Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!news.umbc.edu!cs.umd.edu!zombie.ncsc.mil!news.mathworks.com!uunet!in2.uu.net!hearst.acc.Virginia.EDU!murdoch!viper.cs.Virginia.EDU!ccb8m
From: ccb8m@viper.cs.Virginia.EDU (Charles C. Bundy)
Subject: Re: Is Penrose Right?
X-Nntp-Posting-Host: viper-fo.cs.virginia.edu
Message-ID: <DGpIzG.I7L@murdoch.acc.Virginia.EDU>
Sender: usenet@murdoch.acc.Virginia.EDU
Organization: University of Virginia Computer Science Department
References: <44ucv0$22o@newsflash.concordia.ca> <DG8JH3.4pw@murdoch.acc.Virginia.EDU> <45gsnl$qm1@newsflash.concordia.ca>
Date: Thu, 19 Oct 1995 17:52:28 GMT
Lines: 74

>>>It depends on what you mean by "formal system".  Any computer program
>>>can be viewed as a formal system because it follows a set of
>>>predetermined rules.  Consequently, a "conscious computer" must be
>>>a formal system.
>>
>>Disagree strongly.  My definition of "conscious" involves internal self
>>modification based upon external stimuli.  Thus the "program" changes.
>
>It is easy to write programs that modify themselves.  (It is best to
>use assembler language, and self-modifying programs are not very
>common, but that's beside the point.)  A program can even modify
>itself in ways depending on its input.  Thus the "predetermined rules"
>may actually be predetermined "metarules" (how do I alter my processing
>to accommodate this new situation?) or even metametarules.  Systems

I quote Carver Mead:

"How can you design top down [NN but equally aplicable to AL]? Top down
 implies that you know everything and that isn't realistic"

If you don't "know" everything, a rule built upon environmental stimuli
you didn't anticipate can't be "pre determined", can it?

>that do this have been designed (and even built) by Albus, Powers,
>and others.  These systems exhibit rich behaviour but they are formal
>(or mechanical) in the sense that exactly duplicating the initial
>conditions and input history leads to identical behavior.

Read: "Intelligence without Representation" by Rodney Brooks.  If you
utilize environmental cues/"memory" you will most decidably NOT get
identical behavior out of the same FA.  I leave it to you and quantum
mechanics to decide if physical environmental input can be "identical"
from time frame A to time frame B.

By pre determined do you mean that:

   1) The architecture (Instruction Set) is fixed
   2) The generation of a rule based upon a pre determined instruction set
      is pre determined
   3) The firing of the rule is pre determined?

1) is pre determined (Does the "kernel" reference make sense now? :))
2) isn't pre determined, at least according to Turing.  Let's use the
   halting problem:

   Assume process A generates N rules and quits, then process B generates
   N+1 ... K rules and quits.  According to Turing the question of whether
   or not an arbitrary A will halt (Over all A's) is undecidable. Therefore
   it is undecidable that rule N+1 will be generated.

   So even "formal" automata theory doesn't lay claim to "pre determinism"
   when infinity is involved.

3) Obviously false, given that external stimuli can be truly random rather
   than pseudo random and repeatable.  Make a rule dependent upon external
   stimuli and it's firing is uncertain thus not pre determined.

>
>>Now if a body can "see" into the future you could still apply the term
>>pre-determined to "conscious".  I also believe that "conscious" individuals
>>have a kernel (much like an OS) which is fixed at any given time that
>>another aspect of their selves is undergoing internal self modification.
>>
>>This doesn't imply that the core self can't be changed it just implies that
>>it happens slowly, based upon some "delta" ratio.
>
>I have no objection to having a "kernel" but I don't see that it
>materially affects the argument.
>
>Peter

Charles C. Bundy IV
ccb8m@virginia.edu

