So I was correct all along

********* FROM CHUCK TUCKER 930405-2 ********

The argument between the PCT modellers and the IT folks was about the
question: Can anything "outside" the living system be found "inside"
it? Answer: No.

Thank you very much,


[Martin Taylor 930406 12:00]
(Chuck Tucker 930405-2)

(Reconnected to the Internet after a BBN foulup)

The argument between the PCT modellers and the IT folks was about the
question: Can anything "outside" the living system be found "inside"
it? Answer: No.

Certainly this had nothing to do with the argument from my point of view.
It is an interesting question in itself, as part of PCT theory. But it
has absolutely nothing to do with the argument we have been having, which
has been about establishing a basis for discussing the physical properties
of control systems. Why, for example, is an integrator a reasonable form
for the output function of an ECS? My posting of yesterday, which just
arrived, shows that if my analysis is correct, the integrator is a good
but non-optimum form, if the disturbance is taken to be quite unpredictable.

I have been as frustrated with the non-logicality of the argument as I
gather Greg and Chuck have been. I keep thinking that we have mutual
understanding, and then a posting appears that wrenches the ground out
from under my feet, and I have to go back to even simpler conditions.
I'm trying to deal with the structure and function of control systems,
and I have to explain such things as that you can have almost perfect
information about one signal in another even though they have zero
correlation, but you can't have zero information if the correlation
is non-zero. These road-blocks prevent reasoned development. They
seem to represent strong resistance to disturbances of reference values
that ought to be dependent on technical factors, but apparently are not.

I apologise to the membership of CSG-L for continuing the discussion,
in my futile attempts to talk sense. Once before I said that I would
desist until I could put together a package, but got drawn into the
same discussion once more. This time, I will try to say why I think
the issue important, and I will have to desist, since apart from April
15-20, I am away until June 5.



The physical world has certain constraints, within which any control
system must work. A control system exists so that the internal state
of some chunk of the world can be stabilized in the face of variations
in the influences of the world on its parts. There are other ways to
achieve this stability--a rock is made of crystals that are connected
rigidly enough to withstand most influences from the world (for a while).
Softer things can't. They must control, which means that they must actively
oppose the effects of the external influences. To do this, they must have
an energy source that allows them to act strongly enough to oppose the
potentially damaging influences that they are likely to encounter, and
they must have a way of getting information that allows them to direct
the action appropriately to oppose these influences. The amount of
information must be commensurate with the uncertainties about the
influences. Information is a constraint, like energy.

Information about the influences of the world can be acquired very slowly,
over evolutionary time, if the kinds of influence do not change, and
can be built into the structure of the control systems that evolve. Other
information must be used at the time it is acquired, to oppose a present
danger. The issue we have been discussing concerns that kind of information.
Rick and Bill have been casting the question in terms of WHAT information
is acquired, whereas I keep trying to bring the discussion back to
HOW MUCH is needed and HOW MUCH is possible under different conditions.

We have, so far as I can see, no difference of opinion on how the control
system is structured or in the signal flows. No one has claimed that
knowledge of the amount of information available (a scalar quantity) is
sufficient to reconstruct a waveform (a vector quantity). The only
relevant claim is that if the vector quantity can be reconstructed, that
is prima facie evidence that information about it was available.

There have been differences of understanding about terminology. In
particular, the definition of a "disturbance" has been very fluid. That
causes trouble when statements are made like "there is no information
about the disturbance in the perceptual signal." I tried to bring some
consistency into that part of the discussion by relating "disturbance"
only to something accessible to the ECS. Sometimes, it gets expanded
to refer to the entire structure of the universe that interacts in any
way with the CEV defined by the perceptual input function of the ECS.
It's hard to make a concise, coherent argument under those conditions.

"Output" has been another fluid concept. I have again tried to keep
within the ECS. Once the output signal has been generated within the ECS,
that, to me, is the output of the ECS. But sometimes "output" has been
taken to mean the influence of the ECS's output signal on the CEV, through
whatever transformations are imposed by intervening control systems and
by the variable environment. Again, it is hard to develop a discussion
when important terms change their meanings, sometimes within single postings.
(Bill Powers 930405.1330 MDT)

Bill says that the discussion hinges on us wanting to hold onto old concepts:

Under the best of circumstances, people who believe in
X learn about control theory, and say "Wow, sure, PCT is the most
revolutionary thing since sliced bread, but you know that we Xers
have been saying the same things all along, and when you get
right down to the science of it, X is fundamental. We're glad to
see that PCT fully supports X."

How can anyone refute that? I don't believe it of myself. but it could
be true. What I think is that the last sentence should read "I'm glad
to see that X fully supports PCT and can be used to deepen our understanding
of PCT." And it is not true that I/we claim to "have been saying the same
things all along." Looking back over my old writings, I see a great deal
that could have led me to PCT, including a 1973 paper in which I invoked
a hierarchic control system to account for object perception by touch.
But I never did have the insight of PCT. Why not? I don't know. It
took CSG-L and reading Bill Powers' writings to bring me to understand
PCT and wherein it gains its strength. It is the nature of genius to see
what others could have seen and didn't.

Why I think Information Theory is important is that I think it can show
WHY people do what they do. Bill has analyzed a lot of what they do, but
he didn't come up with the notion of alerting systems and massively
parallel passive monitoring. He now accepts at least the first of these,
which might well be thought of as a structural modification to PCT. That
came directly from an Information Theory analysis. Does it matter? I
think so. What about predictive control? Is the disturbing signal
redundant within the constraints of the perceptual input function? If so,
some predictive control is possible (though it may not be the same as
the "objectionable" predictive control against which Bill has inveighed).
What about the form of the output function? As I mentioned above, we
seem to show that an integrator is not far wrong, though non-optimum,
if the disturbance is non-redundant. And so forth.

Is this important? It depends on whose reference signals it affects,
and through what perceptual filters (aka PIFs) the disturbances are
transformed. To me, planting trees is more important right now.

See you in June, if not April 15-20.