Learning what's right (was Control of (and with) rate of firing)

[From Rick Marken (981129.0930)]

Bruce Gregory --

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.


This paragraph was a red flag for me. What you seem to be saying
here is that anything one does (any pattern of output) will eventually
"work" to achieve control, though some patterns may be more efficient
than others.

Bruce Gregory (981129.0745 EDT)

You seem to have ignored everything after the "if".

It wasn't clear what everything after the "if" meant. But I don't see
how even a generous interpretation of its meaning can save it. It is
simply not true that "all [outputs] will _eventually_ work if you let
the direction of error change determine the next output". In the E.
coli case, for example, some outputs will _never_ work even if you let
the direction of error change determine the next output. Outputs that
move E. coli away from the target will never work to get E. coli to the

Most of E. coli's outputs are "wrong". Does this prevent it from
achieving its "goals"?

No. What would prevent E. coli from achieving its goals would be if
_all_ of its outputs were wrong every time -- which is a real (though
low probability) possibility.

What is the nature of a "disturbance" as far as E. coli is concerned?

Changes in perceived concentration that are _not_ caused by E. coli's
actions. The actual shape of the concentration gradient is one such

Is its organization any different if "disturbances" are present?

No. But I don't see what this has to do with my basic point, which is
that there is right and wrong in control; only the outputs that oppose
the net effects of disturbances to a controlled variable are "right";
all other outputs are "wrong". Learning to control means learning the
control law that produces the "right" outputs. As I said in an earlier

the [real] world (of disturbances, d) determines what _must_ be
done (o) to control a particular perception: o = -1/g(d) [the real
world determines what consititutes the "right" outputs]. What the
[real] world doesn't determine is how (or whether) the system learns
to do what must be done.





Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net