SR and "sorry"; control of meaning; social control

[From Bill Powers (921024.0830)]

Oded Maler (921023) --

... you can let someone
play with the computer such that he/she has to respond in a certain
way to some stimulus (sic.) appearing on the screen (e.g., track it
with the cursor/mouse) within some time interval.

I'm not sure how you see the idea of stimulus and response entering
into tracking a cursor. In our experiments there is always a
disturbance, so "responding to the stimulus in a certain way" will not
result in tracking. There's no response that corresponds to each
target position such that the result will place the cursor at the
target. If you consider the cursor as part of the stimulus, then it's
neither a dependent nor an independent variable.

It seems to me that all the experimental suggestions you made are the
S-R sort of experiment, in which you vary an independent variable and
look for a correlated change in the dependent variable. Am I
misinterpreting your description?

Perhaps I didn't make my question clear. When I asked what kind of
experiment you could do to determine that the "sorry" loop is lower in
the hierarchy than the conscious loop, I was only asking what criteria
you would use to determine that saying "sorry" was lower in the
hierarchy than some conscious kind of behavior, such as evaluating
whether saying sorry is appropriate.

But I'm not an experimentalist.

You must have some sort of actual behavior in mind, and use some way
of interpreting what you observe. How else can any theoretical concept
have meaning?

···

-------------------------------------------------------------------
Gary Cziko (921024.0300) --

Seems to me you mentioned, and Rick accepted, an idea about making
Blind Men a joint paper. Somehow I think that Rick's brilliant idea,
coupled with your brilliant strategy, would do something like square
the importance of the paper. As well as doing what Bruce Nevin
suggested -- trying the same paper with a different slant and leaving
the editor the choice between publishing and supporting contradictory
judgments of the same idea.
--------------------------------------------------------------------
Avery Andrews (921024.1519) --

Suppose that comprehension is effected by a very robust, error-
tolerant, context- & content-= sensitive system. If it fails to
find a suitable meaning for something that gets said, that
generates an error-signal that prompts some reorganization (this is
off the main point, & just included to provide a wider context).

So far so good: the higher system controls for meaning by finding A
lower-level sentence that produces that meaning. Reorganization takes
place when no sentence with a suitable meaning can be found.

But when the comprehension system does find a reasonable meaning
for an utterance, something else happens: that meaning is sent to
the production system, which in effect returns a list of all the
different ways it might express the same meaning (assuming the
perspective of the original utterer).

This gets a little awkward from the modeling standpoint -- sending a
meaning to the production system implies that the production system
knows something about meanings. But it's the higher system that's
concerned with meanings, isn't it?

Suppose we say, instead, that there is more than one higher system
concerned with meaning. If many higher systems are operating in
parallel, each concerned with a different facet of meaning derivable
from the same lower-order sentence, you get an effect similar to what
you describe: one sentence, many possible meanings. But now the
meanings stay at the level concerned with meanings instead of
sentences, and the sentence-level systems don't have to deal with
meanings.

If the original utterance doesn't come up on this list, you get a
big error signal promoting reorganization of the production system,
but you also get an error signal if there turns out to be more than
one way of producing the meaning. E.g., if the current production
system provides alternatives to what was said.

The multiple meaning systems put a slightly different slant on this.
Each meaning system wants to perceive a different meaning. The
sentence that is found produces meanings satisfying one or more
meaning-control systems, but leaves others with errors.

If the original utterance doesn't come up on this list, you get a
big error signal promoting reorganization of the production system,
but you also get an error signal if there turns out to be more than
one way of producing the meaning. E.g., if the current production
system provides alternatives to what was said.

If there are meaning-control systems yet unsatisfied, reorganization
(or at least a continuation of the search strategy, which need not be
random) continues. If the current production system produces UNWANTED
meanings, it can CAUSE error in one or more meaning-control systems.

So if your grammar allows you to say:

John donated the money to the fund.
or:
John donated the fund the money.

to mean the same thing, there will be some error-signal generated
each time the first is used, which can be eliminated by altering
the grammar to exclude the second.

Under the meaning-control hypothesis, the second would be eliminated
only if it produced a meaning that caused an error in some meaning-
control system. For example, there might be a principle that you
mention the most important subject first. If the question is "What
happened to the money?" the answer would be stated "John donated the
money (to the fund)." But if the question is "Who got the money?" the
answer would be "John donated the fund the money, not the Treasury."
To answer the first question the second way would be to violate the
principle by implying that the fund is more relevant to the question
than the money. But if the question were "What did John do?" then
either answer would supply the wanted meaning, and there would be no
reason to eliminate either one.

