[From Bruce Abbott (951027.1600 EST)]
Chris Cherpas (951027.0930 PT)]
[re: >Bruce Abbott (951026.1130 EST)]
I suppose in behaviorese, the "internally specified perceptual states" are
translated as "the rat is always right" in the sense that what is
reinforcing for a given behavior is specific to that organism/behavior
at that specific moment in time. The momentary value of some event
would be said to be a function of "establishing operations" (see Jack
Michael's many writings in JEAB and elsewhere), such as deprivation or
aversive conditions.
Control-systems analysis suggests, as I mentioned in another post, that what
RT calls a reinforcer is anything that, when made the consequence of an
action, reduces error in the system producing the action. If this is so,
then before something can "reinforce" a response, it there must be error.
An "establishing operation" is anything that produces this error. For
example, before food will serve as a reinforcer, there must be error in the
food-intake control system; this can be produced by depriving the rat of
food. Aversive stimuli such as footshock serve produce their own
establishing operation because the shock itself produces error in in a
pain-control system by raising the perception of pain above its reference
level, which I presume is zero. In maintained behavior, the actions
observed (e.g., lever-pressing) have been arranged through the scheduling
apparatus to reduce the error in some such system. The "establishing
operation" acts as a disturbance to the system that, when compared to the
reference level for that system, is registered as an above-zero error
signal; this signal then activates actions (e.g., lever-pressing) that tend
to counter the effect of the disturbance on the controlled perception, thus
reducing or eliminating the error. The observed increase in output under
these conditions is labeled "reinforcement." Note that reinforcement is not
a property of the "reinforcer," but a label applied to the increase in level
of output as a control system acts to counter a disturbance to its
controlled perception. Deprivation and shock (or threat of shock) are
establishing operations; both disturb a controlled perception; the
appearance of food or reduction of shock (or of the threat of shock) are
reinforcers only in that they tend to counteract the disturbance.
ba:
Moreover, purposive
systems constructed along the lines suggested by control theory have been
built, and these behave exactly as a system guided by an internally
specified goal or reference should act.
cc:
If you've seen a simulation that produces cumulative records typical of
FI, FR, VI, FR, and CONC-VI-VI schedules, I am indeed impressed.
Bill Powers has written a simulation that mimics some of these patterns, but
we currently lack data on the controlled variables required to build and
test a more adequate model. I am in the process of setting up to collect
these data.
If you want to see an impressive simulation already written you should take
a look at Bill's CROWD demo, which models several types of "following"
behavior. The simulation implements several simultaneously-active control
systems to produce highly realistic-looking behavior as, for example, when
"ducklings" attempt to follow "mom" or another duckling through a dense
field of obstacles toward a distant goal. Reinforcement accounts of this
behavior sound downright clumsy and ad hoc against the elegant simplicity of
the control-system model that actually mediates this performance.
cc:
Whether its internal references or external stimuli, it seems the problem
is still one of how to find the functional relations between these and the
classes of behvavior that maintain certain values. Out of almost total
ignorance of PCT (see below), it appears to me that instead of "operant" units
one has "control system" units of analysis.
The observed functional relationships emerge naturally from a proper
description of the underlying control systems and their organization. The
trick is to find out what perceptions are under control and how the systems
controlling those perceptions are tied together. The problem currently
facing us is that the needed research has barely begun. Wanna help?
Dennis Delprato has suggested that the proper unit of analysis for behavior
may indeed be the control system. You might want to ask him to elaborate.
Regards,
Bruce