Dear Leif,
I would like to thank you very much for coming by the Club
Kasparov
website, and taking the time to register with us. By the way, if
you have
not downloaded the screensaver yet, you may do so by going to
ftp://ftp.club-kasparov.com and choosing the right file for your
operating
system (pcsaver.exe is the name of the Win95 file, and
macsaver.hqx the
Macintosh one).
As you might have read in our "About the site" segment,
we intend to make
this the best chess website in the world, to provide a
"total chess
solution" for every chess lover out there. We will take all
your input
into account as we grow the site, and I promise to keep you
informed
whenever a new major feature becomes functional. Please continue
to visit
our feedback section to let us know what you think of the site,
and what we
can do to make it better. Also, please stay tuned for the
introduction in
just a few days of our special news section.
Now allow me to move on to what you have been probably waiting
for this
past week: my special post-match report.
This was a very tough match, which demanded a lot of my energy.
It was
also a very interesting match, that captured the imagination of
millions of
people all over the world. Unfortunately, they also got to see
some errors
on my part...
I admit that I was probably too optimistic at the start of the
match. I
followed the conventional wisdom when playing computers of
playing 'ugly'
openings �non-theoretical� to avoid early confrontation, to
accumulate
positional advantages and then I was confident that my
calculation would
stay at a high level once the confrontation occurred.
My whole preparation was a failure because Deep Blue played very
differently from what I expected. My preparation was based on
some wrong
assumptions about its strategy; and when after game 2 it proved
to be a
disaster, I over-worked myself. I actually spent more energy on
the games
in this match than for any before in my life. Every game in this
match took
a lot out of me. There was enormous pressure because I had to
keep my eye
on every possibility, since I didn't want to miss any single
shot.
This is also partly why I lost this match. When Game 6 finally
came, I had
lost my fighting spirit. I simply didn't have enough energy left
to put up
a fight. At the end of Game 5 I felt completely emptied, because
I
couldn't stand facing something I didn't understand. If I had
been playing
against a human whom I knew, then it would have been different.
For
example, I was one game down against Anand in the 1995 world
championship,
but I fought back. Here, I was fighting the unknown.
What threw me off were some of the moves made by Deep Blue, which
a normal
computer would never make - machines usually don't play some of
the moves
that were made in the match. For instance, take move number 11 in
Game 5,
h7-h5.
The logic of the move is that Black wishes to gain squares for
the knights.
The threat is to play ...h4, provoking White into playing g4 so
that the f4
square
can be used after moving the knight back to g6, the bishop to d6,
and so
on. But such positional maneuvering is normally alien to
computers. Unless
there are forcing moves or captures, then lines like this are too
deep for
them. You can take any regular computer and run it for three days
and you
will see that not one comes up with h5. Such positional moves
normally have
no meaning for a computer!
Game 2 was another crucial game. As a matter of fact, I don't
think I ever
recovered after that game, and there are a few questions that I
still have
not found the answers to yet. The ones offered by the Deep Blue
team were
not exactly clarifying either...
The crucial question is, why did Deep Blue decide to take on d6
on move 35?
It spent fifteen minutes for this move, I replied automatically,
and then
in six minutes it changed its mind, and decided not to play Qb6,
the move
it intended to play. Why did it reject it? Also, how can a
machine which is
able to see 20 ply �half moves� ahead, miss a perpetual such as
the one it
allowed with its last move?
These were the kind of questions that tortured me during the
match, and not
finding answers meant not figuring what I was up against. All
this was
especially intriguing, since it followed Game 1, which as any
computer
specialist will tell you, was a typical computer game, judging
from the
moves Qa5, Bc7, h6, and g5. Deep Blue had some clear priorities:
it had to
preserve the bishop. But these priorities didn't stay constant
during the
match... In game 1 Deep Blue clearly wanted to preserve its
bishop, and it
didn't take on f3 even when it had the chance to spoil my pawn
structure,
and then it played 10...h6 and 13...g5 with the sole purpose of
saving the
bishop. But then in game 5 the same machine exchanged its bishop
on f3,
without a second thought.
As you know, the Grandmaster commentaries of the six games,
including some
of my own analysis, are posted in the Garry's Lair section of
Club
Kasparov. Nevertheless, here a few more thoughts on the
individual games:
Game 1
Deep Blue's position wasn't so bad in this game. 13...g5 was okay
for Black
- it played under its own logic, which motivated it. But then
22...g4
wasn't so good - it opened up the kingside to my advantage.
Had it not done that, then Deep Blue would have had some well
centralized
pieces. For instance, it could have made moves like Bd4, which
keep the
position under control.
Later in the game, we saw another typical computer weakness:
taking the
exchange. The computer doesn't understand positions with a
material
disadvantage. I'm sure
it was very pleased with the position, but the consequences were
too deep
for it to judge the position correctly. I think White's position
was
excellent and I am positive this was a correct sacrifice.
Interestingly enough, the machine didn't spend more than three
minutes per
move in this game. It only took longer right at the very end of
the game -
six minutes on a move - when it saw the drop in evaluation. This
game was a
pure machine performance, which is what I had been expecting
before the
match. In order to win the game I had to show some good
qualities. I played
well: I sacrificed an exchange, I maneuvered on the first rank,
then
eventually at the end I played 37. Bh3 and 39 g4 to break
through; it was
very nice. I used the machine, I exploited its weaknesses.
