COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	logist[f86,jmc]		Difficulties with logic approach to AI
C00006 ENDMK
C⊗;
logist[f86,jmc]		Difficulties with logic approach to AI

qualification problem
frame problem

temporal preference

context problem

	McDermott's ``Critique of Pure Reason'' mentions some real
and some imaginary difficulties in realizing the logic approach to
AI.  Besides the difficulties he mentioned there are some others.
We are optimistic about the logic approach, although we think it
might take a long  time to solve all the problems.  We shall discuss
these difficulties with emphasis on the real difficulties mentioned
in the paper.

McDermott: "Unfortunately, very few interesting inferences are deductions."

The question is not which inferences ARE necessarily deductions, but which
inferences CAN BE USEFULLY carried out by deductions. If we
learn that the sides of a certain triangle are equal and conclude that
its angles are equal too, is this a deduction? For someone who has never
studied deductive geometry, this is an inference based on visual images
and intuition, but we know that it can be also interpreted as a deduction,
and attempts to incorporate the knowledge of geometric properties of the
world into AI systems would have been severely handicapped if we did not
have deductive geometry. When we examine an ALGOL program and see that it
checks whether its input is a prime number, is this a deduction? This was
not a deduction before a formal semantics for ALGOL was invented, but now
it is, and this fact is crucial for designing systems that can operate
with programs in an intelligent way.

Every time when another aspect of the world becomes a part of deductive
mathematics, be it the numeric aspect, or geometric, or topological, or
mechanical, or algorithmic, or probabilistic, - this is always the result
of many years of foundational studies, often requiring major inventions,
both conceptual and technical, that could be made only after long periods
of painfully slow progress. The aspects of the world that commonsense
physics is concerned with are not exceptions.