[Martin Taylor 940601 1310]
Bill Powers (940531.1800 MDT) responding to Bill Cunningham (940530.1530)
It seems to me that Bill is right, but nevertheless Bill is right also.
Bill P. deals with Paul Revere's perceptions of lights and so forth,
whereas I interpreted Bill C. as being more concerned with a higher level
perception, which was uncertain. Revere's uncertain perception was that of
the timing and location of a British attack. We assume that Revere had
a belief that there would be an attack on a certain night (or span of nights),
but not when during that night or where it would be. We can call that
pereception "perception U". (If he had had no prior belief that there
would be an attack, the whole business with lights would have been
pointless).
Perception U is not a controllable perception in itself. But the uncertainty
associated with it is controllable. Revere could have invoked the Search
mode by sending out personal spies, but instead he had already engaged to
obtain the information that would reduce the uncertainty by other means--the
lights. So, as I see it, one of the many controlled perceptions in this
situation is Revere's perception of uncertainty about an important perception--
Perception U, the whereabouts and timing of the British troop movements.
Now why should Revere control for low (not zero) uncertainty about the
uncontrollable Perception U? In simple PCT, when we deal with only one
perception at a time, uncontrollable perceptions are of no interest. It
is because Perception U is an element of another controllable perception,
that of the result of a battle agains the British, for which the reference
level is for the British being delayed or defeated--Perception V. There is
(as yet) no error signal associated with Perception V, which exists as
a perception only by means of Revere's imagination. But Revere's imagination
works in some imagined world, and in that world, if the American forces
defend against a British attack by sea, when the British actually come by
land, the imagined Perception V corresponds to a LARGE error signal. There
is no way for Revere to control Perception V, in the sense of acting in such
a way as to reduce this (imagined) large error signal, unless he reduces
the uncertainty about Perception U. And so, control for Perception V
produces output that becomes a reference signal for low uncertainty about
Perception U, which in turn becomes reference signals for attaining
perceptions that enhance the precision of Perception U and thus reduce
the probable error signal associated with the important perception,
perception V.
It seems to me that situations involving choices (presumably perceptions
being controlled at the program level) are often of this character. There
is an imagined world in which the perceptual consequences of the alternatives
are played out, but the likelihood of the different consequences depends
on the precision with which the perceptions that influence the conditional
reflect the real world. What do you do with "If A then B else C" if you
don't know enough to make a strong assertion about A and have no means of
controlling A? You "Search" for perceptions that will reduce your uncertainty
about A (or alternatively for means to control A, such as having an American
staff officer inserted into British HQ with instructions to say "you'd better
go by sea because I have information that the Americans are waiting for a
land attack").
Simply put, I think "planning" is a normal part of perceptual control at
the program level, but (probably) not at lower levels. "Planning" implies
choice of output, and if we are to believe in PCT's canonical form, there
IS no choice of output at lower levels. Differences in output come from
differences in reference levels and disturbances. If both the total set
of reference levels and the total set of perceptual signals (below "program")
were the same in two situations, and the same set of disturbances occurred,
then the same outputs would happen. This is not necessarily true for the
program level, about which I must confess to some vagueness, but which
seems necessarily to incorporate a choice of output. The program level
seems to involve an imagined world in which the perceptual consequences
of possible action are controlled variables.
Perhaps a discussion of the program level might help reduce my uncertainty
here (and might help us to understand how it relates to Bob Clark's DME
as well--I confess that I always translate "DME" as "program level control"
when I read it, though I am probably quite wrong to do so).
Anyway, to repeat, I think Bill is right in augmenting Bill's description,
but less right in the extent of his criticism.
Martin