COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00006 00003	.bb Cellular automata
C00009 00004	.skip 1
C00010 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source;
.turn off "{"
.cb MODELS OF EPISTEMOLOGICAL SYSTEMS

	We propose to model questions of epistemology by questions about a
robot in a hypothetical artificial environment.  For concreteness, suppose
that we model the universe by a cellular automaton array of known
structure and that the robots are mobile configurations of cells large
enough to simulate computers whose programs control their behavior.  The
universe contains other objects besides the robots, and we study programs
for the robots that may discover facts about the structure of the universe
and the objects within it.

	We can ask two kinds of questions about such system:

	1. What are the various ways the robot may be programmed to study
its universe, and what can it discover?

	2. How can we (outside the system) describe the informational
state and propensities to act of the robots, and can we use them
to model mental qualities such as belief and purpose?  For example,
when can we say that the robot intends to perform an action.

	In studying such questions, many problems of philosophy arise
but in a technical form; their answers are matters of mathematics and
not of science and not of philosophical commitment.  Nevertheless, we
hope that these answers will throw light on philosophical problems
that are not confined to such an artificial environment.

	The applicability of our results will be limited by the fact that
the automaton model already limits the possible universes.  However, as we
shall see, many philosophical questions arise in this model, and after we
have studied them we can look for other models.

	There is a hopeful analogy between this method of studying
philosophical questions and the metamathematical method of studying
mathematical systems.  Instead of arguing about the validity of each
axiom, we take the system as a whole, and study it "from the outside"
using ordinary mathematical reasoning.  Once we understand the properties
of the theories, we are in a better position to decide what systems to
"adopt for ourselves".

	The first related work of which I am aware was E.F. Moore's (1956)
paper in which he studied what could be learned about a finite state
automaton by experimenting with its inputs and outputs.
.bb Cellular automata
.skip 1
.begin verbatim
John McCarthy
Artificial Intelligence Laboratory
Computer Science Department
Stanford University
Stanford, California 94305

ARPANET: MCCARTHY@SU-AI
.end

.turn on "{"
%7This draft of

PUBbed at {time} on {date}.%1