Conflicts

[From Bill Powers (940416.0930 MDT)]

Dag Forssell (940415.1200) --

Thanks for the copy of Keillor's wonderful talk.

···

---------------------------------------------------------------
Bob Clark (940415.1733) --

I appreciate your analysis of the conflict between Martin Taylor and
me. It's not quite as serious a conflict as it seems on the surface,
and it's not entirely about mathematics. I think that one day Martin
will develop an information-theoretic analysis of a basic control
system, at which point we will all see what it has to contribute. A
lot of the argument, which Martin and I are only slowly freeing of
irrelevant side-issues, is about interpretation and terminology. I
have no question that Martin grasps the basic relationships as I do
-- just consider his post on the ins and outs of analog modeling on
a digital computer.

As to the resolution, it's true that different kinds of analysis can
be appropriate under different circumstances, but not if they
contradict each other. A contradiction means incompatible premises
or incompatible facts, and that's a no-no in what is supposed to be
a scientific system. It's worth the trouble to peel away superficial
disagreements to get at the direct contradictions.

I don't, by the way, put a lot of stock in the simple canonical
control system model. It's always been surprising to me that it
handles any real behavior as well as it does; I had expected that we
would need nonlinear functions, on-off control, control by
parameters, and multilevel models even to handle simple behaviors.
The fact that we don't, in simple cases, is nice, but only means
that we're dealing with simple aspects of behavior. I don't make any
predictions as to what model will be needed to deal with higher
levels, except that it will bear a strong family resemblance to the
canonical model. When I start work on any more complicated model,
like the Little Man Version 2 or the operant-conditioning model
(which I promise to resurrect), I always try to wipe my mind clean
of previous models and start over with the problem itself. As best I
can.
---------------------------------------------------------------
Tom Bourbon (940415.1738) --

Part of the struggle is won; you sent to the net. I'll try a test
message, too, but I'll wait for a direct send from you to confirm
your address, since Martin couldn't get through.

I agree with Martin about the quote he sent: don't let it happen
here.
---------------------------------------------------------------
Rick Marken (940415.1900) --

Don't go away mad. Let's get a network discussion going on ways and
means of dealing with wrong ideas. I agree with your stand on
sticking to the basic concepts of PCT. Reviewing posts for the last
year and a half, I see that most of what you say is unexceptional
and in fact right on the mark. You don't lose your temper very
often, and then only when an impasse occurs. Certain points in an
argument seem to be reached where there is a direct conflict between
what someone says and what you maintain is the right idea. The
conflict is so direct that there's no point in pressing it. The
problem is to discover where it's coming from. This means
deliberately putting the conflicting ideas on hold and finding out
why the other guy is insisting so hard (and also why you are
insisting so hard).

I came across a remark by Ken Hacker over a year ago: he said you
were telling him he didn't know how to do research. If either you or
I had been practicing control theorists, we would have recognized
that as an opening to go up a level. Instead of doing that we just
said "There, there, Ken, don't be upset" and continued at the same
level. Ken supplied an opportunity to get off the specific conflict
and start a discussion of what doing research means, what kinds of
research there are, what the problems might be, and so on. What's
different about the way we do PCT research and more conventional
approaches? That in turn might have opened up an even higher level:
what do we think we can actually learn about human nature? Why do we
want to do research in the first place? Are there any other ways to
find out what we want to know?

It's hard to let go of a conflict, to let go of the desire to win
it. Going up a level feels like abandoning a struggle, giving up.
That's what it is. It's getting into a position from which the
struggle is seen as futile or secondary, and new considerations come
into view. Maybe the conflict will get even worse; if so, up a level
again. No conflict is worth dwelling on; once you realize there's a
conflict, you know everything about it that's interesting. It's not
interesting if you can prove with impeccable logic that you are
right, and the other guy can do the same for his own view. If the
conflict persists despite impeccable logic on both sides, the
problem is not at the logic level. The problem may be at a higher
level (principles, system concepts) or, come to think of it, at a
lower level (facts). The only place you DON'T need to look any
further is at the level where the conflict is obvious.

