COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	AI issues suggested by "Naming, Necessity and Natural Kinds" edited by
C00011 ENDMK
C⊗;
AI issues suggested by "Naming, Necessity and Natural Kinds" edited by
Stephen P. Schwartz - Cornell University Press 1977

	This book is a collection of articles by Schwartz, Donellan,
Kripke, Putnam, Copi, Evans, Goosens, Plantinga and Quine on questions
connected with meaning, reference, etc.  From an AI point of view,
the question is, "Which of the considerations in the book must
be taken into account in designing AI systems of a simple character".

1. In his introduction Schwartz discusses meanings of proper names
and general terms.  The modern view is that meanings are often not
given by definitions, but by some kind of extended ostension.  Proper
names are given (say) at the time of baptism and are transferred from
one user of the name to another.  The same is true of a common noun
like cat.  People experience cats and the name refers to the animal
they experience.

	It seems to me that this view offers no difficulty for AI
systems and fits in with an initially naive approach.  An AI system
acquires certain names from its trainer.  It assumes that they denote
something - or perhaps nothing.  It is alert to learn more about what
the names denote.

2. Donellan discusses attributive and referential use of descriptions.
In its mentalese, it would seem that the AI system has no need for
referential descriptions, because these are used only in communication.
If I see a certain man, I have no need to describe him internally
as Smith's murderer, because internally he is described by a pointer
to his property list.  On the other hand, when the AI system has to
communicate, it has the same problem as anyone else in finding a
description that its interlocutor will understand as referring
to a certain object.  Attributive descriptions will be used
internally to describe object which are sought, but they are likely
to be replaced by addresses of property lists rather quickly.
Prompt reification might avoid their extensive use, but it isn't clear
that this is desirable.

	Proper names often are transformed referential descriptions,
and maybe these are sometimes transformed attributive descriptions.
Thus someone might have looked for a place to get his oxen across the
river and the next day referred to the place as the ox ford which
became its name.

*****
A major issue to be studied here is that of whether good AI systems
could be developed by people ignorant of the distinction between
referential and attributive descriptions.  On the one hand,
ignoring the distinction leads to the errors pointed out by
Donnellan.  On the other hand, people have lived for thousands
of years without formulating the distinction until recently.
Perhaps we have made some positive error that now has to be
corrected by making the distinction.  It would be worthwhile to
make the distinction implicitly by refraining from jumping to
some conclusion.  On the other hand, we may have another application
for circumscription.  A reference is both attributive and referential
or it doesn't matter or it's obvious unless there is a reason different.
*****


4. Quine says that the notions of ⊗similarity and %2natural kind%1
used by a child to get the concept of yellow are
alien to mathematics and logic.
design stance versus biological stance
  I dunno.  Some psychological
experiments are relevant here.  For example, what notions of similarity
or kind are not readily learned by experience, but once learned by
experience or teaching are readily applied?   For AI we must ask what
must we build in so that it will be able to learn the same kinds of
kinds that humans learn by experience.  In the case of colors, there
is light invariance to consider.  It may impose additional requirements
on what can be considered a color, since same-color judgments should
not depend too much on illumination.

	It would be interesting to ask if
the meanings of color words have drifted.  Presumably blood has always
been red but have apples?  What has counted as brown or black or gray?

Searle also shows that referential descriptions can be assimilated
to attributive. - Referential and attributive.
Kripke 77 paper (referred to in Beyond Belief) defends Russell against
Donellan.

Could I get copies of the Searle and Kripke papers you mentioned?

"Hoover charged that the Berrigan's plotted to kidnap a high
American official" from 
%3Kripke, Saul (1979)%1: "Speaker's Reference and Semantic Reference" in
%2Contemporary Perspectives in the Philosophy of Language%1, Peter French,
Theodore Uehling jr. and Howard Wettstein (eds.), University of
Minnesota Press, Minneapolis.

Kripke, p.15 "Jones looks like he is about to fall off his new motorcycle"
said when the speaker sees Smith on his motorcycle but thinks it's Jones
who has bought a new motorcycle.  We can say this sentence has no
meaning because of the confusion, but remember that the statement might
continue for a long time saying many correct things about the motorcycle
rider without the speaker ever recovering from his confusion.  Moreover,
when the speaker does recover from the confusion, his hearer has only
a small revision to make of what he has heard.

Kripke 21, p.21 The reason there is no English intuition about whether
Donnellan's "Her husband is kind to her is true" may be that this
is only the tip of an iceberg; endless other distinctions problems
would call for intuitive decisions.