like

if N>5 then N = Odd else N = Even

So if 7 appeared on the left then an odd number would follow
on the right if the program was happening. If the number on the
right was 3 then the program is happening. Now the 3 is N so
the number that occurs next (on the left) should be even.
So the following sequence of numbers is consistent with
the program if N>5 then N = Odd else N = Even; 7,3,2,8,1,2...

At some point the sequence of numbers would start violating
the program; the numbers would be generated by a new program,
like "if N>5 then N = Even else N = Odd". The change to the
new program is a _disturbance_ that has an effect on the
sequence of numbers that is seen. The subject can press a button
to restore the original program (pressing the button always
changes the program to the one that is currently _not_ running).
The measure of control was proportion of time that the original
program of numbers occurs. When the numbers appear _slowly_ enough
(I think it was about 1 number per 2 seconds) it was clearly
possible to control the program (keep it at "if N>5 then N = Odd
else N = Even"); the subject can keep the original program going
over 80% of the time. (As I recall, this is about the best that
could be done because there is a substantial reaction time; the
time it takes between the change in the program and the pressing
of the button to restore the original).

One interesting feature of this experiment was that it showed that
it is possible to control a program without actually generating
"programmatic" actions; the program _perception_ was maintained
by simply pressing a button. Another interesting result was that
a program of numbers had to _occur_ much more slowly than a
sequence of numbers in order for the subject to control it.
This "speed limit" is clearly imposed by the perceptual system
(it takes longer to perceive that a particular program is happening
than it does to perceive that a particular sequence is happening)
not by the output system (which was a button press in both cases).

Also, the question was not whether we can control programs.
Certainly we can. The question was, is this the normal
course of business?

I think the normal course of business is to set a reference
for a particular program perception (like a familiar recipe)
and control the perception of that program. I think what you
are describing as "planning in advance" is the planning that
occurs _in imagination_ before (or even during) trying to control
for unfamiliar programs.

. . . And again, the question was whether you typically have a
worked-out-in-advance plan

No. I don't think it's a worked out in advance plan; I think it's
a reference for a program perception.

Best

Rick

···

--

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

2

[From Bruce Gregory (980723.0623 EDT)]

Rick Marken (980722.2220)

I think the normal course of business is to set a reference
for a particular program perception (like a familiar recipe)
and control the perception of that program.

I would have thought that a recipe was a sequence rather than program.

Bruce Gregory

3

[From Bruce Gregory ((80723.0947 EDT)]

Bruce Nevin (980723.0726 EDT)

To Rick:

How do you know that their misuse of words indicates that they are
controlling a perception of the nature of PCT wrongly? Is this
all you need
to Test for the controlled variable? Have I been making this mistake for
the past 8-9 years? (No, I haven't.)

Good point. In my experience, Rick never tests for the controlled variable
when he senses that someone does not understand PCT. In fact, once Rick
makes this judgement there is almost nothing the person can do or say that
will change his mind. If Rick practiced what he preaches (literally) with
regard to the Test, CSGnet might look very different.

Bruce Gregory

4

[From Bruce Nevin (980723.0726 EDT)]

I'm just about to catch the plane for my brother's wedding, only time for
one note, sorry.

Rick Marken (980722.2220)--

That conflict
results from the fact that those other people and me are
controlling the same perceptual variable (nature of PCT) relative
to somewhat different references. There is really no solution to
_that_ conflict (other than one of us giving up). I hope that
lowering the gain of my control of PCT accuracy will keep these
conflicts from getting too far out of control.

This is great as far as it goes. I'm not sure you can sustain it for
reasons I said (you care more about PCT, and so you ought), but maybe you can.

But when you say

That conflict
results from the fact that those other people and me are
controlling the same perceptual variable (nature of PCT) relative
to somewhat different references.

You are begging the question

Part of the problem may be simply jumping to conclusions as to
the intentions of the other person

How do you know that their misuse of words indicates that they are
controlling a perception of the nature of PCT wrongly? Is this all you need
to Test for the controlled variable? Have I been making this mistake for
the past 8-9 years? (No, I haven't.)

When I said "monitoring for disturbances" and you corrected me, was I
controlling a perception of PCT different from the one you control, or was
I using words carelessly? It is this literalism that I am questioning. It
makes the whole thing look like a matter of ideological correctness, which
it is not.

  Bruce Nevin

5

[From Bruce Gregory (980723.1451 EDT)]

Rick Marken (980722.2220)

I suppose so. Actually, I think I did. That's why I don't feel
any conflict between my desire to "control for accuracy of PCT"
and to "control for not hurting others". I went up and saw that
I can control for both; I just can't do this _and_ control for
_either_ with particularly high gain.

Could you say more about this? How does poor control at one level enhance
your ability to control at a higher level? I can't find anything about this
phenomenon in B:CP.

