Xref: utzoo comp.society.futures:377 comp.ai:1486
Path: utzoo!utgpu!water!watmath!clyde!rutgers!cmcl2!nrl-cmf!ames!sgi!daisy!klee
From: klee@daisy.UUCP (Ken Lee)
Newsgroups: comp.society.futures,comp.ai
Subject: The future of AI [was Re: Time Magazine -- Computers of the Future]
Message-ID: <962@daisy.UUCP>
Date: 27 Mar 88 23:28:45 GMT
References: <8803270154.AA08607@bu-cs.bu.edu>
Reply-To: klee@daisy.UUCP (Ken Lee)
Organization: Daisy Systems Corp., Mountain View, Ca.
Lines: 50

In article <8803270154.AA08607@bu-cs.bu.edu> bzs@BU-CS.BU.EDU (Barry Shein) writes:
>
>The cover article this week for Time Magazine is "Computers of the
>Future". Mainly what they mean by that are the two paths
>supercomputing and artifical intelligence (eg. neural networking.)
>
>I haven't finished it, it seems fun, some errors of course but you
>don't expect extreme technical accuracy from such an article anyhow.

What do people think of the PRACTICAL future of artificial intelligence?
For a while, it seemed like it was going to take off.  All sorts of
expert systems and tools houses started to appear.  Most of these are
bankrupt now.  Even the biggies like Symbolics, Teknowledge, and
IntelliCorp are having major trouble.  The only companies that are
successful are Star Wars contractors, and I'm not sure if that's what
you'd call a practical application.

Is AI just too expensive and too complicated for practical use?  I
spent 3 years in the field and I'm beginning to think the answer is
mostly yes.  In my opinion, all working AI programs are either toys or
could have been developed much more cheaply using conventional
techniques.

Why is AI expensive?  No matter how good our theoretical inference or
representation techniques get, we still have the practical problem of
extracting human knowledge transfering it to the machine.  The problem
I see is that every application is sufficiently unique to preclude
automated knowledge engineering.  Since you must solve it by hand
anyway, you could probably more effiently program it using conventional
techniques.

Does AI have any advantage over conventional programming?  There are
claims about the benefits of learning systems, meta-level algorithms,
abstraction, etc., but for these to be cost effective, they must be
developed in an application-independent fashion.  Is this possible?
I don't think so, at least not in the near future.

I'm probably being short-sighted and ignoring the (very) long-term
possibilities.  There are also side-effects, like the popularizing of
object-oriented programming and powerful programming environments,
that are very practical.  Maybe I'm just looking at AI too broadly and
not at specific application areas.

What do you thing?  Thanks for your thoughts.  No, I'm not an AI trying
to clone you.

Ken
-- 
What's the difference between a used car salesman and a computer salesman?
The used car salesman knows when he's lying.