COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00009 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source

	I think this article should be published, perhaps substantially
cut.  Although it isn't very good, it may be the best of its kind.  It is
the most thorough criticism of the ways in which specific computer
programs failed to capture the intellectual mechanisms which their
designers were hoping to embody.  Until someone does this job better, and
it wouldn't be hard for someone with a wide acquaintance with the
literature, the AI field is better off with criticism with Dreyfus's
essentially negative motivation than with none at all.

	I do not agree with Dreyfus's conclusion that formalization is a
hopeless task.  I merely think it is as difficult as many of the classical
problems of science.  For example, the genetic code took nearly a hundred
years after Mendel to unravel, many false leads were followed, and many
criticized the whole enterprise of studying the material basis of
inheritance.  The problem of how embryonic development is controlled still
has no plausible solution.  For this reason, he needs a better argument
than a claim that there has been little progress in 10 or even 20 years.

	Dreyfus's style engenders defensiveness and detracts from getting
workers in AI to pay serious attention to his criticisms of specific work.
His crowing about how right he was will merely divert defenders of AI into
searching his earlier work for wrong predictions to counter those he
claims were right.  His emphasis of "admissions contrary to interest" will
induce people to be more circumspect, which will be unfortunate, because
AI needs more self criticism.  Therefore, the article would be better if
he were induced to tune down its shrillness.

	It would be even better if Dreyfus, like his student John
Haugeland, were motivated to focus his ideas on specific challenges to AI,
and were to come up with a list of the simplest problems that he thinks AI
can't solve.