COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	dreyfu[s86,jmc]
C00007 00003	vijay@ernie.berkeley.edu
C00010 ENDMK
C⊗;
dreyfu[s86,jmc]

vijay@ernie.berkeley.edu
Reply to Dreyfus's
Copyright 1986, John McCarthy

	The history cited by the Dreyfus's is approximately correct as far
as it goes, although I am perhaps naturally inclined to consider
incomplete the list of ideas they mention, since it doesn't mention the
"Programs with common sense" paradigm.  This research programme seemed
difficult to carry out when I proposed it in 1958, and for that reason, I
always considered it likely that AI would turn out to be a difficult
scientific problem.

Remark: The people who place their hopes on parallel processors
are often distinct from those who postulate still vague notions
of holism.

     "Another human capacity which computers  functioning  as
analytic engines cannot copy is the ability to recognize the
similarity between whole images. Recognizing two patterns as
similar,  which  seems  to  be  a  direct  process for human
beings, is for a logic  machine  a  complicated  process  of
first  defining  each pattern in terms of objective features
and then determining whether, by some  objective  criterion,
the  set of features defining one pattern match the features
defining the other pattern."

	It cannot be excluded that other computational processes besides
logical inference will be required for artificial intelligence.  Even
non-monotonic inference may not be enough.  In fact my 1958 paper
mentioned this possibility but proposed to use the processes under the
control of the logical reasoning proposed as corresponding to human
conscious thought.  However, it is important to recognize that our ability
to observe our own brain processes is very partial, and therefore the fact
that we often don't identify subprocesses of the recognition of similarity
doesn't prove that there aren't any.

	The common sense knowledge problem is indeed unsolved
to this day.  However, you need more than that fact to prove
it unsolvable.  After all it was formulated in the 1950s, and
the idea of formalized non-monotonic reasoning didn't come along
till the late 1970s.  It's just a fact that science takes a long
time.

"But  it  just  may  be that the problem of finding a
theory of common sense  physics  is  insoluble.  By  playing
almost  endlessly  with  all sorts of liquids and solids for
several years the child may simply have built up a repertory
of  prototypical  cases of solids, liquids, etc. and typical
skilled response to their typical behavior in  typical  cir-
cumstances.   There may be no theory of common sense physics
more simple than a list of all such typical cases  and  even
such  a  list  is  useless  without a similarity-recognition
ability."

It isn't clear how the usage of "typical" here differs from
the generality required of a theory.  The similarity-recognition
ability mentioned here and above is undescribed.  However, it
must differ from the hologram type similarity hinted at above,
because it has to recognize correspondence of parts as well
as correspondence of wholes.
vijay@ernie.berkeley.edu
1986 May reply to the Dreyfus's

Copyright 1986, John McCarthy

	With their reply, the Dreyfus's enter the arena of theoretical AI
as participants, i.e. they attempt to propose mechanisms for intelligent
behavior.  Welcome.

	An attempt to propose mechanism in AI may be criticized on two
grounds.  First, one may claim that a mechanism is not well enough defined,
i.e. that the attempt to propose a mechanism has failed.  Second, one may
claim that in so far as a mechanism is proposed, it won't work.

	I have doubts on the first grounds, but since they refer
to a book I haven't time to read for this reply, I let this pass.

	As to the second, the mechanism seems to be a variant of
behaviorism, i.e. learned S-R relations are postulated.  The course
of cognitive psychology has been away from this.  Experience suggests
that if any mechanism is to work, it must be much more strongly
based on internal state, i.e. the theory must take internal state
into account to a greater extent than supposing the internal state
to consist of learned S-R relations.

	There is a final point that really should have been in my
first commentary.  We advocates of logic are basing ourselves on
its utility in combining information obtained from different
sources.  The Dreyfus's outline doesn't mention this problem.
Each skill seems to be learned separately.

	For example, Sherlock Holmes combined the fact that
the dog didn't bark with the fact that the horse trainer was
the dog's master and inferred that the trainer stole the horse.

	The Dreyfus theory would ascribe no utility to the fact
that humans communicate mainly in declarative sentences.