So the non-occurrence of the second (a piece of `negative
evidence') becomes accessible to the learning system.

Control theory handles the non-occurrance in terms of a reference
signal that demands the occurrance. When there is a reference signal
spcifying that some perception occur, then as soon as the reference
signal is set there is an error, which persists until the perception
occurs. In this case it is not occurrance of the sentence that
matters, but of the meaning.

A further prediction is that the kind of optionality above will be
an inherently unstable feature of languages: if two such forms are
used with no discernable difference in meaning, the language-
acquisition systems of the speakers will be constantly reorganizing
without being able to find an error-free configuration, &
presumably at some point one of the forms will other drop out, or
they will acquire subtly different meanings (like the emergence of
magnetic domains in cooling iron, perhaps).

The implication is that the mature speaker will come to express the
same meanings in the same words all of the time, never paraphrasing or
varying the wording. I'm not sure you would want to maintain that.
Under the multiple-meaning-control version of your hypothesis, the
choice of wording will be such as to satisfy all of the meaning-
control systems that are involved, if possible. This takes care of
your conjecture that different wordings will have subtly different
meanings -- by which I presume you mean in addition to the main and
obvious meaning of all the different wordings. Reorganization would
take place only if no wording were available (through a learned search
process) that could make all the alternative meanings fit all of the
meaning reference levels at once. I think that a control-system model
would work better than an annealing model based on principles of
magnetism!
---------------------------------------------------------------
Greg Williams (921024) --

In short, we have to include the physics-
model in the discussion. This is necessary even to suppose that
other people exist, for any observer.

This "problem" isn't limited to "social interactions": it is the
same for control that involves "inanimate objects." Yes, it's all
perception -- including the faith that we aren't solipsists (always
a possibility, but one I don't take very seriously -- that's my
ideology).

Right. The non-solipsism assumption is part of the model.

On the basis of the PCT model only -- no ideology -- both examples
simply count as instances of a party controlling some of their own
perceptions which depend on some of another's actions. The first
example corresponds to #1 in my summary (NOT disturbing another's
controlled variables, but altering the environment so as to >perceive

what you want which depends on the other). The second >example
corresponds to #2 in my summary (disturbing another's >controlled
variables so as to perceive what you want which depends >on the
other).

Let's see, I'm not clear now on #1. If you alter your environment to
perceive what you want that depends on another's actions, but altering
the environment has no effect on that other, this must mean that
you're altering something that only you have a reference level for and
that does NOT result in any change in the other's outputs. Would a
valid example be looking at a distant person through binoculars? I'm
trying to make sure that I understand your use of "not disturbing" --
whether you mean not applying a disturbance at all, or applying a
disturbance but within the other's capacity to resist, or only
refraining from applying so large a disturbance that some controlled
variable is materially altered.

The PCT model (which I thought we were trying to get back to)
doesn't require any talk of "social interactions," as it doesn't
need any talk of "importance."

I think we need the term "interaction" to mean the situation where two
systems simultaneously disturb each other's controlled variables in
the process of controlling their own. I would defend the use of
"social" interaction as specifying that the interaction is taking
place between equal organisms (with significant loop gains) rather
than between an organism and a rock (which has no loop gain). The
nature of these two interactions is quite different.

It's true that "interaction" is not a term from PCT. But when we look
at relationships between organisms, the rules of PCT no longer apply:
there is no superordinate system controlling for any form in the
interaction. We can still model social systems, but the approach has
to be more that of the System Dynamics people, with organisms and
their PCT properties being the building blocks along with non-living
aspects of the environment and their properties.

Talking about control of one's perceptions which depend on >another's

actions requires taking no more nor less of an >epistemological
position than does talking about control of one's >perceptions which
depend on the "motion" of a "rock."

True. In both cases we have to assume an environment in which things
happen that are not necessarily represented -- realized -- in
perception. It seems necessary to be able to distinguish between what
a person perceives to be happening and what is "actually" happening.
Pinning down the meaning of "actually" is pretty difficult, unless we
confine ourselves to the physical-world model where disagreements are
minimized.

