COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	searle[f86,jmc]		Further response to Searle
C00009 ENDMK
C⊗;
searle[f86,jmc]		Further response to Searle

	I found Searle's response to my previous criticism difficult
to reply to.  I was accused of abusing him and misrepresenting him,
and I didn't think I did either.  I don't feel like a textual analysis
of what I said and what he said, and I don't think readers would be
much interested.  I notice that Hofstadter has had a similar experience
of being accused of misrepresenting Searle, although when I read his
remarks the first time, they seemed on target to me.

	Therefore, I have decided that this response will concern
some issues that Searle's comments raise in my mind without attributing
any particular views to him.

1. First a remark related to the Chinese room.  Let's take an
extreme form of the Chinese room in which the man has memorized
the rules for manipulating the Chinese characters in order to
carry out a written Chinese dialog but doesn't understand what
the characters refer to.  It is relevant to remark that this would
be very difficult and the processing would be quite slow.  All
facts about the world required for intelligent Chinese conversation
would require a separate representation from those used by the
man in his own conversation.  Some facts would be ``known'' to
the Chinese ``personality'' and others to the normal personality.

In this case we could not identify the personality with the hardware.
Normally we can say ``John Smith knows that Napoleon lost the
battle of Waterloo and has a piece of paper with that assertion
in his pocket'' without ambiguity.  In the Chinese case we might
say that the Chinese personality program Smith executes from
time to time knows many poems of Li Po while Smith knows only
that Li Po was a famous Chinese poet.

The Chinese room puzzle is similar to various stories about
multiple personalities in the same human body.  However, the
stories use a concept analogous to time-sharing in computing
rather than one resembling interpretation in computing, which
corresponds better to the Chinese room.

These remarks give my opinion of the resolution of a certain Chinese
room puzzle.  Any resemblance to a commentary on Searle's opinions
on a puzzle with the same name may very well be coincidental.

2. Non-holism.  Some people have asked ``what is mind'' and come
up with a long list of requirements that a system must meet before
it can be called a mind.  Some of these requirements seem vaguely
stated to me and many other AI people.  For this and other
reasons we have chosen a nonholistic approach.  We have a list
of mental qualities that doesn't claim to be exhaustive.  We
propose to formalize ever larger subsets of these mental qualities.
Such formalizations have both syntactic and semantic aspects, where
semantics is taken in the formal Tarskian sense.  Namely, the semantic part
includes predicate and function symbols, axioms and rules of inference.
The deductive rules of inference don't seem problematical, because
our purposes can be served by first order logic, although we
aren't inhibited from using other more-or-less standard logical
systems when that is convenient.  Choosing good predicate and
function symbols and a good set of axioms is the most substantial
part of the enterprise.  Choices that have been proposed so far
lead to very limited systems.  The Dreyfuses, et. al. consider
these difficulties fundamental, but we don't.  The semantics
of these concepts are treated semi-formally.  We ask what models
our axioms should admit.

The non-monotonic modes of reasoning are more problematical, because
formalized non-monotonic reasoning, an invention of AI, is only
beginning to be studied.  If monotonic reasoning is a precedent,
important innovations in formalizing non-monotonic reasoning are
likely to continue for the next several hundred years.

Mental qualities that have been studied include belief, knowledge,
intention and ability, where ability is not strictly mental.

It seems that human experience provides a justification for
the non-holistic approach to mental qualities.  A small child
ascribes beliefs to other people.  It is easy to show that
a child's notion of other people's mentalities are non-standard
and don't include many qualities that would be mentioned by
adults.  For example, a young child of my acquaintance doesn't
yet distinguish between making a mistake and being bad.