COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	nsf.86[w86,jmc]		NSF renewal proposal
C00008 ENDMK
C⊗;
nsf.86[w86,jmc]		NSF renewal proposal

This is a proposal for renewal of the NSF grant DCR 82-06565-A1,2

Accomplishments

	During the first year of the proposed three year period McCarthy
continued his work on his circumscription formalism and its application
to formalizing common sense knowledge.  His paper ``Applications of
Circumscription to Formalizing Common Sense'' has been accepted by
{\it Artificial Intelligence} and will appear shortly.  In applying
the ideas to the blocks world, i.e. formalizing the effects of moving
and painting blocks, some difficulties were discovered by Vladimir
Lifschitz which give rise to unintended models of the circumscribed axioms
in addition to the intended models.  Lifschitz also found some partial
solutions.  A new issue arose as a consequence of this study.  Namely,
it may be that the circumscriptions that should be done depend not merely
on the physical situation but also on the reasoner's mental state
including even the history of the inferences that have already been made.
The consequences of this possibility were explored but definite conclusions
have not been reached.  Lifschitz is a colleague in the formal reasoning
work whose research is supported at present by DARPA.

	During the few years circumscription has been studied by many
researchers including Tomasz Imielinski (Rutgers), Jon Doyle (Carnegie-Mellon),
David Etherington and Robert Mercer (U. of British Columbia)
Raymond Reiter (U. of Toronto), Martin Davis (NYU),
Jack Minker and Don Perlis (U. of Maryland),
Michael Gelfond, Halina Przymusinska and Teodor Przymusinski (U. of Texas at El
Paso),
Steve Hanks and Drew McDermott (Yale),
John Schlipf (U. of Cincinatti), and
David Kueker (U. Of Maryland).  Some of these have compared it with other
nonmonotonic formalisms, some have applied it, some of tried to determine
its limitations and some have studied its mathematical properties as a
form of minimization in mathematical logic.

	McCarthy's own recent work on circumscription, partly incorporated in
the revisions to the above-mentioned paper, has mainly involved
prioritization, i.e. representing in a logical formalism the structure
of exceptions and exceptions to the exceptions required for common
sense knowledge.

	Some of the problems involved in this were described in McCarthy's
address to the 1985 IJCAI at which he received IJCAI's first award for
research excellence.  The address was entitled, ``The easiest unsolved
problems of artificial intelligence''.


Proposed Work

	During the period  xxx McCarthy will continue his work on
formalizing common sense knowledge, especially the work using
circumscription.  New variants of circumscription by McCarthy and
also by Lifschitz will be studied.

	The possible need for making circumscription depend on the
mental situation has led to the study of a ``mental situation calculus''
and this will be studied intensively in the next year.  The idea is
to formalize a mental situation that is changed by mental actions
and other mental events.  Besides its application to circumscription,
the mental situation calculus seems applicable to formalizing facts
about other people's knowledge, goals and beliefs and to the control
of reasoning.  It seems likely that control of reasoning can be made
more flexible by hill-climbing in mental situation space.  Only
very preliminary results have been obtained, and these will be presented
at the conference on knowledge to be held at Asilomar from March 19
to March 22.