COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Stan, this isn't the editorial, but it's relevant.
C00009 ENDMK
C⊗;
Stan, this isn't the editorial, but it's relevant.

Ed, here's a draft.  I suppose it's too frank in some matters.
What's your reaction?  I think additional material
should be added, but I'm not sure what.
basic[e84,jmc]		Basic research in AI committee

	Artificial intelligence has attained great public recognition
recently, and many companies, new and old, are trying to apply
AI technology to a wide variety of problems.  However, almost all
of this expanded effort is aimed at applying the technology as is,
and few if any companies are making basic scientific efforts in
support of the technology they are trying to use.

	Of course, it is normal for companies and Government agencies
to put vastly more effort into technology than into basic research,
and it is unlikely that AI can fare much better than other sciences.
However, it is currently faring much worse, and we can hope to
correct this.

	It is important for the companies to correct this problem,
because what AI technology can accomplish is seriously limited
by the fundamental state of AI science.  Many desired applications
are simply infeasible today.  Moreover, without activity in basic
AI science it is difficult for a company to decide what projects
current technology will support, what new tools are available or
will shortly become available, and where fundamental advances are
required.

	One can say that AI technology is
treading close on the heels of AI science.

	The basic AI research situation in industry has always been
weak, but just as the Defense Department has begun to work towards
relying on AI technology, Congress has just wiped out DARPA's
allocation for basic research in AI.  As we understand it, the
rationale was that since AI has become a usable technology, it is
time for basic research in AI to be relegated to NSF.  NSF never
has supported basic research in AI very heavily, and it is entirely
unlikely that NSF will come up with the $21 million that has been
cut from the DARPA budget.

	In fact NSF's support of AI has also been heavily oriented
to specific applications parasitic on the state of research rather
than contributing to it.  This is because NSF began support of AI
in the early 1970s when short term applications were emphasized
in every science in connection with the ill-conceived slogan of
``Research Applied to National Needs, RANN''.

	In order to do something about this problem, the American
Association for Artificial Intelligence has formed a Committee
on Basic Research chaired by John McCarthy of Stanford University,
the immediate past President of the Asssociation.

	The decision to form the committee was taken before the
dimensions of the DARPA catastrophe became apparent.  At that
time the main problem was taken to be one of helping companies
active in applying AI technology to build programs in AI basic
research.  These programs might include both in-house research
and support of basic research at universities, but we expected
the emphasis would be on the former.  We propose to begin
by discussing the situation with executives of companies
that have efforts in AI technology.

	Now something will also have to be done about the DARPA
situation.  The present situation resembles what has happened
in the past but promises to be worse.  In the early 1970s,
promising lines of AI research were diverted into short term
``practical projects'' promising results in two or three years.
The researchers often continued to bootleg some basic research,
and today the results of the bootlegging are much more widely
known in academia, industry and government than the applied
projects, which are mostly long forgotten.  They have also
have made much more of a contribution to the AI applications
of today than the ``applied projects''.

	People interested in the work of the Basic Research
Committee are invited to contact the chairman.

John McCarthy
Computer Science Department
Stanford, CA 94305
(415) 497-4430
Arpanet: JMC@SU-AI