Re: Right on schedule?

Date view Thread view Subject view Author view

From: Eugene Leitl (
Date: Sun Jul 02 2000 - 19:05:14 PDT

Adam L. Beberg writes:

> The $1000 point is nifty and all, but when you talk about computer
> intelligence, one machine is pointless, you have to talk about all the
> networked ones together. The entire internet. Not really true yet, but
> it will be very soon *grin*.
This assumes that all these machines are cooperating. You could do
that with a Net worm, but your window of operation would be limited
(watch for peasants with pitchforks), and subsequent infections would
be harder and harder, as exploitable pieces of code are being found
and removed.

If you can use that window to kindle true intelligence, the more power
to you. However, that would be a lot like suicide, so I wonder why
anybody that smart would do something that stupid.
> The human nervous system may do 10^16 ops per second, but almost all of
> that is in coprocessors, the main CPU does like 10 high level ops per

These are remarkably biased statments. The brain is a distributed
architecuture. Talking about coprocessors in this context is extremely
misleading. I thought you were the one locked in constant battle with
centralist thinking?

> second. We just switch descision making "programs" real quick. We really
> are slow and stupid little creatures in the big picture.

Well, we still are very far away from the aggregated computation
capability of those slow and stupi little creatures in a single
installation. Distributed systems don't count, because the pipes are
not fat and fast enough. Unless someone strings a lot of new fiber,
and adds kick-ass switches, that is.

> We passed the "fast food worker brain" point a while ago. Computers are
> actually more at the point of a gorilla right now, if the researches

Strange, I'd assume we're at best at insect stage -- if we had the
right software, and the right massively parallel architecture (memory
bottleneck will get you in single-grain installations). A very large
DSP cluster could be as smart as a cockroach, potentially. If it's
limited to text string processing, it would appear really smart. As
soon as you have lots of resource sinks like a robotic proxy with
machine vision etc., these MIPS suddenly start evaporating real fast.

I think true progress in AI will only come from ALife and wet neuro
people (digitized neuroanatomy route). Mainsteam Comp. Sci. has been
remarkably sterile so far.

Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Sun Jul 02 2000 - 20:10:43 PDT