COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub" source file
C00010 ENDMK
C⊗;
.require "let.pub" source file;
.<<
.FONT 1 "basl30"; TURN ON "%";
.FONT 2 "BASI30";
.FONT 3 "basb30";
.FONT 4 "sta200";
.FONT 5 "ngb25";
.>>
.FONT 6 "SUP";
∂AILProfessor J. Gelfand↓Council of Cybernetics↓USSR Academy of Sciences
↓111333, 40 Vavilov Street↓Moscow V-333, USSR∞

Dear Professor Gelfand:

	Thank you for your letter about the panel on "Mathematics,
Cybernetics and AI" at 4IJCAI.  As you know, I am replacing Prof.
Jack Minker as co-chairman of the panel and did not make the
original arrangements, but here are my ideas of how to proceed from
where he left off.

	Both %2cybernetics%1 as orignally conceived by Norbert Wiener
and %2artificial intelligence%1 are concerned with goal-seeking systems.
However, cybernetics, both as initially conceived and as it subsequently
developed, is to a large extent concerned with the case in which the
state space is R%6n%1 and provided with a metric for evaluating
the distance of a state from the goal state together with a rule
that gives the effects of actions.  It seems to me that the emphasis
in cybernetics has been on those cases in which optimal strategies can
be determined mathematically and about which theorems can be proved.
As the mathematics has improved, the cases that have been treated
have expanded.  Besides this narrow sense of cybernetics, institutions
with the word "cybernetics" in their name have undertaken a variety of
activities some of which they also call "cybernetics".

	Artificial intelligence is concerned with the activities involved
in intelligent behavior, and the main initial activity was the search
of discrete spaces such as move trees and proof trees.  The emphasis has
been rather empirical, and the algorithms studied have rarely lent
themselves to mathematical treatment.
It seems quite clear to workers in AI that the mathematical methods developed
in cybernetics have a limited domain of application, especially where the
intelligent system must itself conceptualize the problem, but also in
simpler cases.

	Nevertheless, there is a certain overlap between activities called
cybernetics and those called artificial intelligence.  In particular, some
kinds of search procedures have been studied mathematically, and some
methods such as dynamic programming have been suggested for application
to AI type problems.

	In my view, it would be interesting for the panel members to discuss
what kinds of problems lend themselves to the kind of mathematics
developed in connection with cybernetics.  Perhaps there might be a survey
of these methods from the cybernetics side, and a survey of problems from
the AI side.  This might then result in a fairly lively discussion of
the prospects for the further extension of mathematical methods.

	Perhaps there will be substantial disagreements between workers in
the two domains about what problems lend themselves to mathematical
methods.  For example, I remember that Richard Bellman
was of the opinion that dynamic programming methods would work for
chess, and we AI workers disagreed, but I don't know anyone who will
be at Tblisi who would take Bellman's point of view.

	As to participants, Bertram Raphael and Alexander Lerner and I were
already invited by Minker and confirmed by the the conference chairman
Professor Winston.  Besides this, I have asked
Max Clowes to take part.  As I recall, Andrei Ershov and Victor Briabrin
were also going to take part, but I have not been in touch with them.
I suppose that you will verify their intentions and invite whomever else
you think will make a good contribution.

	As you may already know and can see from the enclosures, there
has been some concern about the possibility that Alexander Lerner might
not be allowed to participate in the panel.  In order to keep matters
clear, I will take the responsibility for continuing his invitation
to participate as Professor Winston has asked me to do.  This is
entirely regular, because IJCAI is an organization of individuals and
not of national organizations, and the Program Chairman, Pat Winston,
was given complete authority over the program at the last Conference
at Stanford in 1973.  (The previous chairman, Nils Nilsson had complete
authority over that meeting).

	Along with Lerner and Raphael, I am inviting Max Clowes to
participate, because he knows quite a bit about both artificial
intelligence and work that has gone under the name of cybernetics.

	I am looking forward to meeting with you in Tblisi or in Moscow
where I am expecting to arrive on Sunday August 31.

.sgn