Newsgroups: comp.ai.neural-nets,comp.ai.genetic,comp.ai.neural-nets,comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!newsstand.cit.cornell.edu!portc01.blue.aol.com!news-e2a.gnn.com!howland.erols.net!netcom.com!nagle
From: nagle@netcom.com (John Nagle)
Subject: Re: AI hype
Message-ID: <nagleE02KzL.Kq@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <32700CB9.446B@research.nj.nec.com> <3277C198.6723@rmit.edu.au>
Date: Wed, 30 Oct 1996 03:42:57 GMT
Lines: 62
Sender: nagle@netcom6.netcom.com
Xref: glinda.oz.cs.cmu.edu comp.ai.neural-nets:34272 comp.ai.genetic:10172 comp.ai.nat-lang:5550

Andrew Jennings <ajennings@rmit.edu.au> writes:
>Hype is I guess the taking of complex topics (such as intelligence) and
>simplifying them for a general audience. Why are we so critical of hype
>about AI and so forgiving of the hype surrounding other things (eg.
>Windows95, the network computer etc.). Remember the hype about "pen
>based computing"? 
     Because AI never delivered.  Go back and read Feigenbaum's
"The Fifth Generation", from the peak of the AI hype era in the
1980s.  He claimed that the US would become an agrarian nation,
losing out in technology, if the U.S. Government didn't support a 
big national AI lab.

     Worked for him, though; he's now Chief Scientist for the U.S.
Air Force.

>You can get
>caught by hype: worst case perhaps was the Fifth Generation Project
>where they guy generating the hype had the indecency to die before the
>project got really started. But many influential things came out of that
>project despite the hype.
     They did?  Name three.
     Personally, I suspect that the whole Fifth Generation effort in
Japan was an effort by MITI to divert attention from the much larger
effort to make Japan a power in mainframe computers, which was achieved
by Fujitsu and Hitachi buying Amdahl and National Semi's computer division.
The Justice Department might have acted to stop that on national security
grounds had not political attention been focused on the AI sandbox.

>To avoid hype is to avoid those with the real money. Unfortunately the
>whole area of intelligence is thoroughly discredited with the investment
>community.
     That's what happens when all the startups go bust.

>I think there is a deep cultural problem about intelligence that leads
>people to seize on these problems and amplify them beyond what's
>deserved.  More on this theme at
>http://www.cse.rmit.edu.au/~rdsajj/talks/aisig.ppt and
>http://www.cse.rmit.edu.au/~rdsajj/talks/ai_talk.txt. 
     Read that.  Try to find Drew McDermott's "Artificial intellgence
meets natural stupidity", which is almost twenty years old, but still
apt.

>Everybody who thinks about these areas knows that they are 20 year plus
>efforts, so to expect rapid progress is unrealistic. If you'd suggested
>in 1956 that a computer would be the world's leading chess player then
>they would have locked you up. Is this progress? Absolutely.
     People were suggesting that back then.  As soon as Samuels' checkers
program worked, many workers in the field thought chess was just
around the corner.

     The big problem in AI is that every time somebody has a good idea,
they start thinking human-level AI is just round the corner.  It's not.
Worse, you don't get much credit for doing "low-level" AI; Brooks had
a terrible time with the MIT faculty when he was doing insect-level
AI that worked.  Now he's doing human-level AI that doesn't work
(the "Cog" effort), but that's more respectable.

     We ought to be able to get lizard-level AI to work if we try hard.
We have the compute power.

					John Nagle
					nagle@animats.com
