perm filename REPRES.PRO[ESS,JMC] blob
sn#019553 filedate 1973-01-06 generic text, type C, neo UTF8
COMMENT ā VALID 00002 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 An intelligent computer program that understands the
C00007 ENDMK
Cā;
An intelligent computer program that understands the
physical, psychological and social world well enough to intelligently
solve problems in it must be able to represent facts about the world
in general, particular situations, the purposes of others, and the
laws that determine the effects of actions. In our opinion, this is
one of the main problems that has to be solved before artificial
intelligence is a reality. John McCarthy has been working on the
representation problem for a number of years and will continue this
work. Present and proposed efforts are based on the use of sentences
in first order logic and set theory as the primary means of
representing facts. The meaning of other data structures and of
programs is described by such sentences. The work is facilitated by
the existence of a reasonably convenient proof checking program for
this logic and substantial bases of axioms in machinable form. From
various points of view, Luckham, Feldman, Weyhrauch and Newey are
also concerned with the representation problem.
To make this concrete, consider the problem of travelling
to a foreign city. The following kinds of information have to be
represented. (1) The general facts about airplanes, cars, and
walking and the effects of actions using them. (2) The location
and format of information of various kinds, e.g. airline schedules
in the airline guide and in the knowledge of travel agents (3) The
facts about what other people must do or refrain from doing in order
to travel, e.g. the ticket agent must give you a ticket which he
will do in exchange for money in order that the gate agent will
refrain from preventing you from getting on the plane. (4) There
is also information about what information does not have to be
known in advance because it will be available when needed such as
what gate a plane at the second stage of the journey will leave
from and how to get to the gate. The test of whether this
information has been successfully put into the computer can be
made by having a proof checker check a proof that a given strategy
for making a journey will work barring accidents, strikes, and
other acts of God or government. In our opinion, understanding
the kinds of information required to achieve a class of goals is
necessary before it is reasonable to have question answerers or
natural-language-understanders solve such problems or converse
in natural language about them.
In the next contract period the Stanford Artificial
Intelligence Laboratory proposes to increase our ability to
represent information in computers. In particular, we plan to use
the above-mentioned travel problem as the basis for our work.