If there's a problem, let's deal with it. What is the disagreement,
and when that is known, why does it exist?
--------------------------------------------------------------
Best to all,

Bill P.

Hi, Martin --

I'm not mad at you, even when I take you to task. I think you have some
terrific ideas. But I disagree with you about some things, and either we'll
work them out or we won't. I expect we will.

The latest post I send out (about 16:45 or so, delayed by lightning)
answers some of your comments.

I don't take your posts as criticisms of me where not intended, and even if
intended (if any) I wouldn't object to being criticized. We can take it for
granted that we're basically friendly, can't we? My friends have license to
say what they like to me. We're talking about substantive matters, and as
long as we're doing that we ought to be able to speak freely. You must
realize by now how much I respect your intellectual capacities.

RE: VOT study.

But I don't agree that they "concealed the truth." They just reported
what they observed, in the mode of the time.

It's the mode of the time that I'm objecting to. I think the honest thing
to say is "N subjects showed varying degrees of effect, M showed the
opposite effect, and Q were equivocal." It's not honest to say "Subjects
showed the effect" when the truth would be stated "Some of the subjects
showed the effect," and more important, "some didn't." This is a custom I
would like to see abolished. It cripples science.

<...once you get beyond continuum effects into the control of >categories,
you should never expect high correlations to show up in >related measurable
low-level variables. I didn't say it there, but I >say it here, that this
includes the possibility that people may show >opposite effects at the
physical level.

Yes, I agree with this. If we're taking data about category-level
phenomena, however, we will be perceiving the low-level variables in terms
of hypothesized category variables, so there will still be high
correlations (when we get it right). The hypothesis has to be stated in
terms of the appropriate level of variables. The HPCT model is constructed
so that ALL levels of variables can be put to experimental test. I do agree
that the relationship between low-level variables and high-level ones may
be unpredictable in a lot of circumstances. As long as that's true, I
wouldn't try to predict such a relationship, particularly not across
subjects.

I also point out what you have often pointed out, that we do not and
cannot know all the percepts that people are controlling (and I add >that

what they are controlling may change drastically from moment to

moment).[Etc.]

I repeat what I said in my last post: I think you're borrowing trouble.
While I don't say you're wrong, there are ways of getting good data, and I
think we should pursue them.

If you won't accept any data except facts, you will not accept any >data.

Facts are fictitious. There aren't any.

I use the term in recognition that there is no certainty. I'm speaking in
comparative terms: I draw the line for "facts" so that acceptable levels of
unpredictability are very low.

But, statements that are false for a large fraction of the population >may

nevertheless be useful, if one can identify the circumstances in >which
they are true.

But a statement of the circumstances is part of the hypothesis, isn't it?
"[Live] cats can be trained to ..." I'll happily accept a statement that is
true of part of a population, if you can tell me for each individual, with
high accuracy, which part that individual belongs to; of course the
statements must hold true for every such individual, ideally without
exception.

Is there no fact worth pursuing, if some people do better in one >scheme,

whereas others do better in the other?

All such facts are worth pursuing if you can specify WHICH individuals as
clearly as you can state the prediction and the result. This amounts to
explaining why some specific individuals, with specific characteristics,
show the effect and other specific individuals, with other characteristics
and in the same circumstances, don't. In fact I think a scientific
prediction must explain both successes and failures -- in which case there
are no failures of prediction.

If you really mean:

Thank you for bringing this up. This is going to save me a lot of >>time.

From now on, the first question I'm going to ask (if I remember >>to do so

-- if not, remind me) is whether the phenomenon to be >>explained is
actually a phenomenon. If it's not, I'm going to turn my >>attention to
more profitable questions. From here on down, it's all >>peanuts.

then HPCT will continue to be isolated from most of psychological

<science, and most psychologists will continue to fail to see its power

and beauty. That would be a shame.

