COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub[let,jmc]" source
C00006 ENDMK
C⊗;
.require "let.pub[let,jmc]" source;
∂CSL ∞

Dear Colleagues:


	Here are some contentions aimed at showing that the single most
important relatively inexpensive improvement in supporting the science of
artificial intelligence would be the creation of some (up to 20)
postdoctoral fellowships in AI.  These fellowships would be open to
national and perhaps international competition. The fellows could take
them at any hospitable institution, they would have no specific duties;
they would have a duration of three to five years, and would not be
renewale.

1. Progress in AI is idea-limited.

2. Achieving human-level AI requires new concepts.

3. Historically, conceptual scientific breakthroughs have most often come from
well-prepared, strongly-motivated young men free to follow their own ideas.
One example is the Crick-Watson discovery of the structure of DNA.

Comparisons: 100 fellowship years is more likely to produce new concepts than the
next 300 Lisp Machines--at about the same cost.

The best presently available positions are assistant professorships in major
universities. However, these carry many other duties, and often involve 
obtaining external research support, and this encouages adopting short term
research goals already well understood by granting agencies.

Research associate positions are sometimes good, but they usually involve helping
an already established principal investigator achieve his goals.

Who should do it?

1. DARPA is the largest single supporter of AI research, but it may be
institutionally difficult for it to support fellowships.  

2. NSF should surely do it.

3. AAAI can afford to support a few and this would encourage others.

4. Corporations and private donors should be solicited - perhaps by AAAI to
supplement AAAI's own support.

5. We should proceed on a modest scale.  I don't advocate creating enough
fellowships to support even a quarter of the new PhDs in AI.

Assistant professorships, research associateships and industrial research 
positions also play an important role.

.sgn

cc: Saul Amarel, DARPA
Kent Curtis, NSF
Executive Council AAAI
Officers AAAI
Officers SIGART
Past Presidents, AAAI