Re: Maximizing results of efforts Re: Mainstreaming

From: Eugene.Leitl@lrz.uni-muenchen.de
Date: Wed May 02 2001 - 03:03:37 MDT


Ben Goertzel wrote:

> Well, Webmind Inc. had a group working on our AI project for 3 years, about
> 50% focused on building real AI and 50% on delivering AI to the company's
> products. The group grew from 4 to about 50 over the 3 years. Not quite
> Manhattan Project in size I'll admit, but getting there...

Do you realize how many top-notch people worked on the problem? How
many today's $$$s a few G$ in 1940s are?
 
> I'm not sure what your "very specific set of methods" is. Not having any
> practical experience trying to build real AI, how can you be so confident
> that you know what methods are required?

I'm referring to the only tried and true method. Sometimes, when I look
into the bathroom mirror, or read the newspaper, I have my doubts it really
worked.

> I believe I have a complete and workable design for a real Ai, but I don't
> pretend to know what methods are **required** to make real AI. I think that

Ha! I can top that. I don't have a complete and workable design for a
real AI, heavensforbid. It would cover most of the Sahara with documentation.
What I do have, is metaheuristics, a recipe. Two, actually.

You want to make a machine think? Look at prior art. Grab a thinking
critter off the street and stuff him into the machine.

Plan two: figure out, how said critters came into being, figure
out a decent point of departure, and do a fast-forward replay
of the above in machina.

Both plans differ from yours that they don't require superhuman
genius to execute. Sorry Eliezer, you're out of work. Sure,
they nontrivial, large-scale projects, but with a clear structure,
progress metric, and validated by existance proof.

Of course you'll burn many computronium mole-hours just for the
initial search in the parameter space. Then, you can start with
the *real* work. Which might require computronium km^3-centuries.
Then it might not.

> my design is one among many possible workable real AI designs. I think that
> finding **necessary and sufficient conditions** for a real AI design is a
> hard math problem, significantly harder than finding sufficient conditions,

Here's where we differ, again. I don't think it's a math problem at all.
At least not the sort of math you learn at school, or what monkeys do
and can do. I might have to retract that statement, though, depending
on what that mondo Wolfram dude comes out with next.

> which is all we need to do to build one. I can't help but suspect that your
> confidence is unfounded.

I sure as hell hope so. A real AI could clean ruin your day.



This archive was generated by hypermail 2b30 : Mon May 28 2001 - 10:00:01 MDT