COMMENT ⊗   VALID 00005 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00009 00003	This version of hannov[w86,jmc] pubbed at {time} on {date}.
C00010 00004	topics:
C00011 00005	.end
C00012 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source;
.cb "Artificial Intelligence, Expert Systems and Common Sense"

.cb "by John McCarthy, Stanford University"

.<<hannov[w86,jmc] Notes for Hannover Fair lecture on AI and common sense>>

.item←0

	I suppose most of you are from industry and
are mainly interested in the possibilities of applying
artificial intelligence technology.  Nevertheless, I want
to begin by saying what kind of science we are talking about.

	Artificial intelligence is the science involved
in making machines, i.e. computers, behave intelligently.
Many people, hearing that, ask for a definition of intelligence.
I can't give one, and I'll digress into philosophy in order
to say why not.

	``Intelligence'' isn't a defined concept.  It's what
philosophers have come to call a %2natural kind%1.  Namely,
%2intelligence%1 is the name of a phenomenon, and to provide
a correct definition, would require knowing all about that
phenomenon.  Therefore, instead of defining intelligence, all
that can be done at the present state of knowledge is to tell
you something about it.

	Here are some facts about intelligence in general and
AI in particular.

	#. Intelligence is a property of systems that achieve goals
and maintain themselves in complex environments by using knowledge
about the environment in general and also about the particular
situation.  The processes that use the knowledge often predict
the consequences of actions and other events.

	What methods an intelligence has to use obviously depends
on what kinds of problems are presented by the its environment.
However, human physical environments, abstract mathematics and games
present more similar problems than might be imagined.

	#. Intelligence involves abstract properties independent of
whether the goals are being achieved by humans, by Martians or by
machines.  A large part of the theory of intelligence has, for this
reason, a mathematical character.  Certain computational means are required to
to decide what to do in certain situations regardless of who or
what is doing the deciding.

	#. Intelligence isn't a unitary phenomenon.  There are a variety
of intellectual mechanisms and we understand some of them.  We can make
computers perform impressively on tasks that only require the mechanisms
we understand.  Tasks that require mechanisms we don't understand are
beyond present technology.

	#. Having fast computers with big memories can partially compensate
for lack of understanding, but some tasks cannot be done with any
remotely feasible computer until our understanding improves.  In
some areas major conceptual breakthroughs are required, and it is
impossible to be sure whether they will take five years or five hundred.
Remember that it was 100 years from Mendel to the determination of the
genetic code.

	#. The intelligence of a program involves both its facts and
the procedures it carries out.  Without procedures nothing would ever
happen, but progress in AI has involved putting more and more of the
behavior of the program in the form of facts.  The information includes
general facts about the world, the facts about the particular problem
the system is to solve and facts about how to carry out the reasoning.

	#. The present state-of-the-art in AI is characterized by
certain successes and limitations in expressing different kinds of
facts and making the computer carry out different kinds of reasoning.


%3Common Sense%1

	My own work in artificial intelligence since 1958 has concerned
the goal of making computer programs with %2common sense%1.  By this
I mean a program with the knowledge and reasoning ability possessed by
any human who isn't feeble-minded.  This seemed difficult in 1958, and
it is still difficult.  Fortunately, for the application of AI to
practical problems, common sense isn't always required.

	What is common sense?

	We still can't fully characterize it, but we can describe some of
its components.

.item←0

	#. Knowledge of the effects of events including especially the
effects of actions.

examples: effects of travelling, effect of sitting or standing, dying

	#. knowledge of space and objects

	#. relations between appearance and reality
This version of hannov[w86,jmc] pubbed at {time} on {date}.
.if false then begin

expert systems,
prolog
common sense
topics:

What is intelligence.

The history and state of AI

Expert systems - their limitations and strengths

Common Sense

Recent results - non-monotonic reasoning

Some formulas
.end