COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	excess material from a letter to Putnam
C00006 ENDMK
C⊗;
excess material from a letter to Putnam

 	1. The program that operates the robot represents information
about the world by sentences of in a first order language.  These
sentences include those initially put in by the programmer and others
added by the program as a result of its inputs and as a result of
internal computation.  Sometimes these new sentences are added by
processes of deduction involving some kind of sense data sentences
and sentences already present.  Some inputs are interpreted
as communications in the first order language.  These may go into
the database in some such form as  (said John '(all (x) (implies
(piece-of-snow x) (white x)))).  (This is not necessarily advocated
as the best AI way of telling the computer that snow is white).
While many new sentences arise from previous sentences by ordinary
first order deduction, it is by now well agreed that other
processes of inference are also necessary.  This includes the
non-monotonic reasoning processes which have recently begun to
be formalized.  (See the journal Artificial Intelligence, April 1980).
Such approaches to artificial intelligence were first proposed by
me in 1958 and have had their ups and downs in popularity.  Now seems
to be an up.

	A predicate  $true  applied to sentences or to some
more abstract entity, e.g. propositions, is needed for the robot.
The issues I want to discuss don't depend on whether sentences
are used.  Moreover, even if we decide that the proper argument
of  $true is something other than sentences, we may still write
$true1(p,c) to say that the sentence  $p  expresses a true
"proposition" in context  $c.  Therefore, I'll use sentences
in this letter.

	It seems to me that the robot's main use of the predicate
$true  is in applying quantified statements about truth to
particular sentences and then using a reflection principle
to eliminate the predicate  $true.  Thus we have the argument

1. Whatever John says is true. - in the original database of my program.

2. John just said "Snow is white".

3. Therefore, "Snow is white" is true.

4. Therefore, snow is white.

	The constant assertion "Snow is white" doesn't need to be
said to be true in order to be asserted.

	Here are some examples of facts that it may be useful to put
in the robot's initial database.

1.