Turn, turn, turn again

[Richard_Marken]

[From Rick Marken (941005.2200)]

Martin Taylor (941004 18:30) on conventional psychophysical research:

Yes, the perception the subject is (one hopes) controlling is that of the
experimenter being satisfied. The perceptual function the
experimenter is testing is not controlled, but contributes to the
controlled perception. One presumes that on other occasions, the
output of the tested function is a controlled perception--or at least a
more immediate constituent of a controlled perception.

This is an example (another was Clark's post on Sharif) of the lengths
to which even the most committed PCTers we will go to see the relevance
of pre-PCT research to PCT. The goal is to make it seem to oneself that
the people who were doing this research understood something about the nature
of control. From there, it is a short step to imagining that the results of
this research can actually tell us something about the behavior of control
systems. Notice, for example, how Martin refers to "the perception the
subject is (one hopes) controlling" as though psychophysical researchers were
thinking in terms of subjects controlling perceptual variables. Notice,
also, the phrase "one presumes that on other occasions, the output of
the tested function is a controlled perception" as though psychophysicists
are thinking in terms of control loops when they do their experiments.
In fact, psychophysical experiments were not designed with any
understanding of control in mind at all and, thus, no more can be learned
from them about the nature of purposeful behavior than can be learned
from any other pre-PCT research -- nothing. The only connection
between pre-PCT research and control theory is that the subjects in pre-
PCT experiments happen to be control systems.

The desire to believe that the results of pre-PCT research _must_ be
relevant to PCT is strong. I believed this myself for the first few years
that I was involved in PCT; I spent (wasted, it turns out) a lot of time
trying to answer my own and other people's questions about how PCT
explains "such and such". It took me a long time to realize that "such
and such" is almost always 1) a generalization based on statistical
results and, thus, irrelevant to understanding individual purposeful
behavior or 2) the result of an independent - dependent variable design
that does little more than hint at what variables the subject might be
controlling (as in psychophysical experiements).

I used to think that it was important to account for the results of
conventional psychological research using PCT. I finally realized that it
is best to just ignore pre-PCT research or, if anything, use it as a general
starting point for doing some real PCT research. I realized that the job
of PCT is not to "reject" or "reinterpret" existing behavioral science
research; the job of PCT is to start the whole science of living systems all
over again. The reason, of course, is that PCT is about controlling (purpose-
ful behavior); pre-PCT research is not. Explaining pre-PCT research with
with PCT is like explaining fairy tales with Newton's laws.

Many of the questions that seemed important in the context of pre-
PCT behavioral science become irrelevant with PCT. For example, the
psychophysical goal of finding the "actual" relationship between
physical and perceptual variables is really irrelevant (as a behavioral
question); the PCT question becomes "what perceptual variables are
being controlled in a particular situation". The problem of mapping
physical to perceptual variables still exists but it becomes part of the
problem of modelling how a system that controls certain perceptual
variables is able to do this in a world that is modelled by the laws of
physics. So we really don't need psychophysics anymore; and all the
huge effort that went into establishing "psychophysical functions"
based on an S-R model of living systems can now be politely
"depositied into the waste basket" (in Bill's felicitous phrase) --
my own thesis research included.

It is possible to base PCT research on research that has gone before; but
when you do this, it's best to think of the PCT research as a new start
rather than an effort to confirm or reject the results of the earlier
research. Avery Andrews' suggested approach to studying the control
of speech perception is an excellent example of what I think is the right
way to take off in a PCT direction based on hints from pre-PCT research;
Avery's study could be the start of a whole new science of the control of
speech. It may be that some of the results of this research are (at least
qualitatively) consistent with the results of conventional research on
speech -- and that would be great. But problems occur when people try to do
PCT research as an "extension" or "validation" of the results of pre-PCT
research. PCT research can't be an extension or validation of pre-PCT
research because pre-PCT research is not about control.

Again, I am on my knees, pleading with those who are in a position to
do PCT research, or to guide students in PCT research: please turn away
from the pre-PCT research. It has nothing to do with the control of
perception. PCT research will not get done if people don't get out there
and do it; and people won't get out there and do it if they keep trying to
make it answer the Procrustean questions of pre-PCT psychology. We
need more people doing and _publishing_ PCT research. Again, it has
been 21 years since the publication of B:CP and, still, there are only a
few people doing PCT research and only a handful of published reports
of this research. Why is this? It can't because the people on CSG-L aren't
smart enough; they seem pretty bright to me. And it's not because they don't
like PCT; everybody seems to like the theory. And it can't be becuase people
on CSG-L haven't had enough time to learn what the theory is about; I know
many people who have been involved with PCT for over 10 years. And it's
not because there are no researchers involved in PCT; many PCTers have
published lots of pre-PCT type research.

I submit that there is only one reason why there has been so little PCT
research and publication: it is becuase there is a reluctance to turn away
from pre-PCT research. This reluctance is based on the belief that there
must be something of value in the work of those "thousands" of pre-
PCT researchers. And there is, I suppose; by following their gaze, you
can see that they were all looking in precisely the wrong direction.

As Tom Bourbon said in a recent post, there have been numerous,
excellent suggestions for PCT research on CSG-L in the last two days.
We can suggest many more. But someone (plural) has to do them and
publish them (that's what Closed Loop is for) or the basis of PCT will be
nothing but blabbing.

