[From Bill Powers (950228.1100 MST)]
I accidentally send a message for Isaac Kurtzer and Tom Bourbon to CSG-
L; no harm done, I hope.
···
-----------------------------------------------------------------------
Joel Judd (950227.1515 CST)--
I guess if you mean by "explicit" the EXACT, SAME, UNCHANGING
object or concept or thing, then for me there is no explicit
storage and retrieval of information. Maybe I'm reading too much
into your question, but it seems to border on the real reality
problem, or at least the implication that we somehow induct what IS
"out there" into the brain.
I don't follow this. If memory involves recording and retrieval of
perceptual signals, doesn't this still leave the same problematic
relationship between perceptual signals and "out there"?
Without oversimplifying, if HPCT describes perceptual control, then
an implication for me is that imagination and memory are closely
related and may, in some cases, actually be the same thing.
But what about the relationship of memory to past events? If higher
systems select lower-level reference signals from recorded values of the
lower-level perceptual systems, imagination would, as you say, also
involve memory. But where do the recorded values come from, if not
recordings of perceptual signals?
The <car> I recall by replaying, so to speak, relevant perceptual
levels; yet, each time I recall it, and as time passes, the
memory/imagination changes.
This is perfectly consistent with the model proposed in BCP. Recorded
perceptual signals are played back into the same channels that interpret
present-time perceptions, so if the perceptual system changes, so will
the memories. Also, for other kinds of failure of memory, we have to
remember that there are three other components of any memory: the
recording method, the recording itself as a physical trace, and the
playback method. Changes in any of these will also result in changes in
experienced memories. And if the original perception has a dubious
relationship to the external world (as in optical illusions), so will
the memory.
Let me cite part of Campbell and Bickhard (1986) and then run for
the day:
"For the interactive approach, in contrast, representation is NOT
constituted by any kind of structural correspondance between what
represents and what is represented.
This is essentially the PCT approach, as least as far as representation
goes. A signal that represents the presence of chocolate is not itself
made of chocolate; a signal representing presence of a square does not
have a square shape. A signal simply indicates that an amount of a given
function of lower-level signals exists; it's like a meter reading. A
meter reading indicating -10db of audio level does not make a sound.
Representation is an interactive functional property rather than a
structural property ...The ontology of representation and knowledge
is system organization, which could be regarded as a kind of
structure....
I suppose that must mean something, but I can't figure out what. Can you
explain this to me in simple words?
For the interactive perspective, then, learning cannot have
anything to do with structures being stamped in or imported from
the environment.
Is there anybody who has seriously proposed this? This strikes me as one
of those "opposing views" that nobody supports.
Representation and knowledge (and by extension, memory) are
constructed indirectly via new system organization rather than
directly in terms of basic elements of representation (encodings)."
Why "by extension, memory?" If you have some sort of representation in
the system, why can't IT be stored and recalled? I'm not sure what the
implied alternative is here. Why reject the simplest explanation?
-----------------------------------------------------------------------
John Anderson (950227.2200 EST) --
Nice review of that report on infant arm movements, which I had seen.
These people actually did the Test, by applying forces to the arms to
find out whether the movements were intentional or accidental. Or at
least they did a rough version of it.
-----------------------------------------------------------------------
Bill Leach (950227.20:58 EST(EDT)--
Replying to Bruce Buchanan (950227.01:30 EST) --
Bruce B:
Clever people who have read Popper may also try to think of tests which
might show that the model is inadequate or false ("falsifiability"
criterion), a more decisive kind of test than looking only for
confirming instances.
I still will maintain that there is a most significant difference
here, these "thought tests" are not subjected to any experimental
validation. The 'falseifications" can be every bit as incorrect as
the "theory".
Actually, every PCT experiment opens the model to falsification. Even
when just fitting a model to data, we require such a good fit that a bad
fit is taken as evidence that the model doesn't describe the behavior.
This has shown up in the sleep-deprivation data that Martin Taylor was
involved in taking. A subject who is falling asleep at the control stick
produces behavior that the model simple can't imitate; the conclusion is
that this is not control behavior as defined in PCT: the hypothesis that
control behavior (at least of the kind we're modeling) is taking place
is falsified.
PCT models become even more vulnerable to falsification when the
parameters found from one experimental run are used to predict handle
movements with a new disturbance pattern or even the addition of a
disturbance that affects a new aspect of the display. Generally, the
predictions are very nearly as accurate as the fit of the model to the
initial run. But there is no a priori reason that they should be so
accurate, or even close. After all, we have a person holding the control
stick and subject to no other physical constraints at all. The person
can move the handle in any way desired in order to accomplish the task.
PCT claims that there is just one way, and goes further to predict not
only performance, but errors. So it is hard to dismiss PCT as simply a
"perspective" on behavior, because perspectives can't be proven wrong,
while PCT can.
The least falsifiable hypotheses are those obtained from statistical
methods. Large numbers of observations contrary to hypothesis are
permitted, yet they don't count against the hypothesis unless they
clearly predominate.
I have always agreed with Popper's dictum of falsifiability. If an idea
is stated so generally and vaguely that there wouldn't be any way to
find out if it's wrong, then there's not much to be impressed with if it
seems to be true.
---------------------------------------------------------------------
Best to all,
Bill P.