From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Tue Mar 14 2000 - 21:04:32 MST
On Wed, 15 Mar 2000, Damien Broderick wrote:
> (Even that might be putting it a little strongly: the only formal AI expert
> on the list, Robin Hanson, has expressed misgivings over Eliezer's
> broad-brush and non-canonical algorithms/heuristics, and I haven't seen
> anyone cluey like Moravec or Vinge rush to adopt his schemata, although
> they both know about his site.)
That is interesting information. So the pressure will be on for Eli
in May. Even in disguise, I think we will recognize the memes
unless he is totally silent (and somehow I don't expect that to
be the case...) :-)
>
> Sure, it's cheery to mention each other in such glowing terms, but it might
> be preferable to see some code that does something in the world, or to make
> the world jump up and down waving dollar bills, before we assume that one
> of our number is the new Knuth.
Not so much "glowing terms" as respect for education in an area where
you know the waters are deep and I might sink before I swim. Now
at the Contact conference, both Minsky and Lenat gave presentations
and they seem to be of a similar mind. (I.e. you have to think
*really* hard about representations, heuristics and algorithims
to do AI and that most "current" AI research (ala robotic dogs and
facial expressions) was in no way going to advance the field.)
One thing that seems clear is that once the "robotics" aspects of
the brain get encoded into $6.00 chips (ala the recent vision-tracking
chip), the only thing left for the researchers will be to return
to the difficult parts of AI (knowledge representation, algorithms
that actually produce decent results, etc.).
Now, setting aside that debate (please), I will simply offer the
comment that both individuals showed examples of how damn complex
human "intelligence" really is. It isn't a single algorithm or
method that gets us over, its a grab-bag of alternatives that
we can use when one or more methods is utterly insufficient.
Lenat did offer an interesting example where Cyc could come up
with an answer to a question that stumped the collective experts at an
"unnamed" agency. [The questions involved the breakout of war in the Mideast
and the impact of blockades on the shipping/delivery of oil supplies.]
It seemed to me that the essence of the solution consisted
of two parts (a) having access to a very wide information base;
and (b) having a logical reasoning process that exceeds typical
human abilities. [E.g. the reasoning to solve Fermant's Last
Theorem that involved both a very wide information base and the
ability to connect very diverse areas of thought.]
Now what Lenat did not show was what fraction of the questions
whare Cyc did not come up to the level of human conclusions. Still,
one has to bear in mind that computers are still slow and Lenat
has not had the "unlimited" budget that nature has had.
Robert
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:27:24 MST