COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	issues[w86,jmc]		Issues in the foundations of artificial intelligence
C00007 ENDMK
C⊗;
issues[w86,jmc]		Issues in the foundations of artificial intelligence

	This is somewhat in reaction to the preprints of the
Workshop in the Foundations of Artificial Intelligence to be held
at New Mexico State University on 1986 Feb 6-8.  I'm afraid that
the issues listed in the agenda of the meeting miss so much of
what is important to AI that it may be worthwhile to organize
another meeting with a quite different agenda.  The problem with
the New Mexico meeting is the time to be wasted on whether AI
is worthwhile at all, its relation to cognitive science and other
fields (almost certain to be mere haggling over definitions), etc.

	Of course, one can ask whether a meeting on Foundations of
AI is worthwhile at all.  My opinion is that there are enough
foundational issues on which people have different half-formulated
opinions to make it worthwhile to try to lay them out and discuss
them.

	Here are some of the issues I see.

	1. The old issue of declarative vs. procedural is still with
us.  The argument between the most extreme positions, if there ever
was one, is played out, but there's plenty left.  What information
is best represented declaratively and what is best imbedded in
procedures?  Don't forget the possibility that it may be desirable
to represent the same information both ways.  Don't forget that
certain information may have to be represented only procedurally
at present only because of our weak understanding of AI.

	2. What kinds of information can be learned or inferred in the
present state of AI and what kinds must be supplied by the builder
of the AI system?

	3. When does an experimental AI system provide useful information
to advance the field?

	4. What information is representable in logic programs and
what has to be represented some other way?  My ``sterile container''
example purports to show that this depends on the use to which the
information is to be put.

	5. How is deduction to be supplemented by non-monotonic
reasoning in systems that use logical inference for computation?

	6. In systems that use logical inference for computation
what forms of control of reasoning are required and necessary?
How may this be represented?

	7. What is the relation between natural language and logic?
Is there such a thing as natural language reasoning?  My opinion
is that there is not, because human natural language arguments
depend on the listener using facts not represented linguistically
in his brain.

	8. Does mathematical logic provide suitable formalisms
for representing facts about the common sense world?

	9. What is the relation between theory and experiment in AI?

	10. Which philosophical issues are relevant to AI and which
can be bypassed?