COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Dear Dan:
C00008 00003	the meaning of philosophical terminology in terms of approximate theories
C00009 ENDMK
C⊗;
Dear Dan:

	Thanks for your ``Evolution, Error and Intentionality''.  You
seem to have really fathomed the difference of opinion between us
and Searle, et al.  I hadn't heard the phrase {\it derived intentionality}
before; who originated it?

	You seem to have forgotten my old paper ``Ascribing Mental Qualities
to Machines'' that appeared in Ringle's volume, or else perhaps you
only saw the garbled version that was a Stanford AI Memo.  (My secretary
printed the wrong file that contained a false start).  It has some points
that are worth adding to your account.  Here's the exact reference, and
a copy is enclosed.

{\bf McCarthy, John (1979)}:
``Ascribing Mental Qualities to Machines'' in {\it Philosophical Perspectives 
in Artificial Intelligence}, Ringle, Martin (ed.), Harvester Press, July 1979.

	1. The argument from cryptography.  The examples you and others
cite of artifacts whose behavior can have more than one intentional
interpretation are misleading, because they are rare and arise only when
there is some symmetry to the problem.  There are also examples in
which the interpretation is essentially determined except for some minor
detail.

	Most often there is only one
intentional interpretation, and for objects that arose neither by design
nor by natural selection there is almost always no non-trivial intentional
interpretation.  The reference to cryptography is to the fact that
cryptograms don't have solutions that differ significantly from the
original plain text.  Shannon in 1948 discussed how extremely improbable
this is when the key is substantially shorter than the information
content of the message.  Philosophers have cited hypothetical examples
of a chemistry book that might be a novel in another language or something
like that.  A book admitting such an interpretation has a similar probability
to the philosopher dying, because all the air molecules happened to
rush away from his nose.

	Incidentally, I have two examples of simple substitution cryptograms
admitting different interpretations that make equal sense.
  I give one solution of each
cryptogram, and the problem is to find the other.  ``Le prisonnier
est mort, il n'a rien dit.'' and ``An optimist is a guy without
much experience. - Don Marquis''.

	Anyway reasonably elaborate robots will also admit only one
intentional interpretation.

	2. 

	Your discussion of what words mean to the user makes it even
appropriate to tell you about my 13 month old son Timothy, who is just
beginning to talk.  First of all, his primary use of words is naming
for its own sake, rather than making the word part of a request.  This
illustrates a point about evolution.  Since speech is valuable instrumentally,
it is valuable for infants to practice speech for its own sake rather
than only practice it when there is a perceived reason for a particular
speech act.

	The word ``wah-wah'', a mispronunciation of ``bow-wow'', has dog as
only one meaning.  It is excited by seeing a dog or cat or other ground
animal, seeing an animal in a book or on TV or hearing a dog barking.  He
also utters it at random times, which may mean that he is thinking of one
of these.  The word ``ba-ba'' means Timothy going somewhere.  In spite of
instigation, he refuses to use it when someone else is leaving.  A
fortiori it is the name of a phenomenon and not a greeting.
the meaning of philosophical terminology in terms of approximate theories
the computer science interpretation of AI
	- about the relation between goals and mechanisms for their
	achievement.
approximate generic points as an illustration - though not to Dennett
not enough logic and mathematics
Go through the Shannon probabilistic theory in more detail.