Best

Rick

[_Martin_Taylor]

[Martin Taylor 941007 17:30]

Rick Marken (941005.2200) (just arrived)

Martin Taylor (941004 18:30) on conventional psychophysical research:

Yes, the perception the subject is (one hopes) controlling is that of the
experimenter being satisfied. The perceptual function the
experimenter is testing is not controlled, but contributes to the
controlled perception. One presumes that on other occasions, the
output of the tested function is a controlled perception--or at least a
more immediate constituent of a controlled perception.

This is an example (another was Clark's post on Sharif) of the lengths
to which even the most committed PCTers we will go to see the relevance
of pre-PCT research to PCT. The goal is to make it seem to oneself that
the people who were doing this research understood something about the nature
of control.

About the only appropriate comment on this is: "Stuff and nonsense."

Who said ANYTHING about the beliefs of the people doing the research? Where
does control enter into their own thinking? It enters OUR thinking, when
we consider what happens in that kind of experiment. And make no mistake
about it, what happens, happens. If what happens is done by living
control systems, it is a fit subject for consideration using PCT. If
it happens consistently, then PCT had better be able to provide at least
a plausible explanation.

From there, it is a short step to imagining that the results of
this research can actually tell us something about the behavior of control
systems.

Not the behaviour of control systems, but limitations on their behaviour,
perhaps. Or are you claiming that you can control the offset of two lines
to better than one arc-second, given that you deny the relevance of
psychophysical experiments that seem to show that you can resolve the
offset only to within perhaps 15 arc-seconds if you have good eyes? Can
you adjust the perceived loudness of a sound to within 0.1 dB? Can you
even SEE an object, whose perceived location you want to control, when the
intervening fog reduces the contrast of the object to 0.1%? Is conventional
psychophysics so useless in illuminating the limitations on control systems?

Notice, for example, how Martin refers to "the perception the
subject is (one hopes) controlling" as though psychophysical researchers were
thinking in terms of subjects controlling perceptual variables. Notice,
also, the phrase "one presumes that on other occasions, the output of
the tested function is a controlled perception" as though psychophysicists
are thinking in terms of control loops when they do their experiments.

Nonsense. How anyone could read these ideas into my words is beyond me.

In fact, psychophysical experiments were not designed with any
understanding of control in mind at all

True. Of course.

and, thus, no more can be learned
from them about the nature of purposeful behavior than can be learned
from any other pre-PCT research -- nothing.

"And thus"???

An inventory of the densities of common materials is done with no
understanding of navigation in mind at all, and thus no more can be
learned from it about which things float and can be made into boats
than from any other non-navigational research -- nothing.

Yeah, right

If they don't study purpose, why should you expect it? One doesn't learn
much about stellar interiors by studying linguistics.

The only connection
between pre-PCT research and control theory is that the subjects in pre-
PCT experiments happen to be control systems.

And THAT, friends, is a fact. It is a connection. It is what matters.
It is why psychophysical, and possibly other, non-PCT experiments have to
be dealt with if they produce consistent, reproducible results.

For example, the
psychophysical goal of finding the "actual" relationship between
physical and perceptual variables is really irrelevant (as a behavioral
question); the PCT question becomes "what perceptual variables are
being controlled in a particular situation".

What is this "psychophysical goal?" Psychophysics is far more than mapping
physical magnitudes onto presumed perceptual magnitudes, and people have
known for decades that there could be no "real, truthful" mapping. I knew
it as a summer student here, in 1959. I did experiments even then to show
that the so-called mapping depended on what was matched to what (and I used
what you might now call a control method to get the data). I should be
surprised if anyone outside the core group of disciples of Smitty Stevens
ever believed in the goal you say is shown by PCT to be irrelevant. It was
always irrelevant.

Again, I am on my knees, pleading with those who are in a position to
do PCT research, or to guide students in PCT research: please turn away
from the pre-PCT research. It has nothing to do with the control of
perception.

That is a religious position, forcefully presented by the loose canon of
the church. To be on one's knees is not a scientific position. There
are plenty of problems with non-PCT psychological studies--I try to expose
them every day around here. But it is not helpful to take the absolutist
position of the religious convert. A living control system is a living
control system. If one want to find out how it works, one has to notice
that it is a living control system. But once one has noticed this strange
fact, one can make sense of lots of other things one noticed about it
beforehand--things that still are so, but are no longer independent and
surprising.

And then one looks into the living control system further.

···

========================
More generally, Rick often pleads that the only legitimate aim of PCT
research is to find "the controlled variable." If one is to believe the
premises of PCT, this search is doomed to fail, because which variables
are controlled, and the "insistence" on each (to use a word introduced
to the discussion long ago as a generalization of "gain") changes from
moment to moment. What is a controlled perception one moment is not one
the next.

Furthermore, if the perceptual functions of the living system are distributed
and not mutually orthogonal, as must be the case in a robust system, then
there are no identifiable, discrete, "controlled perceptions," but rather
there are controlled "perceptual spaces" of indeterminate dimensionality.
Each ECU, which we presume to exist, has its own scalar controlled variable,
but no external experiment will ever find one. The best that can be done
is to model the experiment using possibly orthogonal perceptual functions
that cover the same basis space as the perceptual functions used by the
living control system. The models may fit to 99.999% accuracy, but all
such a fit would say is that the perceptual space has been closely identified,
not that the perceptual functions used in the model are those used by the
living control system.

Is that clear

Martin