Re: AI and Moore's Law redux

From: John Stick (johnstick@worldnet.att.net)
Date: Sat Jan 26 2002 - 08:46:06 MST


I suspect that once general intelligence is achieved, it will become
possible to run a human equivalent general intelligence on hardware
equivalent to a contemporary pc. There are several ways to approach the
problem, all very speculative, but which seem to me to point in the same
direction.

One approach often used is to compare to the human brain. Consider that
(1) much of the human brain is used for tasks other than general
intelligence, (2) the brain is not used to capacity (that is, an injury
that destroys a substantial amount of brain tissue does not necessarily
result in a proportionate loss of functionality as the remaining brain
tissue can in some circumstances gradually pick up some or all of the lost
abilities), (3) the brain may be extremely inefficient because there is a
likely evolutionary payoff to massive redundency in both memory storage and
processing, and (4) increases in speed and reliabilty in the performance of
simple operations can in part substitute for the absolute number of
conections. There is a lot of room for argument on much of this, but when I
put it together I suspect that a human equivalent general intelligence will
run on hardware at least several orders of magnitude smaller than the human
brain.

A second approach is to look at existing work in AI and ask if it looks like
a mature scientific field where further advances are likely to be only on
the margins and in the details. My impression is the reverse, that AI now
contains a number of widely disparate approaches, some of which don't really
address general intelligence at all. There is little agreement on
fundamental principles. I don't see anything much in existing AI work that
would enable one to rule out the speculations Emil quotes.



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT