perm filename REPRES.P75[1,JMC] blob
sn#172726 filedate 1975-08-09 generic text, type C, neo UTF8
COMMENT ā VALID 00002 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 .SSS Formalization of general properties of the world
C00009 ENDMK
Cā;
.SSS Formalization of general properties of the world
A computer program with general intelligence is one that can
behave intelligently in any goal-seeking or problem-solving situation.
At present there is no computer program that can even understand problems
in general, in the sense that after receiving a presentation of the
problem, it could tell a purported solution from a remark about the
weather. In order to make a program
with general intelligence, we have to formulate general
information about the world and about goal seeking, and we have to
provide a general formalism within which information special to the
particular problem can be expressed. The language of the general information
has to be also suitable for the expression of the particular information.
We will try to be more specific.
Actions are taken and events occur in situations and result in new
situations. The most general information with which a system must be
provided must include such facts, because goal-seeking involves following
a strategy of action that produces a situation with desired properties.
[McCarthy and Hayes] discusses this problem and introduces a formalism
called the %2situation calculus%1. This formalism is intended for
problems in which there is only one actor and the new situation is
the result of his action. It cannot treat problems in which many actions
are going on at the same time. It also requires that the actions be
discrete, i.e. that they terminate in some definite state of the world
rather than merely continue. Thus it can treat the action of Pat going
to New York which terminates in his arrival there, but it
doesn't treat the process of his travel as a background for other action.
Even in this %2quasi-static%1case there are difficulties known
as the %2frame problem%1 and the %2qualification problem%1. The problem
is further complicated by the need for %2epistemologically adequate%1
formalisms that can express the knowledge actually available rather
than, for example, the positions and velocities of all the particles
involved - representation well suited to the formulation of %2physical
laws%1, but not to representing the facts actually known.
Besides the general properties of purposeful action, general
intelligence needs to know the properties of physical objects that
can move about in space and can be made and destroyed. Next come the
properties of knowledge and sources of information. Finally, we have
purposeful actors and what they may be expected to do in situations
given their goals and beliefs.
Recent work has concentrated on the following problems:
1. Devising a better formalism than the %2situation calculus%1
that will permit expressing facts about partial situations and about
concurrent events and actions.
2. Formalization of facts about knowledge, beliefs, and sources
of information. Considerable progress was made in collaboration with
researchers at Kyoto University in formalizing knowledge and belief, and
this collaboration is to continue with a Japanese financed visit to
Stanford next Spring. The work in formalizing knowledge in general has
led to a parallel treatment of how to to describe what information is
contained in files, e.g. how to say that a certain file at Lawrence Berkeley
Laboratory contains information about the distribution of information
by ZIP codes in such a way that a program asked for the income distribution
in Palo Alto can use this description to call the LBL computer on the ARPA
net.
During the period of the contract, McCarthy and interested graduate
students will continue work on these problems. It is not presently possible
to say what will be achieved by when in the general theory, but we can be
more definite about the application to the description of information in
computer files. Namely, we expect to be able to describe %2name and
address files%1 in computers on the net well enough so that a program
using these descriptions will be able to look them up by June 1976.