You're right, and it's a shame. But I mean it. I would like to persuade
psychologists and others that using the principles of HPCT, and MAINTAINING
THE STANDARDS I DESCRIBE, it is possible to do research of an entirely
different caliber from what has been done in the past. There is no way to
say this without pointing out that existing brands of behavioral science
rely on very bad data and accomplish very poorly the task of predicting or
even just fitting behavior. If one doesn't see a serious problem with the
existing system, where is the motive for putting out any effort to find or
understand a better one?

I don't deliberately try to antagonize psychologists. But I don't live or
die on the basis of their approval, either. From my point of view, it is
they who are isolating themselves from science, not I.

The obvious solution to this dilemma would be for me or some other CSG type
to actually start doing some research with the higher levels of
organization, to show how HPCT can be used to get good data and accurate
predictions. I've had the foolish notion that if I could, instead, persuade
some conventional psychologists of the merits of HPCT, and if they learned
to think in those terms, they would do this kind of work (and a few have --
just barely -- begun). That would be the best solution, in that the work
would be in their field of interest. Unfortunately, those who are wise in
the ways of research also have the most indoctrination in the conventional
definitions of what we want to know about human nature, and are used to
methods alien to HPCT methods. Furthermore, the best conventional
scientists also have the most to lose by changing paradigms, including
their pride in work done in the past under the old paradigm. They even risk
losing the respect of their colleagues, who do not understand HPCT and will
think they have gone off the deep end. It's happened more than once.

In case you wonder what I believe I am controlling for here, it is the
sanctity of information. I maintain the belief that even a couple of
bits of information is more useful than none. You are trying to >disturb

this belief, and I am resisting that disturbance. I perceive >you as
believing that only precise (very high information) measurements >should be
processed, and all other information discarded. I am trying >to disturb
that belief, and you resist.

I think you put it well and fairly. This is not a trivial disagreement.
There may be grounds for resolving it; I don't know. I certainly want to.
But there are some fundamental points that will be hard to agree on. I too,
believe in the sanctity of information, but how many bits, amid how much
noise, must we have before we can call them information? I think there are
many circumstances in which a little -- or even a large amount of -- low-
grade information is worse than none, because we tend to build on it in
directions that have little chance of being right, and what we imagine it
to mean leads us in positively wrong directions. I acknowledge that high-
grade information is hard to come by, but I think you're unduly pessimistic
about the possibility of getting it using the methods of control theory.

I know -- put up or shut up. I suppose I'll have to try to put up. Story of
my life. First crowds, then arms, then cockroaches, and now who knows what?
Personality research? Evolutionary models? Economics? Neural networks? I
quail. It's impossible. Unless others understand and adopt the basic
principles and start doing the hard work of reinterpretation and
experimentation, the promises of the PCT paradigm are not going to be
realized for a long, long time. I'll do what I can as long as I can, but it
won't be enough.

We could talk for years on a hypothetical plane without resolving our
disgreements. The first time you design a pure simple PCT experiment and
sweat through getting it to run the way it should run, with a working model
that predicts the data exactly, you will see everything I'm talking about
in a new way. Why do I feel that you're avoiding this step? Are you?

I do realize that you're committed to HPCT and are a major spokesman in its
favor. But without any right to demand it, I'm hoping for a more concrete
kind of committment.

Anyhow, I continue to find CSGnet a source of endless fascination and
challenge, your contributions being a large part of it. I'm at the point
where I have to make specific plans to cut down and get other things done
-- with regret. I hope you continue bugging me for a long, long time. Even
if I'm unreasonably hard to satisfy.

Best,

Bill.

P.S.

I don't think the conflict is really in the area of "old ideas," >because

I have a rather low insistence in that area (I think).

Sure about that?