<Martin Taylor 940205 13:20>
Rick Marken (940204.0930)
In my post to Bill Leach yesterday (940203.1000) I described the
"Marken effect" as evidence for a transport lag in human control
systems. I just realized this morning that the existence of this
effect is really bad news for those who believe that there is
information about the disturbance in perception.
I don't see why you think so. Or maybe I do see why YOU think so,
given prior comments. But there's no reason to think so. In fact,
it is quite a good example of the usefulness of IT in analyzing control.
Why would there be a different amount of information
in one case compared to the other?
(1) Let's ask a different question first. In open loop "control" of a similar
situation with a repeated and therefore completely predictable disturbance
waveform, how long would it take for the "C"EV to get a long way from
where it should be? Not long, I think.
(2) Now another. When you replay disturbance waveforms in other
experiments, how highly correlated are the error signals from run to run?
Not much, if at all (they are, at the higher frequencies, but hardly at
all at the low, I think).
(3) Now think about the "live" situation. How is the disturbance wave
generated? The "disturber" is in conflict with the subject's control system.
The "disturber" is generating a signal based on what the subject's output
does. It is not independent of the subject's output, and hence not of
the subject's perceptual signal.
When you run the same disturbance waveform again, it is now more or less
independent of the subject's perceptual signal, especially at the lower
frequencies where control is relatively good and the information from
the disturbance is effectively used in control. The disturbance is more
uncertain, and the control poorer.
Everything about the environment
(the disturbance waveform) and the subject is the same in both cases;
all that is different is that the disturbance waveform is generated live in
one case and replayed in the other. But there is absolutely no way the
subject can tell whether the disturbance is being generated live or not;
the subject cannot perceive the cause of the disturbance.
Can you tell when someone is controlling something in conflict with you,
as compared to them randomly affecting it? Isn't that the basis of the Test,
that you really CAN tell? If you push, and the "disturber" pushes back,
then you pull, and the "disturber" pulls back, you somehow get the impression
that there is a connection here. In your experiment, I might guess that
if there is an initial transient disturbance you would perceive much the
same whether the disturbance waveform was generated live or from a recording,
but I don't think it would be long before you would know the difference.
And yes, "noise" is at the bottom of the difference. Without noise, you
would always get the same error signal every time you replayed a disturbance
waveform (see 2 above). And I think you would not find the Marken effect.
Also, I don't think you will find the Marken effect when the two systems
(subject and "disturber") are in conflict severely enough that the subject
cannot maintain good control.
How would IT explain the existence of the "Marken effect"?
Simply put, the disturbance waveform is more predictable when it is
generated live than when it is recorded, because of the imprecision of
perception, comparison, and output functions involved in the control loop.