DPRG mailing list archives: [DPRG] Re: Allen Robots (Will's emotions)

Mon Mar 12 14:48:34 CDT 2007

On 3/12/07, Randy M. Dumse <rmd at newmicros.com> wrote:
> > What I question is your belief (as it seems to me) that they
> > are useful to model everything.
>
> My belief is not that they are useful tools for modeling
> everything, but that they are excellent for modeling many things
> real time, which turns out to be almost everything I find
> interesting.

> > I am not convinced that animal behavior can be
> > modelled well by a finite state machine.
> > Neither am I convinced that it cannot.

I'm sorry.  Rather than "modelled well", I should have said "usefully
modelled".  I imagine that if I actually said what I had meant to say,
you would have responded differently.

> If you were convinced RLL (Relay Ladder Logic) could explain
> production machine behavior? If I could show you RLL was a
> limited subset of FSM, would you then say FSM's could model
> production machine behavior?

Yes, I agree that, for this reason, FSMs can model production machine behavior.

> Are you convinced animal behavior can be modeled by BBR? If I
> could show you how BBR was a limited subset of FSM, would you
> then say FSM's could model animal behavior?

I'm assuming that BBR = "Behavior-Based Robotics".  I am neither
convinced nor unconvinced that BBR can properly model animal behavior;
I just haven't collected the evidence.  I will concede, however, that
a FSM can properly model a BBR system, as long as it doesn't include
an infinite stack or queue.  For these, you need at leasst a push-down
automaton (PDA).  Last time we discussed this, you asked for a problem
that can be solved with a PDA but not with a FSM.  The classic example
is a palindrome detector.  If you can make a finite state machine that
can detect palindromes of arbitrary length, you will not only convince
me, but also disprove at least one well-accepted theorem in automata
theory.

> If not, do you know of anything that can for certain model
> animal behavior?

No, I don't, other than animals themselves.

> > I assume that you are talking about quantum physics when you
> > say "physics at very deep levels",
>
> Yes.
>
> > but this seems to rebut your
> > point that physics is, fundamentally, a state machine.
>
> I find your argument including Bell's theorem a bit non-sequitar
> and stretched. But perhaps we try to chew too big a piece.
>
> Just to wave a hand at the problem, here is some of my reverence
> for physics.
>
> To me at the deepest levels physics is an interplay of energy
> and time.
>
> Time is what happens where energy (and mass) isn't. The fastest
> time rates are those far away from gravitating sources and all
> mass/energy/stress gravitate.

Ah.  There's the catch.  You're not talking about quantum physics;
you're talking about relativity.  No one has been able to unify the
two fields with a single theory yet (as far as I know, but there so
many people working on it, I may have missed the announcement).

> My first exposure to state in physics was the Bohr model of the
> hydrogen atom. Certain states are allowed, and certain not. When
> a photon is absorbed by a hydrogen atom, an electron changes
> state. The mass of the atom increases. Energy appears as mass,
> and in this state, time decides how long the photon can stay.

And now you talk about quantum physics.  I dislike the attribution
that time can make decisions, and I don't think that's what you
intended, either.  There certainly are discrete states, but how long
occurs between one state transition and the next is nondeterministic
except in the simplest of cases.  My point with Bell's theorem is
that, although there are discrete states, their transitions cannot be
modelled by a finite state machine (or any other formal automaton).

> So physics is an interplay between the two extremes, energy and
> time.
>
> What state machines are to me is the recognition of order. Time
> is what keeps everything from happening at once, and when energy
> and time are not in interplay, that is a period worth noting.
> Such a period is a state. In a state, no processing is
> necessary.
>
> Since not everything has hardware detection, during a state, the
> only processesing necessary to be done is that to determine if
> it is time to change the state. But no outputs are generated
> during the holding of state.
>
> > I would appreciate being shown to be wrong on either (or
> > both) of these topics, but until then I will likely remain
> > unconvinced of the finite state machine as the all-purpouse
> > real-world modelling tool.
>
> Rather than personally try to prove the finite state machine as
> an all-purpose modeling tool, in particular for computing, let
> me ask:
>
> Where do you stand on UTM (Universal Turing Machine)? Do you
> accept Church Turing thesis?
> http://en.wikipedia.org/wiki/Church-Turing_thesis

The Church-Turing thesis appears to be correct, but, as the article
you have indicated says, is unprovable.

> (Recall that the Church-Turing thesis hypothesizes this to be
> true: that anything that can be "computed" can be computed by
> some Turing machine.)

Where "computed" means derived in an algorithmic manner.  Just as
origami can solve a different subset of geometric problems than can a
compass and rule (trisecting an angle, for instance), I have no reason
to believe that the huge mess of simultaneous differential equations
going on inside our (or an animals') brains is computable, and thus
subject to the Church-Turing thesis.

> In particular recall the Turing machine is a finite state
> machine.

Actually, no, it isn't.  It uses both a finite state machine and an
infinite tape to do things that neither alone could do themselves.

An important question to ask ourselves is, if a Turing machine is as
powerful as anything else, why don't people write programs by
designing Turing machines?  I posit that the answer is, while it is a
correct model, it is not particularly suited to designing such
systems.  Similarly, the computer I am using right now has a finite
number of bits spread between its registers, cache, RAM, and hard
drive.  If you call that number N, its behavior can be described
completely by a finite state machine of 2^N states.  This model will
always produce the correct results.  Why, then, does nobody use it?

  -- Eric