Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!lll-winken!uunet!portal!cup.portal.com!dan-hankins
From: dan-hankins@cup.portal.com (Daniel B Hankins)
Newsgroups: comp.ai
Subject: Re: Simulation vs Reality
Message-ID: <17472@cup.portal.com>
Date: 22 Apr 89 06:56:09 GMT
References: <11314@bcsaic.UUCP>
Organization: The Portal System (TM)
Lines: 49

In article <11314@bcsaic.UUCP> ray@bcsaic.UUCP (Ray Allis) writes:

>Nope.  You assumed the thing you're arguing for.  You said "If there were
>a way to eat the simulated meal, I could eat the simulated meal."  Even
>with that, you didn't succeed in eating the simulation, you cleverly
>created *real* light rays, scent molecules and hand molecules from the
>simulation and ate *them*.

     You're getting closer.  In any case, to someone who experiences it, it
is a difference that makes no difference.  To him, the simulated meal is a
real meal.

     Now, to consider 'simulated' machine intelligence (actually,
intelligence _implemented_ on a machine).  I do not succeed in
communicating with the simulation, I cleverly create *real* textual output,
voice output and arm movements from the simulation and communicate with
*them*.

     If one can't tell the difference between a communication from a
simulated sentient being in a box (the box is important) and a 'real'
sentient being in a box, then what important difference is there?


>Once the simulation is constructed, you can exhaustively work out all the
>internal interactions, implications and entailments of the system of
>assertions and interrelationships, but that's all you get!  There is
>nothing coming out that you didn't put in.  

     Beg to differ.  If the system interacts with its environment what
comes out is what you + environment 'put in'.

     Nothing comes out of any _closed_ system that someone didn't put in.
I'm certainly not proposing any AI system built from scratch and springing,
as it were, from my forehead fully grown and with full knowledge.  You
can't do that unless you can exhaustively read the state of an existing
brain at some instant and duplicate it in the program.  That looks to be
impractical for just about forever.  The protons will probably begin
decaying before we have that capability.
     Rather, one gives the system self-organizing capabilities, programs
_no_ knowledge into it, except perhaps basics like pain, pleasure and so on
(like an infant), and puts it into a learning environment.  Then lots of
things come out that I didn't put in, except as capabilities.
     The interesting thing about self-organizing systems is that one of the
capabilities you give them is to gain more capabilities.


Dan Hankins

Why did the chicken cross the Moebius strip?