COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Dear Lick:
C00009 ENDMK
C⊗;
Dear Lick:

	I think  the new version  of the proposal  will be  very much
better containing an executive summary and meeting your criticisms in
its separate sections.  Versions  of the sections have been  prepared
but the executive  summary has still not been written.   It is partly
because  I had  to long committed  speeches to  make in  the last two
weeks, but  also  because it  took me  a  long time  to  get my  head
straight about why ARPA should  support AI and what form this support
should take.   Besides the proposal  itself, there will  be a  rather
long cover  letter addressing some  of your points  explicitly rather
than implicitly as in the proposal itself. 

	The  main problems that will  be addressed are  the degree of
goal  scheduling  asked   for,  the   concentration  of   immediately
producible applications,  and the  requirement to  fit in  with plans
produced  by the IPT committee  structure.  In  some respects, we can
meet these conditions, but in other respects they  will be disastrous
both for us and for what DoD should get out of the research. 

	Let me  try one argument  on you now,  however.  Part  of the
research results to be obtained in AI  can be planned for, but AI  is
one of those  branches of science in  which some of the  progress may
come in sudden  breakthroughs.  Moreover, such breakthroughs may have
immense practical consequences,  i.e. if it  led to  an RPV with  the
ability to  do human quality  tactical thought at  electronic speeds.
This being  the case, ARPA has bought insurance by its support of AI.
Namely, it has insured a better than 90% probability  that had such a
breakthrough occurred, it would have happened in the United States in
a lab supported by ARPA.  One  thing it can get is a continuation  of
this  insurance.    Let  me  remind  you  that  nuclear  fission  was
discovered  in  Germany in  1939.    It was  great  fortune  that the
discoverers, who realized the  significance of their discovery,  were
motivated to  escape from Germany, and  it is a further  good fortune
that  the discovery was followed  up in the United  States and not in
Germany.   Indeed, all through  World War  II, the Manhattan  Project
people thought they were racing with a corresponding German project. 
Fortunately, the  world situation  is not  as tense  as it  was  just
before World  War II,  but AI  has the  same potential  for a  sudden
conceptual   breakthrough  that  would  lead   to  results  of  great
importance.  The optimal  strategy for optimizing the  probability of
such a breakthrough is not quite the same as optimizing the detail of
a research plan or the probability of bullets in the next year. 

	The  above is  not a plea  for unconditional  support with no
proposal, and I think your criticisms will help make  a research plan
that will increase the probability of a major breakthrough as well as
increasing the number  of definite accomplishments,  but I think  the
insurance argument is an important reason why  DoD should support AI,
maybe the strongest reason. 

	Now  an  additional  complication.   Because  of  yet another
months' old commitment, I  can't get to the Principal  Investigator's
meeting until Thursday morning.  This  will be the first one of which
I have missed any part.  Therefore, I am asking Les Earnest to attend
the first  day in my  place.  I  think this  will be optimal  anyway,
because he will be running  the Laboratory in the Spring, and because
I think  that  it  will make  it  much  easier for  us  to  mesh  our
activities with IPT's  grand plans.  IF  THIS IS NOT OK,  THEN PLEASE
SEND A MESSAGE TO LES@SU-AI, because I will be out of town. 

	On the budget,  we would like to submit a better proposal for
the same money and period as before.  After you have seen it,  we can
rapidly cut it if this proves necessary, i.e. in a day. 

				Best Regards,

				John McCarthy