COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	common[w86,jmc]		Abstract on "Expert Systems and Common Sense"
C00006 ENDMK
C⊗;
common[w86,jmc]		Abstract on "Expert Systems and Common Sense"

Expert Systems and Common Sense

John McCarthy
Stanford University

	Today artificial intelligence is being extensively applied.
These applications are almost entirely based on the technology of
rule based expert systems.  Such expert systems consist of collections
of facts and rules.  Besides the facts built into the system, new
facts arise from dialog with the user and from operation of the rules.
Goals are often included among the facts.
The rules detect patterns in the facts, assign values to variables
accordingly and then perform actions that depend on the values of
the variables.  Sometimes the actions are external, e.g. consist of
advice to the user, and sometimes they are internal and result in
putting new facts in the database.  Few systems generate new rules,
and most limit facts to ground sentences in a domain dependent first
order language.  Although Mycin is one of the earliest expert systems
(1974), it is typical of how they work.  The rules have as input
patterns of symptoms and results of tests and output diagnoses and
recommendations for treatment.

	This technology has important limitations.  It doesn't attempt
to determine the consequences of actions it considers in order to
find actions whose consequences are judged favorable, i.e. it doesn't
plan.  It doesn't know general facts about its domain, e.g. Mycin
which diagnoses bacterial diseases doesn't have represented in its
database in any way the fact that bacteria are organisms that grow
and reproduce and sometimes emit toxins and are transmitted from
one person to another in certain ways.  It can do its job without
this knowledge, but the lack of it imposes limitations.  It also
hasn't ordinary common sense knowledge of life and death.  If told
the patient is dead, it merely replies "unrecognized response".

	The lecture will explore the strengths and limitations of
expert systems without planning and common sense.  This can provide
some guide to where the present level of expert system technology is
and is not applicable.