COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00004 00003
C00005 00004	The common ability to learn
C00006 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source;
.<<common[s83,jmc]		Expert systems sometimes need common sense>>

.cb EXPERT SYSTEMS AND COMMON SENSE

.cb Preliminary report

.cb by John McCarthy, Stanford University

Abstract: We partially identify a body of common sense knowledge and
methods of reasoning which is part of human capabilities.  Present
expert systems all lack most of these capabilities.  For many
purposes this doesn't prevent their usefulness, but for other
applications, common sense abilities will be required.  Only a
small part of this knowledge has been formalized, and we identify
a number of difficult problems to be solved.  These clearly aren't
all, and conceptual advances are required.  It may take a long time.

disjunction
dividing a situation into subsituations
narrative and the conclusions that can be drawn from them.

1. Expert systems, e.g. Mycin.
2. Mycin's capabilities and deficiencies.
3. Declarative and procedural knowledge
4. General common sense knowledge - the big list
5. What we can formalize. Situation calculus and reifying.
6. Common sense reasoning.
7. Introduction to circumscription.
8. Steps toward progress
9. References

common
See common[f82,jmc],common.abs[w83,jmc], slides
The common ability to learn

What must a learning system know before it can begin to learn?

Learning from lectures rather than from brain surgery.