Bruce Gregory

6

[From Rick Marken (980723.2050)]

Bruce Nevin (980723.0726 EDT) --

But when you say

>That conflict results from the fact that those other people
> and me are controlling the same perceptual variable (nature

of PCT) relative to somewhat different references.

You are begging the question

What question am I begging? I thought I was just describing a
conflict.

When I said "monitoring for disturbances" and you corrected me,
was I controlling a perception of PCT different from the one you
control, or was I using words carelessly?

Apparently you were using words carelessly. I know this because
when I corrected you (explaining that control systems monitor
perceptions, not disturbances) you accepted the correction
immediately. You would not have done this if you were controlling
for the idea that control involves "monitoring for disturbances".

Bruce Nevin (980723.0726 EDT)

How do you know that their misuse of words indicates that they
are controlling a perception of the nature of PCT wrongly? Is
this all you need to Test for the controlled variable? Have I
been making this mistake for the past 8-9 years? (No, I haven't.)

Bruce Gregory ((80723.0947 EDT)--

Good point. In my experience, Rick never tests for the controlled
variable when he senses that someone does not understand PCT. In
fact, once Rick makes this judgement there is almost nothing the
person can do or say that will change his mind.

Actually, I do test to determine whether someone is controlling
for a perception of PCT that is different from mine. It's really
very easy; I just watch for _continual_ resistance to my attempts
to correct them. For example, in the "monitoring for disturbances"
case above, Bruce Nevin immediately accepted my correction; so
I could see right away that Bruce was not controlling for an
(incorrect) perception of control as a process of monitoring for
disturbances. When Bruce first said this I may have judged that
he did not understand this aspect of PCT; but Bruce immediately
agreed with my correction (when I noted that control is control
of perception) so I changed my mind and concluded that Bruce does,
indeed, understand this aspect of PCT; what Bruce said about
"monitoring for disturbances" must have just been a careless
use of words So there is something a person can do to change
my mind if I judge that he or she doesn't understanding of PCT -- he
or she can agree with my corrections;-)

It's pretty obvious when someone is controlling for a perception
of PCT that is different from mine (and Bill's); when that happens
there is _continual_ resistance to my corrections. This is what
happened when you proposed that "control of perception can be
described as 'perception drawing forth behavior'". Both Bill and I
tried to correct your proposal but you continually (and vigorously)
resisted these corrections. I think it was pretty obvious that you
were controlling for a perception of PCT that was quite different
from Bill's and mine. I suppose this still could just be a language
problem; maybe what you mean by "calling forth" is what we
mean by "control of perception". But if this is the case, then
communication is hopeless anyway. I would have to assume that
whatever you said about PCT reflected the same understanding
as mine. So there would really be no need for discussion at all.

I guess I still cling to the belief that communication works.
It works for me with Bill and Mary Powers, my wife and my kids.
Maybe those are the only people in the world who use language
as I do; if so, that's plenty.

Bruce Abbott (980723.0910 EST) --

In my studies on preference for signaled over unsignaled shock,
the rats definitely were acting to oppose the disturbance created
by the apparatus when it moved the animals back into the unsignaled
condition at the end of each changeover period. They immediately
pressed the lever to restore the signaled condition. Simple as
this demonstration is, for you refuse to recognize it as a Test
for the controlled variable. I think it pains you to admit it.

I will be happy to admit that your studies of preference involved
aspects of the test for the controlled variable. But it would pain
me to admit that they _were_ a test for the controlled variable
only because they were not. You can tell that they were not a
test for the controlled variable because we still have no idea
what variable (if _any_) was being controlled in those studies.

To set such a reference (for a program), don't you have to have
a detailed specification for what is to occur at each step?
Isn't that what a "reference for a program) is?

Not in the HPCT model. A reference for a program is just a signal;
it specifies the state of a perceptual variable, which in this
case represents the fact that a particular program is happening.

So there is a perceptual function dedicated to perceiving when
a program is present in the inputs?

Yes. And this seems to be what is happening in the program
control experiment I described. The perceptual signal is in
one state (say, 1) when one program is occuring at the input and
in another state (say, 0) when another program is occuring. If
you want program 1 to be occuring you set the reference signal
to 1; when the perceptual signal goes to 0 there is an error
and the subject does what is necessary (presses the button in
my experiment) to restore the desired program perception.

It sounds like trying to force what is an inherently serial
process into a simultaneous model.

See p. 143 and 144 of B:CP for a model of a perceptual function
that can transform a serial input (a word that that occurs over
time) into a scalar perceptual signal. In fact, I think it might
help if you just re-read B:CP in its entirety. I bet you will find
that many of the things that seemed difficult when you first read
it will make more sense now.

Bruce Gregory (980723.0623 EDT)--

I would have thought that a recipe was a sequence rather than
program.

