Habitual well-oiled responses (was Understanding...)

[Martin Taylor 991022 09:56]

[From Bill Powers (991021.2003 MDT)]

(I've changed the subject heading from "Understanding the hierarchy,
because it's becoming hard to find in the message stream which aspects
of the hierarchy are under discussion, and this one seems quite distinct
from the original question that started the thread).

We have a perfectly straightforward case
of two models which can't both be right. One says that perceptual events
trigger off well-oiled habitual responses, and the other says that
responses will vary as needed to create the desired outcomes of behavior,
as appropriate to disturbances -- when independent disturbances of the
outcome are allowed to occur. I think there is already plenty of
experimental evidence to favor the control model and rule out the
straight-line model, but perhaps you don't think so. If you don't, it's
incumbent on you to produce the experimental proof that the control model
is wrong for behavior in the natural world, and your model is right.

I'm not sure the situation is so black and white. Particularly in the case
of sequence control, it takes some time for the error signal to develop,
during which time further acts that form part of the "wrong" sequence
may be executed. I think, for example, of the time when I was writing
a lot about signal detection theory. I type pretty fast, and I found
time after time when I wanted to write, say, "detecting" I would write
"d e t c t i o erase n g". It certainly seems like a sequence of "well-
oiled habitual responses." The same occurs in playing the piano when the
desired note sequence starts like a commonly used one but continues

The models may be mutually contradictory, but the overt phenomenon of
producing a habitual pattern of actions occurs in the real world, and
in both models. When the "habitual pattern" is the appropriate pattern,
as it often is when the low-level systems are controlling well, then
there's no way to distinguish the models. I think the crucial test has
to involve something along the lines of how long it takes to correct
a "wrong" sequence.

"Control" in PCT is often asserted to be something that is continuous,
the perception and the error and the output all varying together, in
contradistinction to approaches where one tries something, tests it
to see whether it was right, and exits if it was (TOTE). But sequences
have beginnings, middles, and ends. It takes some time after the start
of a sequence that sets references for .... that sets references for
muscular outputs, before the sequence control unit can perceive whether
the sequence of perceptions incoming is correct. Moreover, at the level
that set the reference for "this" sequence to be perceived, it takes
even longer to determine that "that" sequence is being executed, and
to output signals that cause the correction of sequence to be made.

The analysis of sequence control is difficult, because sequences are
extended over time, and loop time delays are a critical element in the
analysis of any control system.

All of which is a rather verbose way of saying that even "standard HPCT"
predicts the production of "well-oiled habitual" actions (I won't say
"responses" here).


[From Bill Powers (991022.1606 MDT)]

Martin Taylor 991022 09:56 --

I'm not sure the situation is so black and white. Particularly in the case
of sequence control, it takes some time for the error signal to develop,
during which time further acts that form part of the "wrong" sequence
may be executed.

There's no reason the error signal couldn't appear upon the first wrong
element's appearance. From your typing example, it seems that the first
effect of an error is to reset the sequence and start it running again;
this resetting seems to happen quite soon after the occurrance of the error
(half a second or so). I like Rick's offering: it may take even longer for
the next level to reset the reference signal, so the mistaken sequence
could repeat the same error several times before the reference signal can
be changed. The erasing, of course, does not correct what is causing the
error; it merely attempts to make a correction of the outcome on the screen
after it has appeared.

As to your idea about the category level providing symbols, placing it
beside the analog levels is unnecessary; the basic postulate you're making
is that the category input function can receive inputs from any of the
lower analog levels. As before, I'll take it under advisement. My hangup is
that I can't see how the assignment of symbols happens -- that is, why and
how the signal indicating presence of a category somehow gets associated
with visual or auditory configurations, transitions, and events. The
mechanics of naming eludes me.


Bill P.