X-Message-Number: 14940
References: <>
Date: Fri, 17 Nov 2000 10:35:04 +0000
From: "Joseph Kehoe" <>
Subject: Re: CryoNet #14932 - #14939

>----------------------------------------------------------------------
>
>Message #14933
>Date: Thu, 16 Nov 2000 07:17:04 -0500
>From: Thomas Donaldson <>
>Subject: once more on brains and computers
>
>HI everyone!
>
[snip]

>Anyone who wishes is welcome to produce a proof that such a creature
>(device) follows all the requirements of a Turing machine. I said
>originally that I did not consider it obvious that it does... not the
>same as saying that it doesn't. Why isn't it obvious? Basically because
>by its nature our brain does not in any sense write on a single long
>tape. The different neurons with different connections go off and
>write on many tapes, connected as if they were a treelike graph. Or
>if you insist on a tape, that tape will have to have the ability to
>split into 2 tapes, etc etc. Moreover, one simple feature of our brains
>is not matched at all by ANY Turing machine: they work in real time,
>while a Turing machine is a single computer with a very long tape
>on which it works. If we were only proving math theorems (or any
>other purely theoretical act) then this would make no obvious difference.
>However our literal survival may depend on whether or not we finish
>some computation in a small fixed time. Time is IMPORTANT, and any
>attempt to understand how we work must first understand that point. The
>creation of new neurons, on top of all of this, means that such a
>system does not have a clear limit on the number of tapes it's
>following. (Yes, in practice there is a limit, but it is FUZZY, not
>a definite figure).

Hi,
Just a point.

A single tape turing machine is provably equivalent to a parallel turing machine
with multiple tapes.  The parallel version is no more powerful no matter how 
many tapes there are! Even with fuzzy limits (i.e. any non infinite amount of 
input tapes -at infinity I am not sure what happens but that does not matter as 
brains never reach infinity)

Time may be important but I have seen no hard calculations indicating that chips
can never match parallel brain speed.
Remember neurons are considerably slower in firng than silicon gates.

You would have to show that the brain is NP bounded for it to be impractical to 
mimic in real time and even then...



Also work has been done a Neural net algorithms that allow growth of new 
connections and new neurons. Work is also progressing on NN software that used 
individual neurons with a computational ability much closer to actual neurons 
(i.e. not simple arithmetic functions with a sigmoid threshold function)


Joseph.

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=14940