[From Rick Marken (981128.0915)]
Bruce Gregory (981126.1400 EDT)--
Another way to look at this is to realize that the _only_ way
a system can _know_ that it has achieved a desired outcome is
to compare a representation of the way the world is with a
representation of the way the system wants the world to be.
Me:
I would change "the way the world is" to "the state of the
outcome" and I would change "the way the system wants the world
to be" to "the intended state of the outcome".
Bruce Gregory (981127.2205)--
I'm not sure why you like these changes. They seem obscure to me,
but obviously not to you.
I like these changes because they make it clear that it is a
representation of a particular aspect of "the way the world is"
(the state of the outcome variable) that is controlled.
My point was that there is no need to control your output as
long as you are guided by the difference between the two
representations.
OK.
Me:
And the system can't bring about intended outcomes in "any way"
(using whatever outputs) it likes.
Bruce G.
Oh, but it can. As long as the system is willing to correct on
the basis of its perceptions. I can start out in any direction
I like, but I must allow my perception of whether I am getting
closer to or farther from my desired outcome determine how long
I persist in going in any particular direction. I believe this
is the way the e-coli mechanism works. Or am I wrong?
You are basically right about the E. coli mechanism. E. coli actually
"corrects" (tumbles) on the basis of a _comparison_ of perception
to a reference for that perception. E. coli doesn't _perceive_
whether it is getting closer or farther from the desired outcome. It
perceives the rate of change in the concentration of the chemical
gradient at its sensors. It compares this perception to a reference
for that perception and tumbles when the integrated error reaches a
threshold.
E. coli cannot produce the intended perception (of chemical gradient)
in "any way" it wants; it _must_ swim _toward_ thesource of the
gradient. E. coli finds (usually) this "way" by random means; but
it _must_ find this "way" (it must generate outputs that move it
toward the chemical source) or it will not get the perception of
chemical gradient that it wants.
Me:
The only outputs that will work are those that precisely oppose
the net effect of disturbances to the controlled variable
(controlled outcome). Remember, o = -1/g(d).
Bruce G.
Again, this depends on what you mean by "works". Some patterns
of output may be more efficient than others, but all will
_eventually_ work if you let the direction of error change
determine the next output. This process is sometimes called
learning.
I think you should be able to see now that this is not correct.
Some "patterns of output" (like always swimming _away_ from the
gradient source) will never produce the intended perception.
E. coli's random response generation process _must_ hit on the
_right_ output (swimming toward the gradient source) or E. coli
will never get the perception it wants.
The world (of disturbances, d) determines what _must_ be done (o)
to control a particular perception: o = -1/g(d). What the world
doesn't determine is how (or whether) the system learns to do what
must be done. An E. coli who has a terrible run of bad luck because
every random tumble ends up pointing it in the wrong direction will
simply never produce the perception it wants -- with morbid
consequences for the bacterium.
Best
Rick
···
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/