COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	sigma.abs[s85,jmc]	Abstract for Sigma Xi talk on non-monotonic reasoning
C00005 ENDMK
C⊗;
sigma.abs[s85,jmc]	Abstract for Sigma Xi talk on non-monotonic reasoning

Non-monotonic reasoning applied to artificial intelligence

	John McCarthy, Computer Science Department, Stanford

	Human-level artificial intelligence will require a database of
facts about the common sense world and computer programs that can
reason from the general facts and those relevant to a particular problem
how to solve the problem.  Mathematical logic is increasingly accepted
as providing the formalism in which these facts should be expressed, but
it has turned out that mathematical logical deduction has to be supplemented
as a reaoning method if a program is to draw appropriate conclusions.
The supplements required are conform to Ockham's razor --- the principle
that ``entities should not be multiplied beyond necessity''.  We will
describe a principle of ``non-monotonic reasoning'' that draws conclusions
true in the ``simplest'' models of the facts taken into account, i.e.
not necessarily true in all models.  ``Non-monotonic'' means that when
the premisses are enlarged some conclusions may go away, and this is
different from all systems of mathematical logic.  Understanding non-monotonic
reasoning formally also helps understand human reasoning and argumentation.

Notes for talk:

slides
	subset of last year's slides
remarks on the state of AI

AI is popular, but it is still an open question how much the present
level of technology can do

programs with common sense 1958