[From Rick Marken (960328.0830)]
Bill Powers (9603.26.1400 MST) --
RE: predictability
If I am controlling the relationship between a cursor and a target, and
if the position of either cursor or target is "unpredictable," it makes
a difference whether the unpredictability is inside or outside the
bandwidth of good control.
I think this nicely ties up the "predictabiliy" discussion. The E. coli II
demo shows that a control system can deal with unpredictability that is
_within the bandwidth_ of good control. In E. coli II the results of action
are completely unpredictable in the sense that all results are equally
probable immediatly after a press; but once a result occurs it remains the
same (on the average) for some time; so the rate at which the results of
action change is within a bandwidth that allows control.
If the results of action change unpredicatbly at a _high_ rate (outside the
bandwidth of control) then they can't be controlled. This is consistent
with the point Bruce Abbott made some time ago; both Martin and I are right
(or wrong) about predictability; it just depends on what one means by
"predicatability". If predcitability includes the rate at which changes in
the environment accur, not to just the probability of going from one
environmental state to another, then it's true that, in order to control,
the environment must be predictable within the bandwidth of control.
A human can control in a world that is changing randomly as long as these
changes occur relatively slowly; if the world were randomly different at each
instant in time there could be no control -- which is, I think, Martin's
point.
Richard Thurman (960321.0959) --
How about if you describe what would be necessary to replicate
and extend this [Robertson and Glines] study. And I will try to
gin up a version (in Hypercard or something) that we could use in some
joint research.
Will do; I'll try to write something up on this tonight.
Best
Rick