Bruce's experiments

[From Bill Powers (950608.1400 MDT)]

Bruce Abbott (950608.1105 EST),(950608.1300 EST) --

I guess I was letting my imagination run away with me. I am reassured by
your description of the "shocking experiences" of the rats, and also by
the fact that you did so many studies ancillary to the main one, to
determine critical facts. Despite all my nit-picking, I am impressed by
your talents as an experimenter, a _careful_ experimenter who takes
little for granted.

I should mention my own major experience with shock, which was in the
Navy while inside a cage housing a powerful transmitter that I was
troubleshooting as part of a class. The POWER OFF sign was prominently
posted, and I mistook the map for the territory. Squeezing into the
cage, I laid my bare elbow on the wrong terminal of a 0.5 microfarad
capacitor charged to 3000 volts. My next participation in the world was
about 10 minutes later. This was probably the point at which control
theory began to germinate in my mind: whatever firm beliefs I had had
about human nature were blasted out of existence and I had to start
over. Well, that's a theory, anyway.

RE: Catch-22

We're talking about two different experiments here, aren't we? One in
which the rats could, in principle, reduce the shock rate to zero, and
one in which no matter how quickly the rats responded, they would get at
least 100-300 milliseconds of shock. From the control standpoint these
are very different situations.

In the first case, there is some reason to think that the rats were
maintaining the experienced shock rate very nearly at their reference
level for it, and that if the remaining average shock rate had still
been onerous to them, they could easily have reponded more frequently
and reduced it further. In other words, there is no question that the
demands on their response capabilities were anywhere near the limits of
performance.

In this case we have two possible explanations for failure to prefer the
condition in which the rats have control. As you indicated to Bill
Leach, "a finding of no-preference could indicate either that the rats
were indifferent between the conditions or that they had not learned how
to switch from one condition to the other." Since you demonstrated that
the rats were able to switch conditions when given a warning signal,
that leaves only the "no-preference" option as a reasonable alternative.

It is not clear from this, however, that the rats would not prefer the
controlling condition over the no-controlling condition when there is
any significant loss of control in one condition. In the controlling
condition, the rats clearly respond at a sufficient rate to reduce the
shock rate by a large factor, keeping it close to the presumable
reference level. It is interesting that they continue to improve their
control of the shock rate until it is only a few per hour: evidently,
shock is highly disliked when it occurs too often, and "too often" is
not very often.

However, the required rate of lever-pressing is very low, and the shock
rate is very low. The rats would not, I venture, experience any
signficant loss of control because the error would be close to zero. In
the other condition, the rats would experience the same low error
without having to press the bar every 20 seconds or so; the error would
still be close to zero. This would not be experienced as loss of
control, either, would it? This would remove any basis for a preference
even if rats did prefer control over loss of control.

···

-------------------------
I should pause here to remark that I would be very surprised to find
that rats would have a preference for an abstract condition called "loss
of control", or its converse, "being in control," apart from any lower-
order consequences of losing control. This would give rats capacities to
control at the principle level perhaps greater than we could reasonably
expect. But I am interested in how we could use the data to support my
prejudice, in a way that follows logically from the data. So I'm looking
for loopholes that would have to be plugged to permit us to reach the
conclusion that I fully expect we are justified in assuming. It's a good
idea to plug loopholes, because occasionally they turn out to be drains
down which our comfortable assumptions disappear.
------------------------
In the second experiment involving shocks, we find a different
situation. Here the rats always get shocked, and the only dimension in
which they can control is the duration of the shock. Once again, the
rats show no preference for the controllable condition over the
uncontrollable one, and we must again ask why.

The loophole here is in what the rats would consider to be "in control"
or "losing control." You point out that in the nominal controllable
condition, the rats reduce the shock duration to about 1% of what it
would be if they did nothing. The first crack in the argument appears
when you ask what would have happened if you had made the automatic
cutoff point 150 seconds rather than 15. You would then conclude that
the rats were using _the very same behavior_ to reduce the shock
duration to 0.1% of what it would have been without control. And if you
had reduced the cutoff point to 1.5 seconds, and the rats continued to
limit the duration to 0.15 second, now it would seem that the control
has left a 10% error in shock duration -- still without any change in
behavior.

