[From Bill Powers (960926.1430 MDT)]
Bruce Abbott (960926.0820 EST)]
... if the participant is trying to play the
experimenter's game and get to the goal, then what the experimenter views as
an error (moving away from the goal, "colder") would also be viewed by the
participant as an error, and the participant will change course whenever
that error occurs. Under this condition, to state that "the experimenter is
_not_ a 'source of information as to error' " (emphasis mine) is, strictly
speaking, incorrect.
We should all kick ourselves. KR is effectively part of the controller's
input function. I got hung up on higher and lower levels of control, which
are no doubt involved, but the obvious answer for the higher-level system is
that the external experimenter providing KR is simply part of the (crude)
function that translates the state of the controlled variable into a
perception. When there's no other path through which the controller can know
the state of the controlled variable, KR is the only way to go. I am
assuming that "KR" stands for "experimenter-supplied information about the
state of an otherwise unobservable controlled variable."
The loop consists of the controller's actions that affect the controlled
variable, the experimenter's perceptions of the state of the controlled
variable, the experimenter's translation into words which he or she thinks
describes the state of the controlled variable, the controller's perception
of the experimenter's words and translation into the controller's perception
of the state of the controlled variable, the controller's comparison
process, and back to the controller's action again.
There's no basic reason for KR to be given in cryptic or binary form: the
experimenter might say "The target is 10 degrees to your left, 15 degrees,
10 degrees, 8 degrees, 4 degrees, 2 degrees, you've got it." This, of
course, would enable the controller to control far more accurately than
binary information that lacks directional content, as in "hot-cold." The
controller could even decided to point 10 degrees to one side of the target,
in which case the experimenter's verbal output might go "20 degrees left, 15
degrees, 10 degrees, 8 degrees, 9 degrees, 9 degrees, 10 degrees, 10 degrees
...." All the experiments has to do is keep up an accurate translation of
what he sees into what he says.
The experimenter could also insert disturbances, and control the
controller's action. Whatever the position of the finger, the experimenter
subtracts 10 degrees before reporting it. So the experimenter will end up
reporting a constant angle of 10 degrees left when the controller's finger
is on the target, and the controller will move the finger until the
experimenter is reporting 0 degrees -- at which time the finger will be 10
degrees to the right of the target.
Also, the final state of the controlled variable will be the state as the
experimenter sees it, not the controller. If the controller is trying to
keep the finger 10 degrees to the left of the target, and the experimenter
is standing on the other side of the target, when the controller opens his
eyes he will see the finger 10 degrees to the RIGHT of the target.
Interesting way to play the rubber-band game.
Best,
Bill P.