[From Bruce Abbott (2000.12.09.1050 EST)]
Bill Powers (2000.12.08.0959 MST) --
Attached is a program called "furnace4.exe" which will run on a PC. . . .
This program is called Furnace4 for the reason that it started out as a
program to implement the furnace control system that was talked about some
time ago.
Cute -- I like it!
This could use a little context. Earlier I had presented my little story of
Fred and his furnace, and eventually Bill suggested that I write a model
that would behave as "Fred's furnace" was supposed to behave. The next day
I sent Bill a working model written in Turbo Pascal that worked as advertised.
Bill and I discussed this model and then Bill presented an alternative
model. The only substantative difference between the two models was what
can be called the "selection criterion" -- the condition that would trigger
a switch to a "new" means of generating heat. The criterion in my model was
four consecutive commands to an injector that were not followed by the
"flash" of burning coal dust. The criterion in Bill's model was a certain
difference between the reference for house temperature and the actual
temperature. This difference was set large enough so that it would usually
occur only when the normal means of generating heat had failed.
Both models comprise a pair of control systems organized in the same way.
One system is a typical heating system controlling house temperature. The
other system controls the functional status of the output mechanism
(furnace) of the first system, attempting to keep it operational against
disturbances consisting of injector failures by trying different injectors,
when a failure is detected, until a means of producing heat is found that
works. The two models differ only in how the lower system is implemented.
Bill asserts that neither mechanism involves reinforcement. I say that both
of them will behave in a way that would warm Fred's heart (not to mention
his house).
Furnace 4 presents an elaboration of the same approach, recast now as a rat
searching for food. It is not essentially different from my original
proposal, except for the "selection" criterion (an important difference, to
be sure, but not one that completely destroys the analogy). However,
attempting to relate the behavior of this model to so-called "predictions"
of reinforcement theory, Bill has once again committed certain errors that I
had hoped we could avoid by first arriving at a common understanding about
what, in the model's behavior, corresponds to what, in the EAB describtion
of its behavior.
I had hoped that we could get the track laid _before_ we tried to take the
locomotive for a test drive. Now we're going to have to pick it up somehow
and try get it back on the track.
I'll try to be more explicit about all this in a later post, when I have a
bit more time available for writing.
Bruce A.