[From Rick Marken (930406.1800)]
Gary Cziko (930406.2155 GMT) --
It seems that nobody north of the U.S. border (or elsewhere) wants to try
to reconstruct the disturbance to the perception vectors you sent.
I noticed.
I know my credibility may be questionable
What about mine; the looney who says that control systems can counteract
disturbances without any information about the disturbances.
Toward the end of the second ("more interesting") perception vector you
sent, we find this:[I added the actual disturbance values in the d
column -- RM]
d
10.670372 44
12.904643 47
15.331556 49
15.498774 50
17.244138 52
17.450351 54
18.964253 55
21.933944 57
23.246449 57
23.635379 58
27.322352 59
27.469903 59
29.150621 60
30.24628 60
34.869933 59
37.148115 59
40.032567 59
So during this time the perception is growing monotonically. Now since you
said the reference level is zero and we have a proportional control system
which does not compensate for all of the disturbance, doesn't the
perception vector here tell us that the disturbance vector must also be
rising monotonically?
If so, there must be some information in the perception about the
disturbance (ouch, that hurt).
There, there. It'll be OK. Try taking 17 of the little disturbance numbers
that I have written in next to the corresponding perceptual values. There
is indeed a monotonic increase up until the last five values (about 30% of
the entire disturbance). At that point, the disturbance moves OPPOSITE to
the output -- the output continues to increase (at its fastest rate of
increase) while the disturbance starts to DECREASE. If you compare the
disturbance values to the intergral of p (Martin's Mystery function) the
deviation of actual from expected disturbance values is even more impressive.
Comparing the size of disturbance to perceptual values you can see that
control is, indeed, not impressive during this phase of the disturbance.
Still, control would be MUCH worse than it is if you had added the
outputs of the Mystery function to the perceptual input instead of the
inputs that were actually added during closed loop operation of the
control system. Still, I can get samples of p values that reflect much
better control (p closer to r = 0) that result in predicted d values that
deviate just as dramatically as these do (often more so) from the actual
d values.
Please tell me what I did wrong. If you can't, I will try to come up with
the exact values of the disturbance.--Gary
You done fine.
I think I'm ready to drop this topic now (did I hear a huge collective
sigh of relief). It is clear to me that people will believe what they
want to believe and there's not much one can do about it (I can already
hear the comments on this one --"well, the perception did have SOME
information about the disturbance; at least for the first 70% of the
time; maybe if we put a lag in or something we could reconstruct d from
p, etc etc). It's hopeless.
I just hope that, whatever people believe about this, they start
treating people (and studying them) as though they were perceptual
input control systems (which they are). That is, treat people AS
THOUGH they had no information about the variables that influence
the variables they are controlling (because they don't).
Best
Rick