AI stuff

[From Bill Powers (920904.1100)]

Penni Sibun (920902.1400) --

What the net "is" depends on each user's conception of it, and that
conception isn't out there in the world. It's in a head.

  well, i don't suppose solipsism is very useful.

That's too easy an answer. Control theory is not about solipsism. It
just recognizes that we don't all experience the same environment, so
whatever you say about the environment is probably not true for
everyone else. I don't doubt that there's an environment there. What I
do doubt is that our perceptual representations of it are isomorphic
to it.

···

-----------------------------------

well, i probably confused you by saying ``machine''--let's stick to
``automaton.'' at any rate, neither is a program: an automaton is a
description, a theoretical abstraction. one can perfectly rigorously
say whether an automaton is deterministic or not; i gave the def.
above. determinism does not describe what the automaton does, it
describes how it is built.

OK, I think that's what I said. The automaton itself could behave in
unpredicted ways while still being deterministic. I suppose your
definition would hold even if the automaton is deterministic, but one
of its computing elements computes a random action when it's
operating.

Automata that are theoretical abstractions aren't very interesting,
are they? I prefer mine to be "concrete-situated."
------------------------------------

the natural numbers are the positive integers starting from 1, so a
function from natural numbers to world states implies discreet rather
than continuous time.

OK. The problem with discrete time is that it rules out real physical
phenomena unless it's handled in a way I haven't seen in any AI
models. Physical phenomena take place continuously, and their
properties determine how they behave through time. If you apply a
force to an object at time 1, how far will it have moved by time 2? If
there's no link between discrete time-points and the underlying
continuum, there's no way to tell. How far it moves will depend on the
clock speed of your computer. You have to define a dt in real seconds,
so you can derive distance moved from the double integral of
force/mass TIMES DT. If you make one computing cycle equal to dt
seconds, like 0.01 sec, you'll get one final position. If you make the
computing cycle equal to a longer dt, like 0.1 sec, the object will
move farther in the same number of computing cycles.

This problem is present any time that the computer has to deal with
processes that actually are continuous. When it's ignored, as it
usually is, the results of a simulation don't really mean anything.
Not if you're trying to model a real system, that is.

I've written a persuasive letter to Chapman, by the way. It's worth
another try.
----------------------------------------------------------------------
Best,

Bill P.

(ps 920904.1200)
   [From Bill Powers (920904.1100)]

   Penni Sibun (920902.1400) --

   > well, i don't suppose solipsism is very useful.

   That's too easy an answer. Control theory is not about solipsism. It
   just recognizes that we don't all experience the same environment, so
   whatever you say about the environment is probably not true for
   everyone else. I don't doubt that there's an environment there. What I
   do doubt is that our perceptual representations of it are isomorphic
   to it.

i didn't say control theory was about solipsism. you said everyone
``out there'' is imaginary. i don't think that has explanatory power.
and i certainly don't say that perc. reps are isomorphic to the
env.--shoot me as a cognitivist if i do! and remember, i'm not trying
to talk about the enviroment. a message ago, we were agreeing that
instituions were processes, not objects. you don't hold that
processes are out there in the environment, do you?

   OK, I think that's what I said. The automaton itself could behave in
   unpredicted ways while still being deterministic. I suppose your
   definition would hold even if the automaton is deterministic, but one
   of its computing elements computes a random action when it's
   operating.

automata are just math. when you say ``4 + 7 = 11'' it doesn't matter
4 what or 7 what or what it looks like when you've got the 11
together. that's just not part of the mathematical description.

   Automata that are theoretical abstractions aren't very interesting,
   are they? I prefer mine to be "concrete-situated."

no. but yr original question had to do w/ terminology, and i tried to
explain that terminology. i do think that, as w/ any other field, to
really understand what the practitioners are talking about, you need
to have some idea of the foundations. theory of computation is one
such for ai.

   >the natural numbers are the positive integers starting from 1, so a
   >function from natural numbers to world states implies discreet rather
   >than continuous time.

   OK. The problem with discrete time is that it rules out real physical
   phenomena unless it's handled in a way I haven't seen in any AI
   models. Physical phenomena take place continuously, and their

right. ai models are like this, basically cause the run (or are
supposed to) on digital computers. i don't think you'll find anyone
that would argue that there's something missing by not being able to
model continuous time. the issue is more how close and/or useful your
approximation is.

   I've written a persuasive letter to Chapman, by the way. It's worth
   another try.

what did you try to persuade him of? just curious.

cheers.

        --penni