[From Rick Marken (960105.0900)]
Shannon Williams (960104) --
First- the problem that needs to be solved. (I am going to remove the word
'perception' for right now.)
in the beginning: [input] --> [virgin black box] --> [useless output]
>______________________________________|
in the end: [input] --> [Expert black box] --> [effective output]
>__________________________________________|
I think we are approaching a way to study and model learning. I think we are
narrowing down the learning question to be one of how a system learns to
produce an "effective output". Now I think we have to determine what an
"effective output" is. That is, we have to figure out how to distinguish a
"useless" from an "effective" output. There is no way to tell from your
diagrams _why_ the top output is useless and the bottom effective (except
from the fact that you have labelled them as such).
It looks from your diagram like the black box can produce several different
outputs and that all these outputs have some effect on the input. In PCT, an
"effective" output is one that can be varied to keep the perceptual
representation of the input to the black box matching the reference
specification for that perception. So my preferred definition of an
"effective output" is "an output that allows control of perception". Does
this seem like a reasonable definition to you? If so, then we can move on to
figuring out how "effective outputs" might be learned.
Best
Rick
[From Shannon Williams (960105)]
Rick Marken (960105.0900)--
It looks from your diagram like the black box can produce several different
outputs and that all these outputs have some effect on the input.
No- only some outputs effect the input. For every one million possible
outputs, 999,000 may not have any effect at all.
an "effective" output is one that can be varied to keep the perceptual
representation of the input to the black box matching the reference
specification for that perception.
You cannot think in terms of making a perception match a reference until
you have a reference. Do you see why I am saying this?
So my preferred definition of an "effective output" is "an output that
allows control of perception".
Does this definition mean anything before a reference exists?
then we can move on to figuring out how "effective outputs" might be
learned.
Yes. This is now the key question. I am only sure that, before you build a
reference, you must have an error.
-Shannon
[Martin Taylor 960105 13:00]
Shannon Williams (960105)
I am only sure that, before you build a
reference, you must have an error.
Catch-22, then. Error being the difference between the reference value and
the current value of perception, before you can have an error, you must have
a reference. You are sure that before you have a reference, you must
have an error. Impasse.
Your view of PCT is clearly radically different from mine.
Martin
[Martin Taylor 960105 13:00]
I am only sure that, before you build a
reference, you must have an error.
Catch-22, then. Error being the difference between the reference value and
the current value of perception, before you can have an error, you must have
a reference.
Your definition of error causes the catch-22. I do not have your
definition.
Your view of PCT is clearly radically different from mine.
My view of PCT is evolving. If in the end, my view no longer resembles
all of yours, then I will have to call my view something else. Or maybe we
will all call our new view something else.
-Shannon