Chjoosing actions

[Martin Taylor 920926 19:45]
(Bill Powers 920925 20:00)

Whoa, there. One of the consequences of the PCT approach is the
idea that the same end can be achieved by many different means. If
one possibility is blocked, another can be used. And often the
various means are mutually exclusive in any case.

This is a "levels" problem, which I was avoiding during the argument
to keep from getting more complexities into the act.

A control system that acts by choosing one action from a selection of
others can't also choose the amount and direction of that action that
ensues from disturbances. That takes two levels (more than one,
anyway). Achieving the same end by variable means (which is required
by disturbances of various kinds) is the result of a control system's
reacting to error by altering its output.

I didn't mean to introduce a levels problem at all. What I had in mind was
much simpler and more general. For one thing, I haven't yet begun to think
seriously about ECSs that can direct their output to different supporting
ECSs. I am still thinking of "blind" ECSs that simply produce more or less
output depending on the error. But an outside observer can see that prior
reorganization has given this ECS several different parallel paths (which
I have called micro-loops or something similar) by which this output passes
through the outer world and affects the ECS's percept. Normally, the path
with the lowest resistance (I like the word "impedance") in the outer world
provides most of the effect. If the real-world situation makes the micro-loops
mutually incompatible, then only the one with the lowest impedance has any
effect at all--we choose the path of least resistance. If the car is there,
we use it. If not, we bicycle if that's available. Otherwise, we walk or take
the bus. We can't do all at once, but all have at some time or other been
by reorganization made available as expressions of the output of the ECS.

If a low-impedance loop is not available, perhaps the ECS has to produce more
output to have any effect at all on its percept (these things are not linear
when we get to this kind of level). But it may not have to do so. The
alternate means may be used without any change in output level.

I don't think you need any "choosing" systems when you are actually acting in
the world. I think you need them when you are imagining (planning), because
then there are no blocks to the execution of any path (they all have essentially
zero impedance). And the whole issue of imagination needs a discussion like
the one you have been having with Greg. I don't think it is clear (at least
not to me) yet.

Martin