COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	nsfr.82[w82,jmc]		Final report to NSF on Basic Research Grant
C00005 ENDMK
C⊗;
nsfr.82[w82,jmc]		Final report to NSF on Basic Research Grant

	Making a computer program exhibit intelligence requires equipping
it with common sense knowledge about the world and the ability to do
common sense reasoning.  Both problems were recognized by 1960 and
have proved to be difficult.

	The present research made progress in both domains.  Using two
puzzles as an example, we studied the representation of knowledge about
knowledge.  In particular, we studied how to represent the fact that
a person knows only certain facts and therefore cannot be expected to
know others.  This is important for computer programs that have to
inform people or other programs or conversely have to keep secrets
from other programs.  Knowing what people know and don't know
is also important for predicting how they will act.
Methods for describing limitations on people's knowledge using Saul
Kripke's possible worlds formalism was developed as was another
method involving second order logic.

	Common sense reasoning involves additional methods beyond those
formalized in mathematical logic.  These methods are non-rigorous
in that they "jump to conclusions" that are sometimes wrong, but 
nevertheless at least some of them can be described just as formally
as the methods of mathematical logic, and this is important
when computer programs are to use them.  Since 1978 various proposals
for "non-monotonic logic" have been studied.  The word "non-monotonic"
refers to the fact that conclusions are reached that might not be
reached with more facts.  Our research developed a method called
"circumscription" that allows a computer to conjecture that the objects
it knows about of a certain kind are all the relevant objects of this
kind.  It turns out that people often make such conjectures, and they
are often useful.