COMMENT ⊗   VALID 00006 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	%nonmon[s88,jmc]		Nonmonotonic reasoning in human affairs
C00007 00003	\section{Nonmonotonic Reasoning in Artificial Intelligence}
C00008 00004	\section{Examples of Informal Nonmonotonic Reasoning}
C00009 00005	\smallskip\centerline{Copyright \copyright\ \number\year\ by John McCarthy}
C00010 00006	queries
C00011 ENDMK
C⊗;
%nonmon[s88,jmc]		Nonmonotonic reasoning in human affairs
\input memo.tex[let,jmc]
\title{NONMONOTONIC REASONING IN HUMAN AFFAIRS}

\section{Introduction}

	The object of this paper is to apply some of what has
been learned from the efforts to formalize nonmonotonic reasoning
in artificial intelligence to understand better the unformalized
nonmonotonic reasoning of ordinary language discussion of human
affairs.

	The reasoning in all the logical systems created by mathematical
and other formal logicians prior to 1977 was all monotonic.  If
a sentence $p$ of a logical language is formally deducible from
a set $A$ of assumptions, and $A$ is enlarged to a larger set $B$,
then $p$ is still formally deducible from $B$.  This is what we
mean by saying that a logical system is monotonic.  However,
ordinary human reasoning is often nonmonotonic.  If I hire you
to build me a bird cage and don't tell you more, you will put
a top on the cage in order that the bird can't fly away.  If
I subsequently tell you that my bird is a penguin, thus enlarging
your knowledge, you no longer draw that conclusion.  In the
next section nonmonotonic reasoning is described in more detail
and more examples are given.

	Like speaking prose, nonmonotonic reasoning is something
humans have always done.  However, mathematical logic never attempted
to formalize it, and many people seemed to believe that formalization
and nonmonotonic reasoning were inherently incompatible.  Indeed
this belief may be part of the basis of many people's belief that
artificial intelligence is impossible.

	In the 1970s artificial intelligence research began to
confront the problem of nonmonotonic reasoning.  Since it is
necessary for successful thought, intelligent machines must do
it too.  The first efforts were empirical and didn't involve
a theory of general nonmonotonic reasoning.  The AI language
Microplanner of 1971 (Hewitt 197xx, Sussman and Winograd 197xx)
included the THNOT primitive for nonmonotonic reasoning, and
Jon Doyle's (1978) ``truth maintenance system'' provided
for maintaining a propositional database, where some of
the justifications of propositions were nonmonotonic.

	However, the first attempt to make a nonmonotonic logical
system was included in (McCarthy 1977) and with (McCarthy 1980),
(McDermott and Doyle 1980) and (Reiter 1980), formalized nonmonotonic 
reasoning got properly started.  Since then the field has developed
rapidly, although none of the systems so far developed can be
regarded as definitive.

	Nevertheless, formalizing nonmonotonic reasoning for
AI purposes already seems to shed light on ordinary human
nonmonotonic reasoning.  Partly this is because philosophers,
psychologists, political scientists and literary people
don't seem to have squarely faced the nonmonotonic character
of the reasoning they discuss and often treat it as though
it were monotonic.

\section{Nonmonotonic Reasoning in Artificial Intelligence}
\section{Examples of Informal Nonmonotonic Reasoning}


\smallskip\centerline{Copyright \copyright\ \number\year\ by John McCarthy}
\smallskip\noindent{This draft of nonmon[s88,jmc]\ TEXed
 on \jmcdate\ at \theTime}
\vfill\eject\end
queries
	Where should the definition of nommonotonic reasoning
be put?