Bruce's experiments; PCT & reinforcement

[From Bill Powers (950607.2010 MDT)]

Erling Jorgensen --

(I didn't look up the spelling -- "-son?")

Sorry you won't be coming to the meeting. Your help last year was most

They say that before you name a cat, you should go to the back door and
yell the name a few times as loud as you can.



Bruce Abbott (950607.1155 EST) --

     In the absence of responding, each shock terminated automatically
     at 15 seconds. For a rat able to keep the average shock length at
     150 milliseconds (0.15 seconds, control reduces the disturbance
     (shock) to 1% of its uncontrolled value. By your definition this
     is excellent control, yes?

Depends. A formal observational definition of a reference level for a
controlled variable is that level of the variable at which the system's
action just drops to zero. To find the reference level for the shocks,
you'd have to do something to reduce them until the rats wouldn't bother
to act to avoid them. Then you could judge the loop gain by seeing how
much you would have to raise the shocks again to get the maximum amount
(speed, whatever) of action against them. That would define the range of
control, and you could then guess how important a given level of shock
is to the rat by measuring the ratio of error to effort. I suppose this
would entail adjusting the shock current, since the rats have a minimum
reaction time and you don't know if they've reached it.

     Hmmmm, this sounds oddly familiar. Could it be . . .

     In the avoidance experiment the rat performed on a Sidman shock-
     avoidance schedule on one session and the actual temporal pattern
     of shock delivery was recorded (rats on this schedule occasionally
     make mistakes and receive shocks); this pattern was "played back"
     on the next session, in which the rat had no control over shock
     delivery. As with escapable versus inescapable shock schedules,
     the rats failed to resist the disturbance when the apparatus
     switched them from avoidable to unavoidable shock schedules or vice
     versa. The key here is that during training the rats had learned
     that the shock frequency was the same whether they controlled shock
     delivery let the apparatus determine when shocks would be

     Will that do?

Depends. Did the rats demonstrate that they knew how to switch from one
schedule to the other? For example, if you doubled the recorded shock
rate, would they then switch to the controlled condition, and if you
halved the recorded shock rate would they switch to the uncontrolled
condition? I'm presuming that the rats would show a preference for the
lower shock rate, even if they weren't controlling for controlling. If
they didn't switch to the lower shock rate, maybe they never understood
the switching concept.

Also, I presume that whenever the "controlled" condition appeared, the
rats reduced the experienced shock rate by doing whatever they had to
do. Perhaps their failure to differentiate indicates that whatever the
shock rate was in the controlled condition, it was below the level they
considered an unacceptable amount of error. In the experiments I
analyzed, the rats were holding the shock rate as low as a few shocks
per hour. We could explain the failure to switch conditions as
indicating that the shock rate achieved in the controlled condition was
experienced as zero error. The same shock rate, recorded, would also be
experienced as zero error, or small enough not to bother with.

If the
rats were hovering over the bar in the "control" condition, they must
still have been experiencing a pretty large error.

     What? If your participants keep their hands on the mouse during
     the compensatory tracking task, "they must still have been
     experiencing a pretty large error"? Hmmm.

Depends. If they're leaning toward the screen and moving the mouse as
fast as they can, I assume they still think the errors they're seeing
are pretty large. When you reduce the amplitude of the disturbance, so
the cursor only moves a fraction of an inch from the target at most,
subjects relax and don't try so hard. That's just an impression, of
course. We can judge how much error they're perceiving by seeing what
amount of error will just lead them to stop moving the mouse.

When I test for
control, I want it to be successful. If big errors still remain, we're
not seeing very successful control. So the rats must be trying to
distinguish between hardly any control and none at all, which, it seems
to me, makes telling the difference unneccesarily hard.

     Even in the escapable versus inescapable shock experiment the rats
     had no difficulty distinguishing when they did and did not have
     direct control over shock duration. It's pretty obvious: in the
     escapable shock condition the escape lever is extended and, unless
     you press it, shock just keeps goin' and goin' and goin' (up to 15
     seconds). In the inescapable shock condition the lever is
     retracted (so there is nothing to press) and the shock stops after
     a brief, fixed duration, independent of behavior. z

Depends. It's pretty obvious to you, but it is obvious to a rat? Suppose
that to a rat, a 100 millisecond shock (of the intensity used) creates
95% of the maximum possible sensation of being shocked, and any shock
longer than 300 milliseconds raises the perception to the maximum value
the perceptual signal can have -- 100% of the maximum sensation. That's
a 5% different in perception. So whether the shock lasts 300 msec or 15
seconds makes no difference in the perceptual signal (or the error, or
the output): it's saturated.

