X-Message-Number: 4308
Date: Thu, 27 Apr 1995 20:05:54 +0200 (MET DST)
From: Eugen Leitl <>
Subject: Re: CryoNet #4280 - #4287

> ----------------------------------------------------------------------
> 
> Message #4287
> From:  (Thomas Donaldson)
> Subject: neural nets and computers
> Date: Mon, 24 Apr 1995 19:45:24 -0700 (PDT)
> 
> It's not obvious to me that neural nets are computers at all. They have no

Well, they are. They have been formally proved to be equivalent to
Turing engines. (Don't ask me for the lit reference).

> stored program and do not operate on a sequence of bits in accord with that
> program. (A simulation of a neural net in a computer, which is often done,

Well, they can map input vectors upon output vectors. Since that's what
we humans are doing all the time I consider this a pretty good description
of computation.

> does not make the neural net a computer any more than simulation of rainfall
> makes rainfall a computer). And a suitable neural net might also be trained

agreed.

> to emulate a computer (perhaps only a simple one) but that yet again does
> not show that neural nets are computers.

If it does simulate a Turing engine, it _is_ an all-purpose computer.
( a theoretical NN, as the Turing's is a theoretical engine).


> Moreover, now especially, designers have been using neural nets as attachments
> to computers to do those jobs for which neural nets work much better than 
> the standard computer. That too does not show that neural nets are computers,
> any more than your video screen is a computer.

NN have a much higher computation index, as defined by translatory movement
speed through their's state space times path entropy. Since that abstract
means of computation speed has a lot in common with "real" computation
power, no wonder people are using special software/hardware for a better
(virtual) engine.

> 

> If neural nets are to be considered computers, then I may have a misconception
> about what is the definition of computer. If so I would be happy to be 
> corrected, not by an ex cathedra statement but by reference to some book on
> the subject.

A computer processes information. That's the most fuzzy (but true) defintion
I know.
 
> 
> For those who are eager for an argument, I want to point out finally that 
> the question of whether or not neural nets are computers has at best a 
> tangential relationship with the issue of uploading. I'm not here retreating
> from my previous opinions on that issue; I'm just asking a question.

If:

1) we can scan relevant mind information
2) we can build the hardware capable to emulate the information
   as if it was the original tissue

then we have all the prerequisites of uploading. 
Since the brain hardware is very NN, we'll need appropriate (e.g.
topology) hardware to run it _efficiently_.


> 			Long long life,

The same to you,

-- eugene

> 				Thomas Donaldson
> 
> ----------------------------------------------------------------------
> 
> End of CryoNet Digest
> *********************
> 


Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=4308