Use of Perception

[From Bruce Abbott (960125.1130 EST)]

Rick Marken (960124.1400) --

Bruce Abbott (960124.1530)

Really? How do I control the perception that if I don't stop at red traffic
signals I may get into an accident?

I think you control it in the usual way; you vary your actions as necessary
(pressing the brake, pressing the gas pedal, etc) to keep this logical
perception in the desired state. I guess I don't see your point.

Sorry, Rick, you've "shifted gears" on me. You don't control the perception
that "if I don't stop a red traffic signals I may get into an accident."
It's just somthing you perceive. Pressing the brake or gas pedal does not
serve to control this perception.

We've already covered how a perceptual signal called a "discriminative
stimulus" can be "used" to keep a logic-level control system at a reference
level of "true" most of the time

The perceptual signal represented the state of a logical variable; actions
were varied to keep this variable in the state "true". What we showed was
that the "discriminative stimulus" is not "used" to determine responses; it
is just a disturbance to a logical variable. Responses (which had to be
different on each occasion) varied to compensate for this disturbance and
keep the perception of the logical variable in its reference state.

I deliberately chose this example because we had modeled it and therefore
could agree on what that model contains. Yes, the SD acts as a disturbance
to the logical variable in question; it is not used to "determine"
responses. I am not claiming otherwise. The relevant question is, why is
it attended to at all?

The SD is not controlled, but the perception of its state can be used (by
me, the control system constructor) to improve control over point-gain

I would say that the subject was able to control point-gain rate by
controlling the state of another variable. The subject didn't _use_ the state
of this other perception to improve control of point-gain raten; the subject
controlled one perception Point-gain rate) _by controlling_ another; this is
hierarchical control of perception.

This is not an either-or proposition. If the subject were not aware of the
relationship between the state of the SD and the "active" target, the
subject would not have constructed a control system for point-acquisition
rate that included the logical variable and the perceptual state of the SD.
In constructing this hierarchical system, the subject "made use of" this
relationship to improve control over the perception of primary interest,
point gain rate. If the relationship did not exist or if the subject were
not aware of it, it would not have become part of the system.

Do you really mean to assert that the only perception that affects the
performance of any control system is the perceived current state of the cv?

Worse than that; I assert that perception, including the perceived current
state of the cv (the current state of the perceptual signal), doesn't
even affect the performance of a control system. Rather, the performance of
the control system affects the current state of the perception, forcing the
perception (perceeptual signal) into a match with the reference signal. We
refer to this surprising characteristic of the behavior of closed-loop,
negative feedback systems with the phrase "Behavior: The control of

Perhaps to be clearer about it I should have said "the only perception _of
any relevance to_ the performance . . ., so that you would not have been
able to throw this irrelevancy back at me. But other perceptions are
clearly relevant, as they can affect the _reference_ level and perhaps other
parameters of the ECU in question, as the two-level system involving the SD
demonstrates. The lower-level system doesn't "make use of" the information
provided by the SD concerning which is the currently active target, it just
goes where its reference tells it to. But the person who constructed the
two-level system certainly made use of the perceived relationship between SD
state and active target to build the system in the first place.

Martin Taylor (960124 17:20) got it right:

If ANOTHER control unit has a perception for which the input includes both
the CEV of the first unit and other data, then the output of that other
unit could affect the performance of the first in a variety of ways, not
least by affecting the reference signal value of the first. It could also
affect the gain of the first, or could generate output that disturbs the
perceptual signal of the first...

But WITHIN an ECU, one has to:

assert that the only perception that affects the
performance of any control system is the perceived current state of the cv?

But I'm not talking about WITHIN an ECU! Remember how I got into this mess
in the first place? I said that the organism could "make use of" a
relationship it perceives between its own behavior and the delivery of the
incentive immediately thereafter to improve control over the incentive. The
SD system provides a concrete example of what I mean by this phrase, and
this meaning is consistent with control-system theory.