[From Chris Cherpas (951207.0933 PT)]
[re: > Bill Powers (951206.1500 MST)]
A Conc VI VI is meant to eliminate the "noise" of differentiated and/or
discriminated patterns of behavior found in other schedules (FI scallops,
Ratio post-reinforcement pauses), so you can concentrate on reinforcement
value (of course Shimp and others would argue that interresponse times,
IRTs, get shaped/discriminated). The concurrency helps eliminate floors,
ceilings, nonlinearities -- i.e., the relativity provides a more sensitive
perspective.
If linear VIs are used (I've explained them before, but can provide a
state diagram if we get to that level of detail) then rate of rft
deliveries should be fairly constant. If it is assumed that rate
of rft is the controlled variable, then regardless of how the
relative rate of rft across the two Conc VIs is varied, there
shouldn't be any particular pattern of behavior that shifts along
with it, according to this (admittedly uninformed, since I'm no
PCT expert) PCT view. In other words, there would be no disturbance.
From the view of melioration theory (a theory of reinforcement) the
subject should shift systematically with such changes, because of
discrepancies in local rft rates (# of deliveries while responding
in each of the two separate concurrent components). After several
such shifts, melioration would predict matching.
Here's a rough design:
The subject sits before a computer terminal. There are three places on
the screen where the subject should mouse-click. One each side operandum
is what might be depicted as a weapon that shoots at a target. When a
VI timer has not timed out yet, clicking on the operandum means that the
little cartoon projectile misses the target; whenever a rft is set up
by the VI timer, clicking on the weapon makes the projectile hit the
target with appropriate visual/sound effects. The center operandum is
for switching to the operandum on the other side. When one side is
operative, only it and the center/changeover operandum are visible.
Clicking on the center operandum turns off the current/old side operandum,
makes the other one available, and turns itself off until the first
click on the newly activated side (at which time the center operandum
is now available). When a hit occurs on the target, all operanda are
dis-abled and turned off (i.e., during "reinforcement").
The schedules could be such that there is an average probability of
clicking and hitting every 30 sec. So, the first set of schedules
might be VI60"-VI60". After a stable baseline (we could use a formal
criterion, as used in statistical process control) on a pair of
schedules, we could switch to VI120"-VI40", then VI32"-VI480",
then VI60"-60" again, then VI40"-VI120", then VI480"-VI32,...
you get the idea.
Melioration would make a quantitative prediction for function running
though the points found at stable performances for the sets of
schedule pairs (matching). A quantitative model tracking the dynamic
shifts that would occur with the schedule shifts has not been
definitively stated within melioration theory, but analogies to
Rescorla-Wagner have been done by Will Vaughan, and I've put together
a kind of moving average calculation of the discrepancy between
local rates of rft, which presumably causes the shift in rates of
changeovers to the newly richer schedule...
???,
cc