COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	lenat@mcc.com,guha@mcc.com/cc
C00007 ENDMK
C⊗;
lenat@mcc.com,guha@mcc.com/cc
Some possibly useful prose
%concil[w90,jmc]		Conciliating approaches to AI

In spite of the polemical character of Brian Smith's criticism of
the Lenat and Feigenbaum approach to AI and the logic approach,
his article leaves the impression, especially in the notes
mentioning changes in CYC, that the differences among the
approaches, including the situated approach are significantly
compatible.

	The CYC approach has recently incorporated more logic.
Both the CYC approach and the logic approach are developing
ways of formalizing context dependence although in different
styles from the situated approach.  Thus the logic approach
treats context as an additional parameter of a logical
sentence or, making sentences into objects,
writes  holds(s,c)  to say that the sentence
s holds in the context c.  Perhaps the situated approach is
moving more towards studying the common sense world rather than
concentrating so much on studying the expression of assertions in
natural language.  The description of the situated approach in
Situations and Attitudes was incomprehensible to people not
steeped in it, but Barwise's new book is much clearer, especially the
introduction.

	Any merger is unlikely in the forseeable future, if
only because it would require more scholarship than any
of us has shown to fully understand all the trends within the
various approaches and explain their overlaps and differences.

	The motive for suggesting some reconciliation isn't
pacifism; controversy is helpful to AI.  Moreover,
it isn't unity in face of an external threat, even though
reconciliation with connectionism is unlikely in the
near future.

	The reason is to advance AI.  The comparison and contrast
of approaches should take the form of exponents of one approach
trying to find the easiest problem the criticized approach won't
solve.  exponents of the criticized approach will then try to
solve it and also return the compliment.

	The logic approach is the one best understood by
advocates of differet approaches---perhaps because it is clearer.
Therefore, it welcomes puzzles that others think it won't solve,
the simpler, the better.  The challenge presented by the Yale
Shooting Problem has certainly advanced logical treatments of the
situation calculus.  (Although Vladimir Lifschitz anticipated the
problem it presented in connection with the McCarthy treatment of
the frame problem, its presentation in the context of criticism
of the whole approach excited more activity among the logicians
than otherwise would have occurred.)

	McCarthy's puzzle of Wellington hearing of Napoleon's
death was useful in advancing CYC.

	It would be good if Brian Smith or some other exponent of
the situated approach would give as simple as possible a problem
that they think requires it.  Both logic advocates and expert
system advocates would like to try methods that change the
way logic is used, i.e. new first order axiomatizations and
reifications, rather than radical transformations of mathematical
logic itself.