[Lecture Notes by Prof Yuen]

(This was written in 1997, shortly after the IBM Deep Blue defeated
 Gary Kasparov in a 6-game match)

Deep Blue, AI and Chess
-----------------------

Since Deep Blue's historical win over Kasparov, commentators have tended
to play down the significance of the event. Computing people usually say
that Deep Blue's chess analysis is by brute force, considering every
possible move, instead of the human method of intuitively eliminating most
of these moves as unsuitable and concentrating on the more relevant
possibilities, while chess people say that Kasparov used the wrong anti
machine strategy and played below par, and that Deep Blue's capabilities
have not been tested against a variety of opponents. All that is true,
but the fact remains that Deep Blue played extremely powerful chess, and
will be even stronger next year. It might not yet be guaranteed to beat
any human player, but would certainly reach that stage in the near future.

From those chess lovers who feel the Deep Blue win would provide good PR
for the game and make its future brighter, I must beg to differ: a
decline in the status of chess is difficult to avoid now that the machine
"had it all figured out". It is true that humans continue to have Olympic
games even though machines can run faster, jump higher, etc., but the
analogy is not really applicable: there have always been animals that run
faster and jump higher than humans, and so to have machines doing the same
was no great blow to the human ego. The human's ability to do certain
things that animals could not do was an indicator of its superiority, with
even some religious significance. Now that a machine can go it better, the
very "human" response of the humans will have to be "oh it is not a big
deal; it is only a game, like scrabble or monopoly".

In fact, why is chess more highbrow than scrabble and monopoly? It is
more historical and more international because it involves purer logical
reasoning, independent of the language and cultural backgrounds of
the players. It attracts a more intellectual type of follower, who could
probably have good careers as mathematicians, accountants or architects,
but choose instead to devote their lives to playing, analysing and writing
books about chess. But if no matter how hard they try, they will never be
as good as a machine, it is hard to imagine that such people would 
continue to select chess as their career; again the situation is different
from running and jumping, because sportsmen like Donovan Bailey do not
really have the same career choices.

It is correct to say that Deep Blue is not intelligent in the way a human
is, and it does not learn in the way a human learns. Machines will
continue to solve problems in their own laborious and clumsy way, but its
way is not necessarily inferior. Humans summarize their memories and
experiences in some unknown ways, and use these to make intuitive, often
unexplainable, decisions that seem to be "right", and to seize upon the
important aspects of a problem. Machines cannot do this, but it can
remember great amounts of details accurately, and with the right program,
use the details effectively. 

For example, the performance of chess programs took a significant leap a
few years ago when programmers put in code to guess what moves an opponent
might make and search for the best answer for each possibility, while the
opponent is thinking; when the opponent actually makes one of these moves,
the machine can respond very quickly, thus reducing the use of its own
play time. This occurred frequently during the recent match, with Deep
Blue consistently putting Kasparov under time pressure. A human player
cannot do the same, because he would not be able to remember all the
possibilities clearly; even if he might have guessed and analysed an
opponent's possible moves already, he would still have to spend time
thinking after the opponent actually makes a move.

On learning, the machine has the great advantage that it does not
forget and can add to its stored knowledge constantly. Further, it can
easily transfer its knowledge to another, bigger and better, machine
whereas the next world champion after Kasparov has to learn chess from
scratch; no matter how hard he studies the old games of Kasparov and takes
lessons from him, the new champion cannot be sure that he has acquired all
the experience that Kasparov has stored in the head. Thus, in some ways,
machine learning is more effective than human learning.

Some commentators have mocked the machine's lack of creativity. In fact,
it is easy to add creativity, though that too would involved the clumsy
and laborious machine way: you randomly generate all the possiblities, and
use predefined facts and rules to eliminate unsuitable ones, producing the
better possibilities. Like Kasparov said after the first game in last
year's match, the machine uses quantity to produce quality, and this
year's result shows only too clearly that greater quanity can indeed lead
to better quality.

The victory of Deep Blue may be no breakthrough in computing, AI or chess,
but it is all the same a very significant event. I strongly advise humans
against complacency.

PS: Gary Kasparov lost his (human) champianship match in November 2000. In
fact he had been playing under par ever since the match with Deep Blue,
which shows another machine advantage: it does not get tired or demoralized.