From newshub.ccs.yorku.ca!ists!torn.onet.on.ca!utgpu!csd.unb.ca!morgan.ucs.mun.ca!nstn.ns.ca!news.cs.indiana.edu!att!linac!pacific.mps.ohio-state.edu!zaphod.mps.ohio-state.edu!mips!darwin.sura.net!cs.ucf.edu!news Mon Jun 15 16:04:46 EDT 1992
Article 6217 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!ists!torn.onet.on.ca!utgpu!csd.unb.ca!morgan.ucs.mun.ca!nstn.ns.ca!news.cs.indiana.edu!att!linac!pacific.mps.ohio-state.edu!zaphod.mps.ohio-state.edu!mips!darwin.sura.net!cs.ucf.edu!news
>From: clarke@acme.ucf.edu (Thomas Clarke)
Subject: Re: Physical versus Computaional (was Re: Transducers)
Message-ID: <1992Jun11.201743.11470@cs.ucf.edu>
Sender: news@cs.ucf.edu (News system)
Organization: University of Central Florida
References: <1992Jun11.184125.13172@mp.cs.niu.edu>
Date: Thu, 11 Jun 1992 20:17:43 GMT

In article <1992Jun11.184125.13172@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil  
Rickert) writes:
> In article <1992Jun11.172846.9370@cs.ucf.edu> clarke@acme.ucf.edu (Thomas  
Clarke) writes:
> >In article <1992Jun11.153432.4670@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil  
> >Rickert) writes:
> >> In article <1992Jun11.132823.7139@cs.ucf.edu> clarke@acme.ucf.edu (Thomas  
> >Clarke) writes:
> >>   (in reply to Neil Rickert <4138.708217481@mp.cs.niu.edu>)  
> >> 
> >> >You are using "physical" in a much different way than I would.
> 
> >As I suspected, "What we have here is a failure to communicate."
> >According to my trusty on-line Webster, I am using physical:
> >	2b (2): (2): characterized or produced by the forces and operations 
> >               of physics
> 
>   Strange.  That is exactly how I am using the term "physical".
> 
> >I think you are using physical:
> >	3a: of or relating to the body
> 
>   I am most certainly NOT making that mistake.

Not so sure it's a mistake.  I vaguely remember some theory of mind in 
which the body played a distinct role.  Since people actually tense
the relevant muscles when thinking and visualizing, someone (who?) 
maintained that somatosensory feedback was essential to thought.
 
>   As I said in my message, a computer implementation uses electrons, and
> you can consider that physical.  In that trivial sense everything
> computational is physical, but surely that misses the point.
> 
>   Let's look at your multiplication with neural nets.  Consider
> the following 5 ways of doing it:
> 
> 	1:  Implement the NN on a standard digital computer using
> 	    transistors and dynamic RAM. 
> 	2:  Implement on a 1950's era computer using vacuum tubes,
> 	    mechanical relays and mercury delay lines. 
> 	3:  Implement on a digital computer built using fluidics. 
> 	4:  Implement on a digital computer using photonics. 
> 	5:  Implement on a huge room full of trained clerks using
> 	    mechanical calculators, paper, and a typewriter.
> 
>   To avoid quibbling, lets agree that all acquire the multiplicands
> from an IBM punch card, and all produce the output in standard
> typewriter format on paper.  Let's also agree to slow down the
> faster implementations so that they are all at the same speed.  Let's
> also agree that all systems uses the exact same degree of precision in
> their floating point data, and the identical random number sequence if
> you are using pseudo-random numbers.
> 
>   Then I claim that all implementations will produce the same result
> result.  If so, the result is a function of the computational
> algorithm, and not of the physical mechanism used to implement that
> algorithm.  It is computational, and not physical.

Of course they will, they're all digital.  All produce the same
answer by design.  Only number 5 might produce novel results if a 
clerk got bored :-)

As an alternative I present an inherently analog, but silicon, 
neural network implemented using field effect transistors in CD4007 
chips as variable weight resistors, and LM324 op amps and 1N914 
diodes for the sigmoidal function.  You can simulate the 
CD4007/LM324/1N914 neural network on a digital computer with SPICE,
but you must make approximations in the simulation.  Whether these
approximations are important depends on the details of the network.
I wouldn't begin to guess about the significance for the 10^10 
neuron network in our heads.  (I would guess but I can't prove it :)
 
> >If I may elaborate a bit, not all neural nets run on digital computers;
> >brains use something like spiking rates as analog variables, and
> >Carver Meade's chips use voltages as variables.
> 
>   I would suggest that if you really believe that a physical implementation
> is important, you should stop fooling around with these silicon neurons,
> and start building some chemical neurons using the same neurotransmitters
> as the brain.  For if what you are looking for is physical, and not
> computational, then the specific physical details of the implementation
> are crucial.

Some day I hope to stop fooling with neurons simulated in a digital 
computer, but right now I really don't know what analogish aspects 
would be important.  Off hand, I don't think it matters though whether 
analog neurons are implemented in silicon, in brain tissue, or whatever,  
provided the non-linear dynamics are preserved.  (If quantum effects are
important then physical scale might be a factor ).
  
I could even be wrong, and it might be the case that it doesn't matter
whether neurons are implemented digitally, but I maintain it is an
open question. 

--
Thomas Clarke
Institute for Simulation and Training, University of Central FL
12424 Research Parkway, Suite 300, Orlando, FL 32826
(407)658-5030, FAX: (407)658-5059, clarke@acme.ucf.edu


