COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	BLUMENTHAL@A.ISI.EDU
C00013 ENDMK
C⊗;
BLUMENTHAL@A.ISI.EDU
ai[e88,jmc]		Artificial Intelligence for NAS study

	Artificial intelligence (AI) is the study of how machines can
be made to perform tasks of kinds that require intelligence in human
beings.  Mostly the machines in question are computers, and making
them behave intelligently involves programming them appropriately and
providing them with suitably encoded database of facts about the
domain in which they are to work.  The most successful work in AI
has involved treating it as a branch of computer science, i.e.
as concerned with the relation between a situation involving
knowledge and possibilities for action and the achievement of goals.
Athough information about how humans solve problems is used as a guide,
AI is not primarily concerned with psychology.  Of course, AI research
interacts with research in information processing psychology.

	AI has both scientific and technological aspects, and
these are not very sharply separated.  The present scientific
understanding of intellectual processes is rather limited,
and these limitations are reflected in the technology the
present scientific knowledge can support.  Nevertheless, a
useful technology of ``expert systems'' has developed in the last
15 years, and hundreds of companies world-wide are involved in
exploiting various aspects of it.

	Scientific research in AI is largely concentrated in universities,
especially American universities.  The field is served by
international scientific journals and by national and international
conferences.  Results are published in an entirely open literature.
At least there is no case of some important result coming into
the open after a substantial period of secrecy.

	The American lead in AI science is probably larger than in
most fields of computer science.  The most prominent AI researchers
are mostly American.  This lead is based on an early start in the
1950s and early support by the U.S. Department of Defense via DARPA.
American lead in computer technology was also important, because good
computers are important for AI research.  However, for the last ten
years, the progress of AI science has been primarily limited by
conceptual difficulties rather than by hardware.  In principle, the
next major discoveries could be made anywhere in the world, but in
fact the conceptual advances have mainly been made in the U.S.

	Communist ruled countries, i.e. the Soviet Union, China
and Eastern Europe have done little important work in AI.  This
is the result of their own policies rather than a result of any
attempt to keep information from them.  The ideological objections
to AI became unimportant in the 1950s, so this can't account for
it.  The hierarchical and gerontological structure of their scientific
organization has probably been the main factor in delaying their
start in new fields.  Whereever progress in AI has depended on the
patronage of prominent physicists, mathematicians and biologists,
progress has been slow.
  Also they don't send people abroad enough for
them to absorb the culture of new fields.

	The technology of AI has the following components.

	1. The educational system produces trained and creative
people.

	2. Computer companies produce suitable computers.  Some of
these, e.g. the LISP machines and now a few logic programming
machines (especially Japanese) are designed primarily for AI work.
However, general purpose computers and workstations are also
suitable for AI, except that the IBM 370 and its imitations
have made this rather difficult.  Since the Soviets have based
their largest computer effort on imitating IBM, this has certainly
delayed them.

	3. Languages suitable for writing AI programs are required.
Mostly this has involved dialects of LISP, but Prolog and its variants
have more recently become important.  The Eastern countries haven't
done much here, although Hungarians have become active in Prolog.

	4. Commercial ``AI shells'' have recently been developed.
These systems allow the more convenient development of expert
systems.  This is important, because in expert systems, development
time and the ability to modify the system is the most important
component of cost.  There doesn't seem to have been export of Western
AI shells to the East.  The Institute of Cybernetics in Tallinn,
Estonia has developed and expert system shell called PRIZ that has
even been exported to Sweden.  It seems to be rather limited.

	5. For many applications, expert systems have to be
connected to databases and other programs.  Numerous specialized
products exist for doing this, especially for making connections
to commercial systems based on IBM technology.

	6.  The previous components are all auxiliary to the 
main technology, the writing and installation of the expert
systems.  This is done both by companies needing the expert
systems and by specialist companies working under contract.
Fitting the expert system into the user's operations can be
a larger task than writing it.  Both tasks involve interaction
with the people possessing expert knowledge of the domain in
question.

	Expert system technolgy is not mature in the West and
is just beginning in the Eastern bloc.  Their institutional
systems which often involve major discrepancies between how
institutions are supposed to behave and how they normally do
behave surely make difficulties for computer systems in
general and for expert systems in particular.  Whereever
under-the-table deals including bribery are required,
it will probably be impossible to use expert systems.  Secrecy
also brings difficulties.

	Successful Soviet use of Western AI technology
will depend on their maintaining reliable continued access to manuals,
changes in software and doctrine for usage.  The theft or
illegal export of programs leaves the recipient in a difficult
position.  In the Soviet and East German case this has often
involved giving up compatibility and going on their own after
a certain point.  One can imagine maintaining continued access
through dummy Western companies that transmit bug reports in
one direction and updates in the other, and whether such conduits
exist might be worth investigating.

	If political developments take place that involve increased
Soviet access to Western technology, e.g. in exchange for concessions
in some other area such as Soviet territorial secrecy, the most
important step would be their use of conventional international
channels for maintenance of software systems.