Unfortunately, after game 1 Deep Blue never played in the same
style again.
No positional mistakes of that magnitude, and no fixed time per
move.
Sometimes it thought for four or five minutes per move. In two
very
important games - games two and four - it spent 8 minutes and
then 15
minutes for two of its moves. Deep Blue probably sees more lines
in 15
minutes than all chess-players in the world for the rest of their
lives. I
would really like to know how the Deep Blue team was able to so
radically
alter the machine after the first game.
Game 2
I didn't play the opening very well - I was passive, because I
thought the
computer would not be able to understand and handle well the
resulting
position.
I wish I would have played otherwise, however, in order to play
'normal'
openings you have to spend a couple of months checking your
openings with
your computer. Every line. Because one mistake in the Najdorf
Sicilian
could be fatal. The level of preparation would be different.
Obviously Deep
Blue had a large team behind it studying the openings.
The rest of the game is history. Including my resignation in a
drawn
position. The truth is that I was very tired and couldn't believe
the way
the machine had just played. I trusted its calculations. I
assumed that
if the machine allowed a move such as Qe3 at the end, it had
calculated
everything that could follow and found wins, so I didn't even
bother
checking it. Costly error, as I soon found out.
Game 3
I think I tricked the machine very nicely in the opening, thanks
to 1.d3.
Unfortunately, I played 9 Ng5, instead of the better 9.b4. After
9 b4 I
thought Deep Blue would play 9...e4, and then I saw a line, but I
didn't
evaluate the consequences correctly: 10 dxe4 Bxc4 11 Nd2 Be6 12
Nd5 a5
13.b5 Ne5 14 a4, and here I thought maybe 14...Bd8 and just c6
would be ok
for black, but the good news is that I can play f4 first: 15 f4
Ng4 16 Nc4
and my position is clearly better.
If I had played 9 b4 I think I would have won the game.
I thought 9 Ng5 was strong, but I only realized later that I
couldn't get
my
kingside play going with g4. When I played 13 Nf2 I finally
understood that
Black had a fairly good position.
Game 4
This was a very good game. I don't think there is another human
who would
have saved the game. If I were playing with a human being, I
would not have
played 30...Qxe3. Instead, 30...Rf7 is a better move and my
position would
have been extremely powerful. But I wanted to exchange queens,
because
everyone was saying that this is the right way to play against a
computer... Then Deep Blue defended like a 2800 player.
Game 5
This was another game of very high quality. It could have easily
been from
a world championship match, and my opponent would have been
really proud to
save the game against me, or vice versa. I was exhausted and
upset after
game 2, but still, I am pleased to have managed to play some
outstanding
chess afterwards.
Game 6
I came to the board that day, but I didn't really want to play. I
am sorry
about my decision to play 7...h6. I simply didn't realize what I
was doing
when I played the move. It was a big mistake, and it shows the
mood I was
in. Maybe if I had gotten a nice tactical position in this game,
then
perhaps I could have fired up the gas. But it didn't happen that
way...
Clearly, I hadn't really prepared the opening. I thought it would
play
something else. It is really hard for me to explain the whole
thing...
This was not a real game. It was something that was beyond my
understanding. All I can say is that perhaps if I had one more
day of rest
between games 5 and 6, things might have gone differently.
Despite the score of this match, I am firmly convinced that this
thing is
beatable. Having said that, I don't think there are that many
players in
the world who would be able to beat it. I think only four or five
players
in the world would stand a chance against Deep Blue You need
outstanding
chess qualities to play it - you simply can't make comparisons
with other
chess computers. Take my case: I have an enormous score in
training against
the best PC programs, but it didn't help me to prepare for Deep
Blue. As a
matter of fact, I think I made a mistake in doing that. In the
future I
have to prepare specifically for Deep Blue, and play normal
chess, as well
as normal openings.
Is there a future? Yes, I think so! I just challenged IBM for a
rematch,
to take place later this year, under slightly different
conditions, such as
10 games, with one rest day between each game. Further, I want to
receive
ten practice game played by Deep Blue against a Grandmaster, as
well as the
nomination of an independent panel to supervise the match and
Deep Blue,
making sure there are no suspicions whatsoever. If this match
takes place,
and I hope it will, I am so confident I can win it, that I am
even willing
to play for a "winner takes all" prize. My score
prediction? 6-4 in my
favor!
I hope you enjoyed the match, despite my loss. I also hope you
enjoyed the
coverage that Club Kasparov provided during the days of the
challenge. The
earliest you will be able to see me in action again is June 10th,
when the
Novgorod super-tournament starts in Russia. Club Kasparov will
cover the
event, so don't forget to come by! Meanwhile, I am heading to Tel
Aviv,
where I will officially inaugurate the Kasparov International
Chess
Academy, which I will try to develop into a global center for the
creation
of special school chess courses. It is my hope that these courses
will
soon be used by schools all over the world to teach chess to
millions of
children.
Thank you for your support Leif, and keep playing chess!
Garry Kasparov