COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	quine[e87,jmc]		Notes on Quine's Natural Kinds
C00007 ENDMK
C⊗;
quine[e87,jmc]		Notes on Quine's Natural Kinds
(In his Epistemological Relativity)

	Some philosophical problems must be solved for successful AI;
others must be evaded.  Quine's natural kind problem seems to be
of the second sort.

	He begins with Hempel's paradox of confirmation.  Seeing a
black raven confirms the hypothesis,

ARB:	"All ravens are black",

but seeing a green leaf doesn't even though it is an instance of 

ANBNR:	"All non-black objects are non-ravens"

which is logically equivalent to "ARB.  He proposes that the complement of
a "projectible" predicate, using Goodman's terminology, need not be
projectible.  Quine then goes on to discuss similarity and kinds.

	From the AI point of view, it looks like we want to try to
evade this problem rather than solve it.  Presumable a baby learning
language is prepared to regard a black raven as confirming ARB.
His language isn't adequate to formulate the equivalent ANBNR, let alone
realize that it is equivalent to ARB.

	A simple approach to the problem involves following the example of
the baby.  Namely, make the robot use a subset of predicate logic for the
formulation of such hypotheses.  In this subset the contrapositive of a
sentence doesn't necessarily exist.  This subset may be used as a reified
sublanguage L1 of a larger language L2 including full logic.  Thus the
sentences of L1 are objects of L2.  Exactly what should be in L1 is in
question.  How much logic do we want including how much quantification?

	The baby has certain built-in prejudices about what similarities
and contrasts are important.  Color words are not learned early in spite
of parental efforts, and this suggests that similarities and contrasts
of color are not primary.

	Quine considers certain concepts of similarity to be intrinsic,
but he doesn't consider that some of the kinds may also be intrinsic.
This possibility somewhat vitiates his discussion of apple, because
he supposes that the individuation of apples also has to be learned.
However, it seems likely that individuation of objects is intrinsic,
or at least is learned long before there is any question of language.
Contrary to what would be expected from Quines treatment, apple is
easier than yellow.  There is also the work that suggests that the
individuation of the colors is actually intrinsic.  All we have to
learn is the names for them.  Indeed sharp distinctions of color
are especially characteristic of artificial objects.  Many natural
objects are not sharply distinguished by color.

How kinds change.  Science discovers new projectible predicates, e.g.
positively charged particles.

p135 "A man's judgments of similarity do and should depend on his thoery,
on his beliefs; but similarity itself, what the man's judgments purport
to be judgments of, purports to be an objective relation in the world."
This may be questionable.

Must the dispositional definition of solubility be legitimized by the
 science of chemistry --- and then
be discarded as superfluous?