Randall's experiment; Challengs 2

[From Bill Powers (930330.1330 MST)]

Allan Randall (930330.1330 EST) --

Sorry to be so slow, but reading your conversation with Rick, I
realized that I don't even understand what experiment you're
proposing to do (with or without a real subject).

Let's get it into a diagram so I can tell what's connected to
what under what conditions. Here's a start, the basic control
system. It's simplest just to use an integrator in the output

                     > ref sig (constant)
            p----- comp ---- e= error
            > >
          inp fuct integrator
            > >
        i=input o=output
          +| |+ |
           > ---------------

for simplicity we assume that p = i; the input function is just a
unity multiplier.

As I understand it, you want to do a short run and record p, o,
and d. Then you want to cut the input line and insert an
artificially generated signal a:

                     > ref sig (constant)
           p ----- comp ---- e = error
            > >
          inp funct integrator
       / |
     a (i) (o)
          +| |+ |
           > ---------------

Then you want to start creating all possible signals a, while
comparing the output o with the disturbance d. The search ends
when o(t) is the same as it was with the loop closed.

In the successful case, d + o will produce a value of i ( = d +
o) that has the same waveform as a. So you're really searching
for the waveform of a that produces a waveform of i such that

i(t) = a(t) for all t

But this is no different from simply solving the closed-system
equations for the value of the input variable, which is the same
as the perceptual variable. If I understand what you mean, you're
proposing to find the solution by systematically going through
all possible solutions until you find the one that works.

Is this correct so far? I don't want to try to follow the rest
until I'm sure what the actual procedure and method are.


Bruce Nevin (930330.1316) --

Did the person moving the car even perceive the space that the
car covered, before or after moving it? What if she was
controlling line of sight, or shade?

Right, those are the questions that the Test would have to
resolve. I didn't pick a wonderful example, but you get the idea:
actions can have more than one outcome, and generally only one of
them is intended, although more than one may intended.

In the case of language, the ambiguity, the alternatives of
structure (information), is a socially established property of
the utterance. You need some version of the Test to determine
which, but you would never hedge the question with "if any".

Agreed. The same actually holds for any action, because according
to the basis hypothesis, the only reason for which any action is
ever produced is to maintain control of some perceptual variable.
The proviso "if any" was meant merely to acknowledge what you
said: that the actual intended outcome of the observed action may
not have been on the offered menu of choices. It's conceivable
that the purpose of uttering the words "Flying is dangerous" is
merely to check that the microphone at the podium at the
linguistics conference is working, with neither possible meaning
being intended.
Greg Williams (930330 - 2) --

I think it would be a lot MORE reasonable to perform tests
aimed at understanding (by experimenter and subject alike)
one's behavior which are not COMPLETELY dependent on possibly
fallacious subjective reports.

Subjective reports can be fallacious if they contradict evidence
that can be obtained about the same thing without relying on
subjective reports. They can't be fallacious if the question is
how the world appears to the observer. The investigation of
illusions, for example, depends completely on subjective reports:
if a person says that the crossbar of a T looks shorter than the
stem, one has to accept that report as truthful in order to
establish that the illusion even exists.

I can accept your statements as truthful without a qualm, because
what I am doing is trying to determine how your experiences
appear to you. I'm not trying to catch you up in a mistake by
showing that what you report to me isn't true. Of course I have
to trust that you will report, as truthfully as you can, what
seems to you to be going on in present time. This is no different
from asking you if it seems to you that "if A is greater than B
and B is greater than C, A must be greater than C." That is a
subjective impression which most human beings who reason
experience as the truth, even though sometimes it isn't true. I
recognize that your words may mean something different to you
than they mean to me, but we don't generally do too badly in
spite of that. At least we can come to an understanding that can
be checked out in other contexts.

You may not have noticed, but I've been applying Tests all over
the place, attempting to disturb variables that I hypothesize
that you're controlling for, and noting whether you resist or
not. So far I've established that it's difficult to get what
would normally be accepted as an answer to a question -- that is,
a response cast in the same terms as the question. I've found
that the answers I do get tend to invalidate my trial hypothesis
with interesting regularity. I'm still trying to learn enough
about this phenomenon to make it worth while to offer a real

Appropriate use of the Test for Controlled Variables needn't be
so dependent on subjective reports, I think -- or am I wrong?
(It seems obvious that the Test could be applied to non-human
animals -- not over the net very easily, of course, as you so
wittily pointed out yesterday.)

Yes, you apply a disturbance to the variable you hypothesize to
be under control, and look for systematic resistance to it. In
nonverbal situations, this means direct physical interaction of
some sort, which we can't do over the net. In purely verbal
situations, like the one we are in, it means uttering words
calculated to have a certain effect on the words that another
person speaks, under the hypothesis that the other has a certain
intent in uttering the words. If there is resistance, one is on
the track of something controlled; if not, one has to try another
hypothesis. Success will depend largely on how well we can each
communicate meanings to each other. Fortunately, we speak much
the same language.

When I said you were an easy case, by the way, did you experience
anything that you didn't tell me about?

But (a big "but," I think), the Test cannot be applied to
behavior which happened in the past (at least not with current
technology) ...

Right. But I'm interested in controlled variables and control
organizations, not behaviors. There's not a lot anyone can do to
understand a behavioral organization that appears and disappears
in five minutes. But I don't think that real human control
processes come and go so whimsically, unless the changes are
intentional (in that case, there is still regularity to be
found). If you were controlling for something important to you a
week ago, chances are that you're still controlling for it. What
other assumption can one go on, when trying to determine the
characteristics of any system?

... and the Test, as you have noted, is difficult to apply in
field conditions.

Not as difficult as you think. I would say challenging.

The point I think (but don't know!) I'm trying to make NOW (not
necessarily when I first issued the challenge) is that
understanding (and especially predicting) complex individual
human behaviors in the field by using The Test looks to be much
more difficult than laboratory tracking experiments.

Well, I don't necessarily buy that. Outside the laboratory there
are far more controlled variables to be found. The chances of
finding one that will open the door to another are much greater,
and the kind of controlled variables that will be found are much
more natural. It's harder to quantify them, to be sure, but at
the higher levels that's not our immediate concern.

... it is at least possible that, IN PRACTICE, PCT won't be
able to generate population measures which some behavioral
scientists will continue to desire.

Just curious -- is that all you intend to say about my proposal
the other day for predicting population demand curves from models
of individual behavior?
Back, as you say, to the challenge.

It now seems to me that you are reluctant to report subjective
impressions and offer them as answers to my questions. Is this
correct? If so, this might explain some of the failure of my
questions to elicit answers cast in the same terms. But I won't
know whether this impression fits your experiences unless you
tell me.
P.S. primers on the way, ASCII. Yes, it was the new Simcon.
Unless I've lost another dozen neurons.
Best to all,

Bill P.