Where's Bill, World Models, Santa Fe

CSGnet Archive CSG1996
1

[From Rick Marken (960821.1315)]

                        Announcements:

                      -- Where's Bill? --

I noticed that Bill Powers hasn't posted in some time so I called him
today and found out that he has been cut off from his mail since last
Saturday (8/17) due to a failure of the phone lines at Ft. Lewis College.
Hopefully, a local provider will return Bill to CyberSpace by this Friday.

                -- A Control System Named Stella --

I've been playing with Stella II, a "systems simulation" package, on my
Mac. I finally managed to Build a simple control system that controls
a variable against a sinusoidal disturbance. The Stella software is
pretty easy to use (once you learn how to do it, of course) and it
provides nice graphics. I think I recall seeing that Remi Cote was
using Stella. If he, or anyone else with access to Stella, would like a
copy of my little control model just send me a note (marken@aero.org)
and I'll try sending the file.

···

--------------------
Me:

There is no implicit or explicit world model (a model of the environmental
physics that relate system output to controlled input) in PCT.

Martin Taylor (960821 14:10) --

Ah, so you are asserting that a control system will control its perceptual
variable equally well no matter what the form of the output function,
no matter what the gain, and no matter whether the output for positive
error is positive or negative.

No, I said that there is no implicit or explicit world model in PCT.
You can show me that I am wrong by pointing to the implicit or explicit
world model (a model of the environmental physics that relate system
output to controlled input: that is i = world model(o)) in PCT.

Me:

Do any of your modelling efforts speak to the idea that the number of
systems involved in control of a variable influences the ease with which
the variable can be changed?

Martin:

Perhaps not Kent's (about which I know very little), but there is a lot
of related work (from a source you seem to despise--the Santa Fe
Institute).

I don't despise the Santa Fe Institute. I just don't think the Institute
has done anything worthwhile (living systems-wise) other than have an
attractive Institute in Santa Fe, NM. If I were willing to go into the
trendy science business I would have an even nicer Institute in Carmel
(to be near the Bach Festival AND the most beautiful coastline in the
universe).

The Santa Fe Institute doesn't do anything worthwhile (living systems -
wise) because no one there has any idea what control is or that controlling
is what living systems DO. The Santa Fe Institute is just a sad waste of
glossy paper in a nice location;-)

Best

Rick

2

[From Bruce Gregory (960821.1625 EDT)]

Rick Marken (960821.1315)]

I noticed that Bill Powers hasn't posted in some time so I called him
today and found out that he has been cut off from his mail since last
Saturday (8/17) due to a failure of the phone lines at Ft. Lewis College.
Hopefully, a local provider will return Bill to CyberSpace by this Friday.

Thanks. I was starting to wonder if Bill was ill.

I don't despise the Santa Fe Institute. I just don't think the Institute
has done anything worthwhile (living systems-wise) other than have an
attractive Institute in Santa Fe, NM. If I were willing to go into the
trendy science business I would have an even nicer Institute in Carmel
(to be near the Bach Festival AND the most beautiful coastline in the
universe).

Sorry Rick, but you are wrong. The West coast of Ireland is the
most beautiful coastline in the universe AND Californians are
not in the same league with the Irish when it comes to being
neighbors!

Regards,

Bruce

3

[Martin Taylor 960821 16:45]

Rick Marken (960821.1315)

Oh goody. An easy one!

I said that there is no implicit or explicit world model in PCT.
You can show me that I am wrong by pointing to the implicit or explicit
world model (a model of the environmental physics that relate system
output to controlled input: that is i = world model(o)) in PCT.

If the environmental feedback function is such that the CEV (and the
perceptual signal) increases when the output increases, and the output
increases when the error (p-r) is positive, then the system will not
control. The system has to model the environmental feedback function
at least in respect of its sign. Bill P's Artifical Cerebellum shows
that there is at least one way of improving on this rudimentary one-bit
model, but that's not inherent in PCT.

What is inherent is that _every_ loop has to have at least this one-bit
model of some aspect of the environment, and that these one-bit models
are intrinsically independent, their correct setting being determined
only by the environment of the hierarchy. A hierarchy with N levels
each having M ECSs would have a model of the environment at least N*M bits
big. Realistic values of M and N imply quite refined models of the
environment, even when they are limited to just one bit per control loop.

Clearly the models are more precise than that. Loops in which the gain
is too high or too low for the environment don't control very well, and nor
do loops with output functions that react too fast for the environmental
loop delay or too slow for the disturbances characteristically experienced.
Even though the useful ranges of gain and time-constant are quite wide,
nevertheless they may be different if the environments changes sufficiently.
Those parameters represent more bits per control loop, though how many bits
has to be determined case by case. They all form part of the implicit model
of the environment.

I don't know how many bits are required to define one of Hans's models,
in an environment where the implicit PCT model has, say, K bits. It would
be part of the Occam's razor test: the theory needing the fewer bits in
its model would be preferred, all else being equal.

The Santa Fe Institute doesn't do anything worthwhile (living systems -
wise)

This was in response to my posting in which I suggested a Santa Fe answer
to a question you posed as worthwhile (living systems-wise). Odd that you
should be so sure that the answer cannot be worthwhile, before you look
at it.

Martin

4

<[Bill Leach (960821.1936 EDT)]

{I fainally got a Linux box to work on Worldnet but I do not yet have
any marcos to handle things like CSG headers|

Rick:

>There is no implicit or explicit world model (a model of the >environmental physics that relate system output to controlled input)
in PCT.

Martin Taylor (960821 14:10) --

>Ah, so you are asserting that a control system will control its >perceptual ...

Rick:
No, I said that there is no implicit or explicit world model in PCT.
You can show me that I am wrong by pointing to the implicit or > >
explicit world model (a model of the environmental physics that
relate system output to controlled input: that is i = world
model(o)) in PCT.

Rick, I think that this is a major terminology issue again. I believe
that Martin's presentation with regard to implicit "world model" is
quite consistent with Bill P.'s. I certainly agree with what I
understand Martin to be saying.

The "world model" is in the structure and tuning of the control loops.
At higher levels, symbol manipulation and inference may play a part in
the existance of a "world model".

In all cases (as Martin pointed out) what we might _think about_ our
world model(s) may or may not in any way be related to what actually
comprises that which IS our world model.

This modeling may differ substantially from what Han's considers
modeling (though I am not actually certain of that). Again, Han's
interests seem to reside at a much higher level than any of the rest of
us and with matters that probably not only are not of concern much
presently but probably will not be for a lifetime or two.

···

--
bill leach
b.leach@worldnet.att.net
ars KB7LX