Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!unixhub!slacvm!doctorj
From: DOCTORJ@SLACVM.SLAC.STANFORD.EDU (Jon J Thaler)
Newsgroups: comp.ai.philosophy
Subject: Re: Continuous vs discrete
Message-ID: <91093.195412DOCTORJ@SLACVM.SLAC.STANFORD.EDU>
Date: 4 Apr 91 03:54:12 GMT
References: <91082.223501DOCTORJ@SLACVM.SLAC.STANFORD.EDU>
Organization: Stanford Linear Accelerator Center
Lines: 25

(...I've lost the name of the person who posted this ...)

>> Marko, it all comes down to this. There are simulations of
>> neural nets which are close enough to the biological version to
>> reproduce _waveforms_. Example: predator evasion reflex of
>> tritonia. Example: swimming behavior of the Lamprey. Example:
>> stomach ganglia of the lobster. You want ref's?
>
>> Since these all do work well, doesn't that kind of indicate
>> that these networks are simulable?

I don't think so.  In fact the examples given illustrate the point that
there is a vast difference between simulating relatively simple systems
with hundreds or thousands of components and doing it with systems which
contain about 10**13.  I wonder what *qualitative* features of the behavior
are lost by the inevitable simplification that will be introduced.  I am
not talking about minor disagreements in the magnitude of an effect, but
about modes of behavior that will be missed altogether.

I think there is a good analogy in physics.  There is a theory of the nuclear
interactions (QCD) which may be correct.  The equations are not wildly
complicated, but they are nonlinear.  As a consequence of the nonlinearity, the
theory cannot (yet) be used to compute even the simplest of phenomena.
Given this, how can we think that computer simulations can even begin to
provide a realistic model of the human brain, or any other "intelligence".