If the controller is not around at the finish, where is the
control? Don't confuse prediction with control.

The controller WAS around at the finish (of control -- of getting
what he/she wanted), but is NOT around at the time the other party
begins to perceive a problem (and is not controlling, then, for >what

he/she was controlling before). And, after the finish of >control, the
controller needn't care whether the other party EVER >perceives a
problem, since he (the controller) got what he wanted >and split.

OK, this eliminates the idea of disturbing a present perception in
order to control for a future effect on the person. I'm just as happy
to let it go, because if any such control were possible, it would be
extremely loose, and might take several lifetimes before an error
could be corrected.

(Note that the same sort of analysis holds for nonproblematic
"beneficial" (to the "controlled") future consequences: the
"teacher" can be controlling all along for his/her perceptions of
certain types of actions by the "student" and be long gone -- maybe
even dead -- by the time the student realizes the benefits of the
exercises done so long ago.)

Predicting that teaching will have "beneficial" results is not a
control process, except in the imagination at the time the prediction
is made. Your note explains the motivation of the teacher, but says
nothing about whether beneficial effects can be systematically
produced in this way. It's pretty hard to produce a systematic result
if you can't detect errors and vary your actions to correct them.

I think you've brought out a major deficiency in the commonsense
approach to teaching. You know -- "You should learn algebra now,
because some day you'll be thankful that I made you learn it." This
approach doesn't work very well, if at all -- think of how many people
learn algebra this way, and how many of those have any use for it in
later life or even remember it. This suggests that perhaps we should
focus on teaching algebra in a way that has observable benefits NOW,
that helps a student solve a problem that the student wants to solve.
Or else we should postpone teaching it until such a problem arises.
Many people will never encounter such a problem. They will be no worse
off for not being taught algebra than if they had learned algebra and
then forgotten it. And they might have spent their early years on
learning something nearer to their lines of interest. It occurs to me
also that if we didn't force people to learn algebra at a time when
they gained nothing relevant out of doing so, it might not be so hard
to learn when the need for it did arise. Teaching people what they
don't want to know is a great way to turn them off to any subject.

Most psychologists were forced to pass algebra courses. Perhaps if
they hadn't been forced to learn algebra, they wouldn't spend the rest
of their lives avoiding anything that requires it, like modeling
behavior.

When we introduce the time dimension, we can't arbitrarily cut it
off after a single time-spanning experience. Over time, people
learn from experiences, mostly from doing things or having things
done to them that create error for themselves.

And if you lose your life savings in the first scam you've ever
been hooked by, that you will learn to be wary next time is
woefully miniscule consolation. That's one reason why many folks
are interested in understanding control of one's perceptions which
depend on others' actions: to help PREVENT such catastrophes. (Many
folks are also interested so they can improve counseling/teaching
techniques, too.)

If you're looking for a way to engineer society so that man-made
catastrophes will never occur, I think you will look in vain. Maybe
that's what people want, but they won't get it. Such catastrophes are
caused by people trying to get what they want through controlling what
they perceive. You can't prevent people from trying to get what they
want. Laws and punishment don't do it; persuasion doesn't do it;
teaching doesn't do it; counselling doesn't do it. People will
continue to try to get what they want by any means that works, until
they are dead, even if you have them locked up in a prison cell.

Let's start looking at instances of control of one's own >perceptions

which depend on others' actions, REGARDLESS of what >ANYONE calls it.

Right.

I have nothing against trying to cope with manmade catastrophes, or
any other kind, both before and after they occur. To do otherwise
would be like giving up all attempts to control what matters to us,
and I don't think that would happen even if I didn't agree that it's
not a sensible thing to do. Coping with catastrophes and recovering
from them demonstrate the basic human capacity to control and
reorganize.

But how well we do this depends on how well we understand what's going
on. The first thing we have to understand is that people will continue
to try to get what they want by any means that works, until they are
dead. That's just a fact of nature. It is not changed by anything that
either other people or the environment can do.

Another fact of nature is that people will change whatever they must
change in order to get what they value more or at a higher level. They
will change their actions and their goals in order to continue to
achieve more important or higher goals. They will never act against
their own higher goals -- that is impossible. The nearest they can
come to that is to adopt conflicting goals, and then they can't act at
all.