A recipe is a _contingent_ sequence. There are branch points:
bake until a crust appears, stir until thick, etc. These are
basically program loops; you keep baking (or stirring) until
the test condition (crust = true; thick = true) is met. Also,
the components of a recipe can often be carried out in parallel;
you can bake _while_ you look for the ingredients for the next
step, for example. But you can always perceive the fact that you
are in the process of following a particular recipe _while you are
actually carrying out the recipe program_. The fact that you
can perceive that "baking a pie", say, is happening (just as you
can perceive that "if N>5 then N=odd else N = even" is happening --
when it is -- in my program control experiment) shows that we
have some means of constructing a perception of a program in
progress. HPCT hypothesizes that program perceptions are perceptual
_signals_ and that these perceptual signals are constructed by
perceptual functions.

Best

Rick

···

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

7

[From Bruce Gregory (980724.0445 EDT)]

Rick Marken (980723.2050)

A recipe is a _contingent_ sequence. There are branch points:
bake until a crust appears, stir until thick, etc. These are
basically program loops; you keep baking (or stirring) until
the test condition (crust = true; thick = true) is met.

Ah, yes. TOTE.

Bruce Gregory

p.s. I have forsworn the "calls you forth" and "many worlds" heresies.
You'll need new examples.

8

[From Rick Marken (980724.2120)]

Me:

A recipe is a _contingent_ sequence. There are branch points:
bake until a crust appears, stir until thick, etc. These are
basically program loops; you keep baking (or stirring) until
the test condition (crust = true; thick = true) is met.

Bruce Gregory (980724.0445 EDT) --

Ah, yes. TOTE.

Well, a side effect of the operation of a TOTE, yes -- if a TOTE
could actually behave. Unforunately, Miller, Galanter and Pribrum
(MGP) understood neither the _phenomenon_ nor the theory of control.
Their TOTE unit was really just a verbal description of what
they saw as "goal oriented behavior". The main problem with the
TOTE unit as an explanation of goal oriented behavior is that
it views goal achievement as a sequential process: Test _then_
Operate _then_ Test then (if the Test is passed) Exit. Such a system
can work only if the loop gain is < 1.0 (see Powers, 1978); when
the loop gain is >= 1.0 the system goes into positive feedback
oscillations. A system with a loop gain of <1.0 can barely control
(achieve its goals in a disturbance prone environment) at all.

Not only did MGP get the dynamics of control wrong with the TOTE
"model", they also didn't notice what is most important about
a control model when it works: it controls s perception. This
means that if, say, a recipe is actually under control (if
carrying out the recipe is the system's goal) then the system
must be able to continuously monitor a _perception_ of the recipe
and act to keep that perception in the intended (goal) state.

This is why I say that the recipe is a side effect of the operation
of a TOTE -- at least, as the TOTE was conceived of by MGP.
According to MGP, what you see when you see a person carrying out
a recipe is the operation of a hierarchy of TOTE units; a TOTE
unit that brings the crust to the goal state; a TOTE unit that
brings the lemon filling to the goal thickness, etc. According to
this conception of behavior, the recipe program itself is not under
control; it is just a side effect of the operation of individual
TOTE units.

Because PCT unambigously calls our attention to the fact that
behavior is the _control of perception_ we can see that, if
carrying out the recipe program is itself under control, then
that program must (somehow) be perceived and continuously
compared to a reference specification (that is of the same type
as the perception; in PCT, both are scalar signals). That is
the genius behind Bill Powers' conception of a hierarchy of
perceptual variables; it is a model that accounts for the fact that
people seem to be able to control not only muscle tension, arm
positions and body movement but, also, sequences (melodies),
events (golf swings), programs (pin the king), principles (control
the center) and system concepts (member of a chess club). You can't
control these things unless you can perceive them. The hierarchy
of perception says you _can_ perceive and, hence, control these
variables.

Best

Rick

···

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

9

[From Bruce Gregory ((80725.0836 EDT)]

Rick Marken (980724.2120)

Because PCT unambiguously calls our attention to the fact that
behavior is the _control of perception_ we can see that, if
carrying out the recipe program is itself under control, then
that program must (somehow) be perceived and continuously
compared to a reference specification (that is of the same type
as the perception; in PCT, both are scalar signals). That is
the genius behind Bill Powers' conception of a hierarchy of
perceptual variables; it is a model that accounts for the fact that
people seem to be able to control not only muscle tension, arm
positions and body movement but, also, sequences (melodies),
events (golf swings), programs (pin the king), principles (control
the center) and system concepts (member of a chess club). You can't
control these things unless you can perceive them. The hierarchy
of perception says you _can_ perceive and, hence, control these
variables.

It appears that when you adopt a Plan, the level that activated the plan no
longer experiences the error that it did before the Plan was adopted, even
though perception that the upper level goal has been reached may be many
years away. This seems to me at least a very good reason for paying
attention to the program level, as MGP do. I don't disagree with anything
you say.

Bruce Gregory