COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	@style(indent 5)
C00006 ENDMK
C⊗;
@style(indent 5)
@flushleft(@B(The three lectures will cover the following two topics.))
@blankspace(3 lines)
@center(Requirements of Expert Systems for Common Sense Knowledge)

Expert systems are computer programs that embody the abilities of an expert in
some activity.  Most of them represent knowledge by a collection of production
rules.  A production rule has a pattern part which is matched against 
input data.  A successful match assigns values to the variables in the
pattern.  These values determine the action taken by the action part of the
pattern.

Much expert knowledge and ability can be expressed in this way, but other
intelligent activities require a body of "general common sense knowledge"
expressed in a way that is independent of specific purposes.  AI research
has identified some of this knowledge and expressed it in languages based on
mathematical logic.

We will discuss common sense facts and their representation in logic in
the following areas:

@begin(enumerate)
Partial knowledge of situations expressed in a language called "situation
calculus"

The effects of events and actions

Facts about the knowledge, beliefs, goals and intentions of oneself and others.

Concurrent events and actions

Partial knowledge of extended objects

Substances and quantities
@end(enumerate)

@blankspace(3 lines)

@center(Formalizing Non-Monotonic Reasoning)

Logical deduction is monotonic in that when the set of premises is increased
the set of deducible conclusions never decreases.  In common sense reasoning 
increased information often leads to the withdrawal of a previously inferred
conclusion. Intelligent computer systems require such non-monotonic reasoning and
"default rules" have been used for some time.  Recently it has been discovered
that non-monotonic reasoning can also be understood mathematically and 
formalized for computer use.  A form of non-monotonic reasoning called
circumscription will be described and applied to common sense AI problems.