From newshub.ccs.yorku.ca!torn!cs.utexas.edu!sun-barr!olivea!spool.mu.edu!caen!nic.umass.edu!titan.ucc.umass.edu!quilty Thu Oct  8 10:11:22 EDT 1992
Article 7128 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!torn!cs.utexas.edu!sun-barr!olivea!spool.mu.edu!caen!nic.umass.edu!titan.ucc.umass.edu!quilty
>From: quilty@titan.ucc.umass.edu (Humberto Humbertoldi)
Newsgroups: comp.ai.philosophy
Subject: Re: Simulated Brain
Message-ID: <BvoFn3.Ext@nic.umass.edu>
Date: 6 Oct 92 01:47:26 GMT
References: <1992Sep29.151801.8240@Informatik.TU-Muenchen.DE> <1992Sep29.225005.4267@usl.edu> <1992Sep30.173422.4220@utdallas.edu>
Sender: usenet@nic.umass.edu (USENET News System)
Organization: University of Massachusetts, Amherst
Lines: 68

In article <1992Sep30.173422.4220@utdallas.edu> ggraham@utdallas.edu (Gregory S. Graham) writes:
>In article <1992Sep29.225005.4267@usl.edu> mhf4421@usl.edu (Flynn Matthew H) writes:
>>erlebach@Informatik.TU-Muenchen.DE (Thomas Erlebach) states in
>>Message-ID: <1992Sep29.151801.8240@Informatik.TU-Muenchen.DE>
>>  
>>What on earth makes you think that any algorhythm througha single processor
>>is  capable of simulating a brain.   I find there's very little reason that 
>>the brain would rely on a single processor, and I think the evidence would 
>>show that there is  multiple processing going on all the time in our skulls.
>>The concept of a super pc just seems like a romantic notion somehow.
>>
>>M.H.Flynn
>>
>I think it can be shown that any multi-processor system can be simulated
>on a single processor.  I see a multi-processor solution as merely a method
>for making an intelligent system run faster, but I don't see it as essential
>to intelligence.
>-- 
>
>Greg Graham                                ggraham@utdallas.edu
>
Greg Graham is quite obviously correct here.  This is, in fact, just
the Turing thesis:  any computational system has a subset (possibly
improper) of the computational powers of a Turing machine.  Since the
PC on which I write this notice (as well as the computers on which
anyone reading this use) performs only Turing-computable functions --
and since it may model a universal Turing Machine -- this machine can
compute ANYTHING WHICH IS COMPUTABLE.  The mistake which Ms/Mr Flynn
makes is confusing a hardware level with an algorythmic level.  Any
hardware (with slight restrictions) can implement any algorythm, the
only difference is speed and efficiency.  
	Now, of course, if Ms/Mr Flynn is claiming that human brains
are not central processor cpu's with seperable memory/storage
subsystems and all those other features of conventional first-fourth
generation computers which we are familiar with, s/he is quite
correct.  The architecture of human brains is unquestionably much
different from any existing humanly-designed electromechanical
computational devices.  Then again, the question of single processor
vs. multi-processors is a bit of a misunderstanding of this
architecture, I take it -- since I assume that some variation of a
connectionist model of human brain architecture is correct.  That is,
there is a great deal more interactivity amongst the "nodes" of the
brain than would be allowed in any discreet multi-processing model.
On the other hand, single input/output channels are not isolatable in
the human hardware, as would be required if we were to characterize
the human nervous system as one single processor (even a huge one).
Clearly, something like parallelism exists across many computational
levels within the human (or generally "chordate") nervous system.  So
for example, in pattern recognition lower level feature recognitions
combine in a multivariate fashion within total object recognition; but
total objects are themselves similarly conditioned by all the "other"
sensory objects which contextualize and shape recognition.  Basically,
the single-processor/multi-processor distinction breaks down and
becomes meaningless within "Connectionist" systems such as chordate
nervous systems (I put "connectionist" in scare quotes to indicate
that I do not claim that chordate brains are fully describable, even 
at a hardware level, by any finitely and discretely striated
connectionist model.  Indeed, nature generally tends to be much too
"indiscrete" for such artificial cognitive models)

David Mertz


-- 

					ciao, humberto
--
      <***************************************************>


