COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	common[s84,jmc]		What is common sense? for AAAI Presidential Address
C00007 00003		I want to try to distinguish between common sense knowledge
C00010 00004	Draft of the actual lecture
C00013 ENDMK
C⊗;
common[s84,jmc]		What is common sense? for AAAI Presidential Address

common[f83,jmc]		Common sense reasoning is not natural language reasoning
common[w84,jmc]		Definition of common sense for Kay Mills
common.abs[e83,jmc]	Abstract of common sense paper
common[e83,jmc]		Common sense paper for NY Academy
common[s83,jmc]		Expert systems sometimes need common sense
common.abs[w83,jmc]	What programs need common sense
common[f82,jmc]		What is common sense?
common[s82,jmc]		What is common sense?
COMMON[W78,jMC] 03-May-78	TOWARDS A SCIENCE OF COMMON SENSE
COMMON[CUR,JMC] 02-Jan-75	Common sense and computers
COMMON.MEN[ESS,JMC] 11-Jan-76	TOWARDS A SCIENCE OF COMMON SENSE
PROLEG[S79,JMC]		PROLEGOMENA TO A SCIENTIFIC THEORY OF COMMON SENSE
IDEAS[S80,JMC]		Language of common sense
COMMON.[E80,JMC]	Programs with Common Sense
slac[f82,jmc]		AI lecture for SLAC; common sense physics
handou.226[w83,jmc]	Handout for cs226 - Common sense data base, 1983 winter
lectur[f83,jmc]		Formalization of common sense lectures

An important domain of common sense knowledge is the relation of
appearance and reality.  This is of equal importance to the facts
about the effects of events.  The direct laws concern how reality
affects appearance.  Inferring reality from appearance is a process
of inversion.

The observation-action rules of the kind embodied in expert systems
are also a form of common sense.  They may even be the basic form in
which the more declarative forms are somehow encoded.

Domains of common sense knowledge.

1. The effects of events including actions and concurrent events.

2. The relations between appearance and reality.

3. Facts about planning the achievement of goals.  a must or can be
done before b.

4. The principle of rationality.

Perhaps it is worthwhile distinguishing facts from knowledge.  It
is worthwhile thinking about the set of facts qua set even though
what is known is only a subset of the facts.

Contents:

1. Main contents are a list of areas of common sense knowledge.

2. Realist philosophy, roofs and boxes, evolution, philosophical
relativity - Is philosophy important to AI?  For me it has had
great heuristic value.

3. Do we need a discussion of what are facts and what is built into
program and the redundancy thereof?  Yes, we need it.

dog and trash cans as paradigm problem.  Leave the listeners and
readers with problems to think about

reification

fasten the cans to something

Challenge problems:  need one with a clearly small domain of knowledge.
	I want to try to distinguish between common sense knowledge
and common sense ability.

  In logic this is the distinction between
axioms and rules of inference.

In Lisp or Prolog there is a similar distinction between what is
in the program being interpreted and what is built into the
interpreter.  There are also analogs to the interpreter's data
working structures, which are not accessible to the
program.  Exceptions to this are Brian Smith's 3-Lisp and an
unpublished "introspective Prolog interpreter" written by
Peter Szeredi.

  In general some information is
represented only as facts, some is represented only by built-in rules
and still other information is represented both ways.

	In humans also, there is knowledge that is explicitly
expressible in the form of natural language sentences and information
that is built in.  Some of this is hereditary and expressed in
the course of development, and another part is learned but not
ordinarily expressed linguistically.

June 5

	I should refer to the principle of rationality, because this
is an important common sense fact and because it recalls Newell's
1980 lecture.

Don't forget to refer to SRI's common sense summer if this turns
out to be appropriate.

wishes, goals, intentions, state of being pleased or disappointed,
likes and dislikes and preferences
form an integrated collection of concepts.
Draft of the actual lecture

.require "memo.pub[let,jmc]" source;
.cb What is Common Sense?

.cb "by John McCarthy, Stanford University"


#. Introduction

	
	It has long been realized, e.g. (McCarthy 1959),
 that humans have certain common
sense knowledge and reasoning ability that enables us to achieve our
goals and that we have not yet succeeded writing programs with
such general abilities.  It is becoming apparent to many people
that this is one of the major obstacles to progress in AI.

	However, it isn't easy to be precise
and comprehensive
about what these abilities are, and this address may be one of the
first attempts to do so.

	One view (Feigenbaum ?) is that common sense should be regarded as
just another knowledge domain suitable for use in rule based expert
systems.  According to this view, the only reason present programs don't
have common sense is that the amount of required knowledge is very large,
and no-one has yet undertaken to express it as a collection of rules.  I
don't agree.  Common sense requires abilities no-one has yet succeeded in
programming and won't be easy.  Moreover, I don't think the amount of
knowledge required is very large, provided it is expressed in as general
a way as humans apparently express it internally.