If we want to change the way in which another person acts, without
getting into conflict with the other, we can only arrange ourselves
and the world we perceive so that the other has a different way of
getting what the other values most, a way that suits us better. This
means that we must also change what matters less to us in order to
continue to control what matters more. This process requires us to
know ourselves well, and to help the other to the same degree of self-
knowledge so that the other can tell us what matters more. It requires
mutual consent to change, a mutual sense of advantage in maintaining
the interaction.

The only human hierarchy of perception and control that we can
understand in real time from top to bottom is our own. The Test for
the Controlled Variable is useful for proving control, for
demonstrating principles, for helping us to choose the control
explanation over other explanations of behavior. It can help us see,
here and there, roughly what others are controlling for, or what we
would be controlling for if we were behaving the same way. But it can
never give us a view of another person's organization comparable in
detail or accuracy to the view that the other person can have. We are
simply in the wrong place to understand another person fully: we are
outside, looking at superficial indications; the other person is
inside and knows exactly what perceptions are involved.

This is why I maintain that changing society in any substantial and
lasting way requires changing the understanding that people have of
themselves, and therefore of others. I think that HPCT provides a
useful understanding. I don't think that HPCT is going to work through
developing tools and techniques that experts can use on society or its
members to make them behave more harmoniously or to prevent them from
causing more catastrophes. I think that HPCT provides an understanding
which will have its effects on society primarily through being
understood by the members of that society. Out of that understanding a
new kind of society will emerge (as we have seen in our own little
one).
--------------------------------------------------------------------
Best to all,

Bill P.

From Oded Maler (921030)

* [From Bill Powers (921024.0830)]

···

*
* I'm not sure how you see the idea of stimulus and response entering
* into tracking a cursor. In our experiments there is always a
* disturbance, so "responding to the stimulus in a certain way" will not
* result in tracking. There's no response that corresponds to each
* target position such that the result will place the cursor at the
* target. If you consider the cursor as part of the stimulus, then it's
* neither a dependent nor an independent variable.

My personal meanings of stimulus and response are much less loaded
than that of psychologists. What I meant was that in some moment the
tracking system receives a reference signal that makes it respond in
certain way (that is, try by variable means to achieve a certain
perceptual goal in the presence of disturbances, bla-bla, etc.).
Perhaps in your setting this event occurs only once when the
experimenter tells the subject the rules of the game ("try to track
everything that appears"). Now suppose
that the figures appear and reappear on the screen, and that after
some training the rules change: some shapes you have to track, but
others you ought to ignore. All this was just a suggestion of how
to reproduce a "sorry"-like effect in the laboratory.

* Perhaps I didn't make my question clear. When I asked what kind of
* experiment you could do to determine that the "sorry" loop is lower in
* the hierarchy than the conscious loop, I was only asking what criteria
* you would use to determine that saying "sorry" was lower in the
* hierarchy than some conscious kind of behavior, such as evaluating
* whether saying sorry is appropriate.

I think hierarchies are personal and cultural. I can observe some
parts of it in, say, species from the French culture, because I'm an
outsider from a culture where this reflex (and others, such as looking
backwards before you close the door while entering a building) is
situated in the higher conscious level (if at all..) The ideal way to
determine it would use a crude hypothetical (and probably impossible)
mapping of the hierarchy to the brain, and that use some lesions/drugs
that are known to affect certain areas (areas in the broad sense of a
generalized geometric/chemical topology). Till then I don't see any
*scientific* way to speak about it.

* You must have some sort of actual behavior in mind, and use some way
* of interpreting what you observe. How else can any theoretical concept
* have meaning?

They have private meanings to me as a pertt time theoretician of my own
mind.

The question of how to determine whether "something" is higher or lower
leads to the general problems of whether CEVs really exist (otherwise,
what is that "something" in the subject's mind that you and me are talking
about?). In your experiments, you assume that the percept of, say,
a cursor on the screen has some universal characteristic so that you
may assume the existence of a corresponding signal somewhere in the
hierarchy of every subject, as well as the possibility to refer explicitly
to this signal by verbal means, while instructing the subject.

Well, I don't seem to converge to an answer. The answer should be however
related to the notions of conflicts and bounded resources within levels.

--Oded

--

Oded Maler, LGI-IMAG (Campus), B.P. 53x, 38041 Grenoble, France
Phone: 76635846 Fax: 76446675 e-mail: maler@vercors.imag.fr