I'm not saying that's true. But without testing, how can you rule it
out? The only real way to find out if the rat is controlling
successfully is to find the level of the controlled variable at which
the rat would just cease trying to correct it, and then the level at
which it couldn't try any harder.

     That depends on whether they were in the escapable or inescapable
     shock condition at the time. In the former condition, the shock
     duration was clearly under the rat's control.

Depends. The rats were clearly _trying_ to control the shock when they
could. They succeeded in getting its duration down to 100-300
milliseconds. While I don't know anything about a rat's reaction time to
a shock, that strikes me as getting close to the fastest possible
reaction time. This could be tested, of course: you could double the
shock intensity and see if they reduced the duration to 50-150
milliseconds, and if you halved it they increased it to 200-600
milliseconds (not literally proportionally, of course). The point is to
find out whether they're trying as hard as they can and still
experiencing a very large error, or whether they're successfully
reducing the shock duration to just the duration they prefer. To
experience maximum control means experiencing a match between the
perception and its reference level.

     No real control system can keep the controlled perception exactly
     at reference at all times regardless of disturbance.

True. The problem is how to decide what is a "large" error and what is a
"small" one. The best way to judge is in terms of the whole control
range. The control range can be determined by seeing how much error is
needed to produce the maximum possible effort in opposition to it.

If you see that the effort is at its maximum, then you know that you are
on the edge of the controller's range of control -- or perhaps beyond
it. If you're beyond it, increasing the disturbance further will not
increase the effort further; it can't get any larger. When that is
observed, you know that the system is trying to control, but is not
actually able to control. The only way to verify that the system is
really controlling is to vary the disturbance and see if the effort
varies in opposition to it. If you find that an increase in shock
intensity results in a substantially shorter reaction time by the rats,
then you know that they were controlling, rather than just trying to

An analogy. Suppose you were told that by pressing a lever with one
hand, you could cause the immediate withdrawal of a plunger on which is
mounted a needle that is inserted under one fingernail on your other
hand (which is immobilized). At the beginning of each trial, the plunger
moves forward in 5 milliseconds, driving the needle 1/2 centimeter into
the interface between fingernail and finger tissues. If you could react
to press the button in 250 milliseconds, would you consider that you are
in control of this experience? If you could keep your free hand just
over the button and by intense concentration reduce your reaction time
to 150 milliseconds, would you now feel in control?

I wonder if the rats ever try hammering on the lever as fast as they
can, and if the experimenter has to think of a way to keep them from
doing that so he can measure the true reaction time. I think you might
come up with that strategy as a subject in the above experiment. I
wonder if the rats ever vocalize when they get shocked, and what a
recording of that vocalization might sound like if slowed down so the
mean pitch was, say, 500 hz. I wonder if you might vocalize in the above

     My study demonstrated that rats can indeed compare overall shock
     exposure in the two conditions and, when those exposures are the
     same, do not act to control which condition they are in.

Try that out with my hypothetical experiment above. The needle could be
withdrawn by you in one condition, and in the other condition would
automatically withdraw after 200 millisec. Would you be thinking in
terms of whether you have control or not? I rather think that any
cognitive functions you might have would be pretty well shut down. This
tends to happen under torture.

This says either that
they had no preference for being able to control the shocks, or that
the difference in degree of control they could obtain in one condition
relative to the other was so slight that they couldn't tell the

    Nice try, but no cigar.


     Right? I think I've shown that the second possibility can be ruled

I don't think you have ruled it out, unless there is some other
observation you made that you haven't told me about. We don't know that
the rats could have shortened their reaction times if they had wanted to
reduce the duration even more. Which is better, having no control at
all, or trying as hard to control as you can and failing?

