[From Bill Powers (951121.2100 MST)]
Martin Taylor 951121 17:05 --
WTP: The cat _never_ controls for "the perception of the correct exit
mechanism"
MMT: Do I have to take the long-winded wording EVERY time, and say
"the perception of the CEV that the experimenter knows that if
properly controlled would result in the exit mechanism being
activated"? We'd get even more bogged down in words. You really
have to take some shorthand wordings for what they (should)
obviously mean. Making the point once should be enough, with
shorthand to follow. ... You yourself are not guiltless of saying
things like "controlling the angle of the steering wheel" in place
of "controlling the perception of the angle of the steering wheel"
When I do that I am omitting saying "perception of", not, as you did,
specifying a perception of the wrong thing. Actually I think I would be
more likely to say "varying the position of the steering wheel" and
"controlling the position of the car," omitting "perception of" in the
latter case but properly leaving it out of the former.
Your longer sentence would have been preferable, even if you had omitted
"perception of" from it.
You say "the cat would not have to reorganize if you were careful
to use only disturbances that did not prevent control" under
conditions in which it is taken as a foundational assumption that
control is NOT PRESENT for any CEV so far intuited by the
experimenter, but in which every sequential condition of the Test,
as quoted by Chuck Tucker from you, is fulfilled.
The first step of the test, the main one on which you're concentrating,
is primarily a method of ruling out a false identification. If a
disturbance is applied and it has exactly the effect predicted under the
assumption of NO control, you can stop because you have shown that the
postulated variable is not being controlled. In all your examples, you
are being vague about just what the postulated variable is, the exact
kind of disturbance you would apply, and just how you would predict what
the effect of the disturbance would be if there is no control. And I
think you are interpreting the test to mean that if there is apparent
resistance to the disturbance, the Test has proven that control is
taking place, which is not the case. All it has then shown is that
control is not ruled out, a very different matter.
If you have been unable to rule out control using one disturbance, there
is nothing to prevent you from trying to rule it out by using other
disturbances. The coin game illustrates this process very clearly. You
form hypotheses, apply the test, and rule them out. Sometimes they seem
to be satisfied, but before you venture to predict three different moves
that will yield a "no error" response and three that will call for
corrections (the rules of the coin game), you would be wise to test the
definition of the controlled variable further, by disturbing other
aspects of the situation, until you are satisfied that there is no
loophole in your conception of what is being controlled. Otherwise you
are likely to come up with a confident declaration that you understand
what is being controlled, only to find that when you use a different
disturbance in the denoument, you predict incorrectly. Instead of "no
error," you get a correction, or instead of a correction, you get "no
error."
The test is not a recipe, a set of instructions you follow to get the
right answer. It is a _strategy_ which has to be applied with
understanding of what you're trying to do.
The most difficult session with the coin game that I was ever in lasted
for half an hour. At the end, I gave up: I had to admit that I couldn't
discover any controlled variable. My partner in the game then admitted
that he was changing the target pattern on every trial -- the equivalent
of continuous reorganization. Many times during this session I was
convinced that I had the solution, but each time I would try still
another kind of disturbance and find my prediction wrong. So in giving
up, I came up with the right answer after all. Giving up was exactly
what I should have done.
I still don't know what point you're trying to make, but my experience
with the coin game has pretty much convinced me that the Test can work
even in very difficult situations. The situations you've described don't
look particularly difficult. Perhaps you haven't yet described the exact
situation you're thinking of.
We are obviously talking about quite different experiments if you
start by assuming that the cat is controlling some perception that
the experimenter already has intuited.
That was not my assumption. I merely said that whatever the cat is
controlling, the experimenter can discover it by using the Test. You
presented a case in which the cat appeared to be controlling the
position of the stick when it was actually controlling something else. I
came back by offering additional kinds of disturbances which could rule
out the variable "intuited" by the experimenter. This is what anyone
experienced with using the Test would do -- double check and triple
check to see that there was no other variable that could actually be
under control, with the one first proposed being only a side-effect. The
experimenter, after all, can see not only the stick, but all the things
the cat does that could affect the stick. Any one of those other
variables might be the controlled variable, if regularly related to the
stick position. The experimenter must not only show that one of the
variables is not ruled out by the Test, but that all the other equally
plausible definitions ARE ruled out. Perhaps this should be spelled out
in Chuck Tucker's list. It would not have to be spelled out to anyone
who had played the coin game for a while. At any rate, the experimenter
would most likely eventually discover the mistake, because the
happenstance side-effect could be changed by some subtle change in the
situation, and the Test would eventually fail. But it is better not to
leave that to chance.
···
-----------------------------------------------------------------------
Bruce Abbot and Rick Marken:
RE: Newton's Law
I left out one important detail about Newton's model. The concept that
each particle experiences a force due to each other particle, according
to a specific quantitative expression, predicts that the force between
homogeneous spheres will be equivalent to the force between point-
objects, and will obey an exact inverse-square distance law. This is
what was verified by the planetary predictions. However, the same model
predicts something that was not then observed: that forces between non-
spherically-symmetrical objects close to each other would NOT obey a
strict inverse-square law. A good model not only fits observations after
the fact, but predicts new observations under conditions different from
those of the original observations. Newton (I think, or else it was
someone later using Newton's model) predicted that the perihelion
position of Mercury's orbit could not remain constant in space, but must
precess because of the oblateness of the Sun (Einstein predicted from
relativity theory a very slight departure of this precession from the
Newtonian prediction, which was eventually verified). The precession of
the Earth's axis of rotation was likewise predictable from the influence
of the Moon and Sun together. These were necessary consequences of the
model, and fortunately it turned out that Nature agreed within the
limits of measurement.
To make a model, one must venture a guess about invisible mechanisms,
not simply generalize from observations. I believe that the correct name
for this process is the "hypothetico-deductive" method. As I interpret
these words, they mean hypothesizing an underlying mechanism, then
deducing from the properties of that mechanism some observable
phenomenon for comparison with real observations. This is what Newton
did. This is what PCT does. I do not believe it is what reinforcement
theorists did.
-----------------------------------------------------------------------
Best to all,
Bill P.