[From Bill Powers (941028.0845 MDT)]
Op Cond experimenters and modelers:
I've been thinking about a novel experiment to explore the question of
learning of behavior. The basic idea is to create a field, one- or two-
dimensional, of places where an animal could respond, with a gradient of
reinforcement extending across the field. In the one-dimensional case,
an example would be a long position-sensitive bar enabling a computer to
tell where on the bar the rat was pressing on it. The number of presses
required for a reinforcement would be a function of x, the distance from
some point x0 on the bar. So this would be an FR(|x-x0|) schedule.
The question to be answered is whether the rat would find the best place
to press by a systematic method, by a reinforcement-based method, or by
the E. coli method. A sub-question would be whether we could distinguish
a reinforcement-based method from an E. coli method.
It now occurs to me that we could combine this experiment with one I had
wanted to do while in graduate school in psychology at Northwestern (I
studied Hull there, and one of my professors, Janet Somebody, married
Spence. Enough said.). Being told to find a more conventional experiment
for my Master's thesis was one reason I quit after one year.
Suppose we mount the bar on a lever attached to a potentiometer. The
lever (counterweighted) can be pulled down through an arc of perhaps 90
degrees against a mild spring. The potentiometer can be connected to the
game-paddle input of a computer, or to an A/D converter.
The gimmick is that the rat is rewarded for bringing the lever to a
certain angle within some small range and holding it there for some
length of time. If the reward dish is far enough from the lever the rat
will have to let go to collect the reinforcer, thus letting the lever
rise to its home position for the next trial.
You can see that successive modifications of this experiment are
possible, leading to a rat making a cursor track a target.
But we could begin by using this experiment to explore reorganization,
by making the length of time the bar must be held still depend on the
position in which it's held still. The only problem is how to do the
shaping without giving the show away.
Perhaps the long bar is the best idea, if we can figure out how to do
it. Or maybe someone else can come up with a more practical method.
Some day I hope we can try out analog control as an alternative to the
"rate of behaving" measure which seems to be the only thing studied in
EAB.
···
----------------------------------------------------------------------
Best,
Bill P.