X-Message-Number: 8594
From: Thomas Donaldson <>
Subject: Re: CryoNet #8546 - #8551
Date: Sun, 14 Sep 1997 16:18:27 -0700 (PDT)

Hi again!

I note that John Clark is aware of neural nets.

While it's probably possible to make a Turing machine which would implement
the workings of billions of neurons together, it seems unlikely that it could
do so in any time we (or even we if we attain the immortality we are seeking)
would consider reasonable. Recognition is not a task performable by a
PRACTICAL Turing machine.

Our nervous system (for some of this you must be a reader of PERIASTRON,
sorry!) also has a feature which MIGHT be impossible for a Turing machine
to emulate. The best current ideas about how memory forms (in brains, guys,
in brains) involve growth of new connections between neurons. This comes
from actual observations. It may also involve growth of new neurons, including
in adults (readers of PERIASTRON will know why I say this). 

Question (I'm not fooling, but then I'd want to give this more thought
before I answered, so I really mean it as a question): the classical Turing
machine is given a tape, which it moves along marking individual digital
spaces. If we allow formation of new connections and even new neurons, this
tape becomes endless not only at both ends but even in the middle. (A
system which does this might also signal the addition of new tape with
some code at its front and back, just so the Turing machine doesn't
get confused). This system allows addition of new tape upon new tape
upon new tape, it's not a single-time thing. Among other effects would
be that what was once a simple code written by the Turing machine would
become quite complex over time, with the overlays).

I am certainly aware of all the ideas for mapping, say, a digital N-dim
infinite set of points to a single 1-dim sequence of points. Perhaps
that solves the problem of equivalence and perhaps not: remember that
such a mapping would have to change constantly as the Turing machine
proceeded.

For PRACTICAL purposes the case is good that neural nets, especially
ones sprouting new connections and new neurons, can't be imitated by
a Turing machine. Perhaps Mr. Pietrzak will save me the trouble of
looking up the proofs in books by giving an opinion here.


                         Long long life,

				Thomas Donaldson

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=8594