What this would tell us is that we are using the wrong basis for judging
the quality of control. In fact, you now report several side-experiments
that support this conclusion:

     The 1.0 mA level is close to the minimum value that will produce
     rapid, reliable escape behavior. At 0.5 mA the escape latencies
     are longer as the rats are likely to be spending more of their time
     exploring the chamber, grooming, and so on, and thus tend to be
     some distance from the escape lever when the shock occurs. At 0.25
     mA they may not acquire the escape response at all.

So evidently, at 0.5 milliamp the rats relax their efforts to respond
quickly and take longer to press the lever. At 0.25 milliamp they may
not press the lever at all. And at 1.0 milliamp they respond as quickly
as they can. As you indicate, duration should also be taken into
account, but presumably these were all shocks of the same duration.

So for shocks of whatever this duration was, we find that the fastest
response comes for a current of 1.0 milliamp. The speed of response is
then near the upper limit that the rats are capable of producing. If
either the current or the duration were further increased, the rats
could probably not respond in any significantly shorter time.

This tells us, or would tell us if we verified some details, that the
rats responding in 100-300 msec are near the upper limit of the range of
control. It also tells us that the intensity-time product is near the
upper limit of the controllable range. From previous data, the
intensity-time product representing zero error is slightly less than
0.25 milliamp times the duration. If the rats were able to keep the
intensity-time product near that lower figure, they would be
experiencing no loss of control. But if they were responding as quickly
as they could, yet the intensity-time product was near the upper limit,
they would be close to the condition of "no control" (or as you propose,
"out of control").

The second ancillary experiment supports these conclusions:

     When the inescapable shocks were about 100 milliseconds or more
     longer than the _average_ duration of the escapable shocks, the
     rats preferred the escapable-shock condition. When the reverse was
     true they preferred the inescapable-shock condition. When the
     shock durations in the two conditions matched within 100
     milliseconds, preference, if any, was below the sensitivity of the
     experiment to detect.

If the rats in the escapable-shock condition were at the upper limits of
the control range (almost zero control), then the main difference
between the two conditions was simply in the total experience of being
shocked. The preference that was indicated was for the least amount of
shocking. This, therefore, was still not an unequivocal test of the
rat's preference for "being in control," or for "not being out of
control."

So neither shock experiment could clearly distinguish between preferring
fewer shocks from preferring to be in control. But either one could have
been modified to improve this discrimination.

In the first experiment, you determined the reference level for
experienced shock rate: the shock rate that occurred when the rat could
easily prevent shocks altogether. You could vary the experienced shock
rate in either of two ways: by shortening the interval between
resettings of the timer, or by requiring more than one lever-press to
postpone the next shock. The more rapidly the rat has to press the
lever, the greater the number of shocks will be experienced under the
controlling condition. So you could determine the upper limits of the
control range. Finally, you could pick a condition comfortably in the
lower range of shock frequency where you know the rat is definitely able
to control and that the shock rate is definitely higher than the
reference level. Using this same shock rate in the no-control condition,
remove any preference for the lower shock rate as a consideration. THEN
if the rats failed to prefer the controlled condition, you could say
that they have no abstract preference for being in control for its own
sake.

In the second experiment you could vary the shock current and determine
the range of latencies. Then you would pick a shock current such that
the rats were definitely not responding as quickly as possible, and
where the experienced shocks were definitely above the reference level.
Again, matching the shock rate under both conditions, you could then
rule out preference for a lower shock rate, and if any preference did
occur, reasonably attribute it to a preference for being in control (or
maybe out of control, who can say?).
-----------------------------------
We see that to make the determination you were trying to make, some
rather fine tuning of the experiment is necessary. There is no way you
could have known that such a stragegy was needed, without control
theory. As it happened, the two experiments demonstrated two extremes of
control: nearly perfect, and nearly absent. The question you were trying
to ask required finding the middle of the range, which could only be
done with an understanding of PCT.
-----------------------------------
     Psychophysical studies show that saturation is not a problem:
     perception of shock duration follows Weber's Law.

Yes, but in the second experiment the limit was imposed by the maximum
possible speed of the action (or maybe the maximum possible error
signal), not saturation of the perception.

     If you are correct, then there is no way to test for control of a
     logical variable when the output is a step function.

It is difficult to separate a higher-level controlled perception from
control of lower-level perceptions on which it depends. You have to make
sure that the lower systems would receive the same disturbances under
all conditions, with the only difference being a disturbance of a
logical relationship. This can be done, but it takes careful work with
the lower-level systems first.

     I don't want you to get the impression that I miss your point.

Never thought so for a second.
-----------------------------------------------------------------------
Best,

Bill P.