[From Bruce Abbott (960717.1730 EST)]
Bill Leach (960717.1433) --
[From Bruce Abbott (960716.2005 EST)]
O.K., seven hours and no bites on my "troll" post. Second verse, same as
the first:But i did bite!
Ah, sorry Bill, so you did. I overlooked it! And now that I've read it, I
agree.
If the rat only pressed the lever following the receipt of a shock we would
have a "normal" PCT control situation (that looks very much like a
stimulas-response situation). However, the rat creates a control loop that
sets a reference for pressing the lever periodically (appearently). If such
a system exists, a reference for "receiving no shocks" would not itself
create an error signal to set a reference for perceiving the lever being
pressed if shocks were not received.
A more likely version would be one in which the rat uses a "leaky
integrator" perceptual function whose current value is a time-weighted
reflection of the recent shock rate; Presumably the rat would keep
responding until this perceived rate fell to some value near zero.
Some, I think, interesting questions:
1. Will the rats continue the lever pressing operations if the rat and
lever are moved to a different environment?2. If changes are made to the existing environment will the rat continue to
press the lever?3. If there are two levers and one delivers food, will the rat press the
shock lever regularly but press the food lever for some food intake rate?
Some provisional answers:
1. Whether the rats would continue pressing would depend on how similar the
new environment is to the one in which training occurred. If the
environment is too different, there is little reason to expect that the same
"world model" would apply. On the other hand, I have switched rats to
(nearly) identical chambers in mid-experiment with no detectable disruption.
2. Probably again depends on the degree of similarity, and also on the
speed and magnitude of change. You might have seen something similar when
naturalists have photographed birds nesting inside a hollow tree -- by
_gradually_ cutting and widening an opening into the tree, and _gradually_
increasing the level of illumination, they have been able to photograph the
birds and their chicks without any disruption of the birds' normal
behaviors. Evidently each change is too small to trigger any alarm.
3. Yes, but only if shocks are not too strong. In that case "defensive"
control systems tend to be asserted, suppressing intake control systems.
Bill Leach (960717.1348) --
Looks to me as though we need to resolve this difference of opinion
concerning the meaning of the terms if we are to successfully discuss model
based control, yes?I think that the issue of model based control is complex because there are
several way in which the concepts of modeling may be used and these uses
will differ markedly.
Bill, your discussion of this issue was superb! Model-based control is
indeed a broad concept that covers a number of somewhat different
strategies, all involving acquiring some kind of knowledge about the world;
the term should not be restricted only to the modeling of how the CEV will
be affected by the system's actions, or how the CEV is likely to change over
the short run. My guess is that those rats continue to respond, and I
continue to make those amulets and send those checks in to the IRS because
the rats and I have acquired specific models of our worlds, and those models
tells that there will be dire consequences if we do not respond. At the
same time, we are responding at the present time to present perceptions
(even if imaginary).
Oops! There goes another empty carton of milk! Sorry, I had perceived
(imagined) that it was full, and set my reference for lifting force accordingly.
Regards,
Bruce