[Martin Taylor 2013.02.11.09.10]
[From Rick Marken (2013.02.10.2000)]
Martin Taylor (2013.02.10.14.30)–
MT: Your approach (step
- of sticking with a theory X so long as it explains all
the results is fine, if you have come across a good theory X
by some happy chance.
RM: Yes, I came across PCT by happy chance.
MT: How do you do that
without having first gone through what you characterize as
my approach? Did theory X spring fully formed from your
mind, as Aphrodite from the brow of Zeus?
RM: Well, I wasn't the one who came up with PCT but however it
was come up with, it was come up with and so far it works
great.
I used "you" in the generic sense, not meaning "you -- Rick". And I
agree that it “works great”. I stipulated that in the message to
which you were responding. The point was that it is usually the
small anomalies in a theory that “works great” that lead to greater
understanding.
MT: In point of fact,
you violate your step 4 in what you describe as your
approach, when theory X is “HPCT using the “standard” type
of control unit”. We all know of the characteristic
deviations between this control model and human performance,
but instead of following your “step 4”, you say “It fit well
enough for me – isn’t it perfect”.
RM: Actually, I don't "know of the characteristic deviations
between this control model [PCT/HPCT I presume] and human
performance". Please tell me what they are; and tell me what
changes to the PCT model should be (or have been) made to
account for these deviations. Finding deviations from PCT
would be very exciting; I can’t believe I haven’t noticed
them.
Nor can I.
Have you never looked closely at the fitted tracks at places where
the target changes location or velocity abruptly?
Actually, as I said before, I wasn't talking about deviations from
PCT or even from the HPCT model, but from the leaky-ingtegrator
output function. When I discussed the characteristic anomalies with
Bill a few years ago, I had the impression that the little
deviations between human and model in low-level tracking tasks were
well known and ill-understood. And if I knew what changes to the
output function would resolve the anomalies, I would not have
mentioned them as an open question within the HPCT structure.
MT: Experimentalists can't get very far
without theorists. The converse is not always true.
RM: I don't think either can do without the other.
It depends what you mean by "go far". If you mean "gain general
acceptance", I agree with you. If you mean “advance understanding of
some aspect of science” I don’t. Think of relativity, which was a
novel way of interpreting the idea that what matters is perception.
Einstein realized that this was important and theorized a whole lot
of things that would never have been thought of in the “nearly
finished” physics of 1900. There were no data to suggest relativity
might be a more accurate way of describing the world than the
prevailing “absolute universe” view, so it didn’t catch on generally
for nearly 20 years, until an out-of-left-field prediction turned
out to be correct. Then, with data that preferred relativity to the
Newton-Maxwell view, it did gain general acceptance. But the
discrepancy that was predicted exactly by Einstein was
observationally tiny and the observation was sufficiently hard to
make that probably nobody would have looked for it if he had not
predicted it years earlier.
In respect of comparing theories, let's go back to a couple of
possible alternatives to the HPCT structure, and ask which of your
many experiments and demos would give different data under these
different alternative possibilities. I think that the answer is
“none”, but I expect you to correct me if I am wrong.
The alternatives are based on the HPCT structure, each with a slight
modification (your preferred “hill-climbing” approach).
Possibility 1: Lateral feedback loops are permitted within a level,
perceptual signal to perceptual input or output signal to reference
input at the same level, rather than having lateral connections
limited to “the imagination loop” connection between an output and a
perceptual input at the same level. Such connections can
theoretically create sharpened perceptions or decisive outputs that
are hard to achieve with a strictly tree-structured perceptual or
output configuration. Experiments would be needed to show that such
lateral connections cannot exist, or theory is needed to argue that
if they existed, reorganization would eliminate them. I know of no
such experiments or theory.
Possibility 2: Rather than there being a category level, above which
all perceptual types are based in their specific kind of logical
relationship, there is a category boundary that can be visualised as
lying vertically alongside all the ascending analogue levels. There
is no well-defined “top level” type on either the analogue or the
logical side of the category boundary (though the number of levels
is necessarily finite). Such a configuration allows for categories
of intensities, transitions, sequences, etc. without “level jumping”
perceptual input structures. Such categories do occur in my own
subjective perception (remember that his subjective perception is
the basis on which Bill’s HPCT levels have been developed).
I rather think that all the available experimental data would give
each of these possibilities (and probably many more) exactly the
same cachet of “working great”, because with those studies they
would all give the same predictions. There’s nothing particularly
novel about these possibilities, but it might be possible to develop
experiments to discriminate among them, if anyone was interested in
seeing whether HPCT is a unique “working great” configuration of
PCT. Maybe it has been done already, but not to my knowledge.
Martin
