COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub" source
C00004 ENDMK
C⊗;
.require "memo.pub" source

	This is a first class proposal and should definitely be
supported as proposed.  It is one of the few AI proposals I have
seen that has a proper balance between theory and experiment.
The discussion of the limitations of current problem expression
and problem solving formalisms is excellent.

	McDermott is right in paying attention to the philosophical
work on temporal logic and counterfactuals, but I hope he isn't too
optimistic about finding useful results there.  In particular, the
notions of counterfactuals in the papers he cites by Goodman and
Lewis are unlikely to help.

	He should look at McCarthy's recent "Ascribing Mental
Qualities to Machines" for another view of counterfactuals and
his "Circumscription" for an approach to non-monotonic reasoning
that supplements his own and may be more computational.

	I hope he will make the discussion of these problems
contained in the proposal into a paper.