COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Pages 25-37
C00011 ENDMK
C⊗;
Pages 25-37

June 20, 1985
Restriction of Information Transfer in AI and Related Fields
by John McCarthy
Stanford University

These remarks were triggered by a remark by David Brandin of SRI International
at a U.S.-Japan conference on artificial intelligence. He said that SRI was
negotiating the formation of an SRI-Cambridge University research center in
Cambridge, England to focus its efforts on natural language. He said (as a side
 remark) that the U.S. Government (he didn't say what part of it) favored the
project and that the necessary technology export licenses would be forthcoming
 except perhaps in AI. (I hope this is an approximately correct rendition 
of his remarks).

This triggered in my mind a picture of bull-in-the-china-shop behavior on the 
part of uninformed Government bureaucrats that might seriously jeopardize 
international co-operatin in AI research. On further thought I'm still 
worried. I don't know exactly what U.S. Government policy is, but here is what
I think is feasible and desirable.

1. In so far as AI technology is incorporated in military systems, the AI 
part of the systems should be treated like any other part. Programs can be 
classified as can their exact function and the manuals for their use. 
Because AI systems are hidden in computer memory and not 
apparent in the external appearance of
 the equipment, the opportunities for maintaining secrecy are better than
 average. Moreover, since knowing a program helps to fool it, there are big
 advantages in maintaining this secrecy. Also, it should be possible to keep
 programs secret for many years even in equipment that is in wide use.

2.Under peace time conditions it is infeasible and undesirable to put any part 
of AI basic reserach under classification. It will not be possible to get 
highly capable research people to do research in this area if they can't 
publish research papers.

3.l There will be continual tension between companies and their researchers
 about keeping the results of research proprietary. It will be harder for 
companies to get researchers to observe stiff proprietary conditions, because 
AI research has traditionally been led by universities. If matters get too 
nasty, I would advocate that AAAI and similar organizations make certain
workshops open only to people prepared to sign a statement saying that they
have no obligation to withhold the results of research in the area of the
workshop. Such measures have not up to the present been desirable.

4. Proprietary products, hardware and software, by companies are subject to
the same considerations as with other products. I don't want to discuss the
the software piracy problem in these remarks.

5. It is feasible and disirable to delay the distribution of unpublished 
research in AI to the Soviet Union and other unfriendly countries. Keeping
such addressees off report distribution lists will do the main job. It is
desirable for two reasons. First, delaying Soviet acquisition of military 
useful technology reduces our defense costs. Second, beyond that we 
can use our technology as a bargaining point in reducing their secrecy in 
technology and in other areas. Their secrecy in technology is far greater
than anything we could possibly adopt without serious damage to our progress.
(They probably damage their own progress, but that't their problem). Besides,
bargaining about their technological secrecy, we could also try to trade some 
of our technology secrecy for their restrictions for their prohibition or 
inhibition of travel by foreigners in most of their territory.

While they can steal  some technology by clandestine purchases and smuggling
equipment, stealing is no substitute for licensing which provides bug fixes,
help in using the technology, and preview of the direction in which it is
moving. It is relatively easy to delay the transfer of technology to the
Soviet Union because of their restrictions on travel by their own people.
These restrictions have existed since about 1930 and are unlikely to change
without a substantial liberalization of the Soviet Union that might greatly
reduce the threat they present to world peace.

6. I believe it is a bad idea to try to prevent technology transfer to
friendly countries for competitive reasons by any means other than proprietary 
restrictions by companies. A technology secrecy was is likely to be far more
harmful to progress than any trade war. The U.S. is likely to be the main
loser in technology secrecy wars with any countries whose native language isn't
English, especially Japan. If foreign researchers are persuaded to stop
English language publication and report writing by their governments, we are
in trouble.

7. A single big scandal of unreasonable behavior by the U.S. Government 
may have effects t6hat will take a long time to reverse.

8. Enforcement of technology transfer licensing requirements applied to
reports on AI research can destroy the international co-operation that has
existed in the field. It can also triggter the formation of ani-U.S. 
combinations in technology.
end
-------