     O.K., O.K., you caught me. All this time I've been part of a top
     secret, high-level plot to undermine PCT and reestablish
     traditional reinforcement theory as the "top dog" in the field of
     learning and behavior

I don't see that as the problem. What I see is the Necker Cube problem,
very difficult to overcome. Consider this:

     I believe that you have drawn an incorrect conclusion from one
     graph showing what happens to asymptoticly maintained response
     rates as the ratio requirement of a ratio schedule is varied.

This way of saying it makes the assumption that it is the response rate
that is being maintained by the reinforcement rate. However, when for
example the schedule is changed from FR1 to FR2, the first thing that
happens is that the reinforcement rate drops in half. It will remain at
that level unless the behavior rate changes. As the behavior rate
increases, the reinforcement rate will increase, because it is dependent
on the behavior rate. What you end up with is a reinforcement rate that
is somewhat less than before, and a behavior rate that is almost twice
as high as before.

Clearly, it is the reinforcement rate that is being maintained by the
behavior rate. The direction of the dependency is easy to verify in a
number of ways, a simple way being to examine in detail what happens
immediately after a change in the schedule. Another simple way is to
examine the apparatus, which will show exactly how variations in
behavior cause variations in reinforcement rate. On a ratio schedule, it
is impossible for the reinforcement rate to be anything but the behavior
rate divided by the ratio.
     My performance may be such that I am reducing the effect of the
     disturbance on the controlled perception, but not really doing a
     good job of it (overshoots, undershoots, etc.) Do I have control
     or not? A given performance lies along a continuum of control, from
     total out-of-control to extremely well controlled.

I agree. The best sense of control comes when you're keeping the
controlled variable exactly where you want it. Loss of control is sensed
when you're trying as hard as you can and the errors are still
unacceptably large. As you say, there is a continuum of "being in
control;" it's not an either-or thing. That's what I was talking about
above: trying to determine where the rats really were on this continuum.
I don't blame anyone for not making this determination when the
experiments were done, but the determination was not made.

     Would "have the means to control" provide the needed distinction?

It would certainly help.

     Also, this difference in meaning suggests that distinctions can be
     drawn among three situations in which a perception is not in
     control: (a) you do not have the means of control and therefore
     are not attempting to control it, (b) you have the means of control
     but are not attempting to control, and (c) you have the means of
     control and are attempting to establish control, but have not yet
     succeeded. In cases (a) and (b) the variable is "not in control,"
     in case (c) it is "out of control."

Good distinctions.

     In cases (a) and (b) the individual is doing nothing, in case (c)
     he or she is engaged in a frantic effort to achieve control.

Right. We can refine this even a little further, in that what seems like
"frantic" action to us may still be within the control range, although
nearing the state of zero control. The step beyond (c) would be "futile"
efforts, meaning efforts that are too small to oppose all of the
disturbance, yet that are the maximum efforts that can be produced.
Going back (hard to keep up with all this):

Bruce Abbott (950606.1800 EST)

     For the "blind tracking" test you have them wear a blindfold, but
     you discover that two of your subjects are cheating by peeking
     under the blindfold. What should you do? You should:

     (a) allow them to cheat because this is their preferred mode of
     control over cursor position.

     (b) prevent them from cheating, perhaps by taking steps to assure
     that they will not be able to peek.

     (c) abandon the research because you have determined that the
     question is silly, as the nonvisual mode of control is not
     the one the participants would select for themselves. As for me,
     I'd shave their backs. (;->

I would ask them to reconsider their agreement to be part of this
experiment, explain what I am trying to find out, and ask if they are
willing to play it straight. If they are, we can go ahead; if not, I can
get other subjects. I would not shave their backs.

Actually, what I DID was to turn off the display, so the question of
peeking never arose.

Those rats who flipped over on their backs would be worth studying. Were
they smarter than the other rats, or just lucky enough to have fallen
over at the right time and alert enough to realize the effect on the
shocks? If I found myself starting to play "behave, dammit" with a bunch
of rats, I think I'd have to pause for a little self-examination, and
maybe even have a critical look at previous data. And if I overlooked an
interesting phenomenon because I wanted to get on with my idea of what
was important, I'd have to turn in my serendipity license.

Being lazy, I would like to do experiments with rats in which they would
wander over to the apparatus now and then and work it to get something
they want, then go on about their business. Of course this would mean
that whatever they were doing must have seemed to the rats worth doing.
Associating shocks with the apparatus wouldn't help achieve this.

Bill P.