Man vs Machine World Team Championship
in Bilbao, October 6 – 9, 2004
This encounter between chess playing entities, biological
and electronic, took place in the city of Bilbao in the Basque region
of Spain. It pitted three strong grandmasters against three top programs.
The humans were Veselin Topalov (Bulgaria, Elo 2757,
world number five); Ruslan Ponomariov (Ukraine, 2710,
No. 13); and Sergey Karjakin (Ukraine, 2576, who at
12 was the youngest GM in the history of the game).
The computers were: Hydra, a hardware
machine running on a 16-processer
array that is located in Abu Dhabi, UAE; Deep Junior,
the reigning computer chess world champion, playing on a remote 4 x 2.8
GHz Xeon machine located at Intel UK (Swindon); and Fritz 8,
running on a Centrino 1.7 GHz notebook in the tournament hall. The organisers
felt that this would provide interesting comparative data, for amateurs
and press, about program performances on three different platforms.
|
|
Final standing of the Bilbao Man vs Machine
Drama in Bilbao
David Levy
Just as I was writing my previous
article for ChessBase.com, Kramnik and Leko were putting on a great show
of fireworks in Brissago (game 8). It was as though these players had read
my thoughts and decided to introduce an element of excitement into their World
Championship. Alas by game 9 and 10 it was back to “business as usual”,
with more boring draws.
The last two days of the event were held in the savings bank Bilbao
Bizkaia Kutxa (BBK)
In Bilbao, however, the 1st Man vs Machine World Team Championship continued
to provide high drama until the very last move of the event. When the final
day’s play began the computer team led the Grandmasters by 6 points to
3, and no-one expected anything other than another comfortable round for the
programs. After an hour or so, however, it appeared as though mankind might
be striking back with a vengeance. Karjakin had demonstrated a significantly
deeper knowledge than Fritz of a very sharp line of the Najdorf Variation,
and appeared to have a winning advantage when he played 18 c3. Ponomariov steadily
built up an ideal-looking position on the Black side of a Scheveningen, with
the two bishops and excellent-looking prospects for the endgame. And Topalov
was outplaying Junior, picking up a pawn before too long. Was this going to
be a 3-0 whitewash for the Grandmasters?
Sergey Karjakin (right) and Ruslan Ponomariov, who both lost unluckily
in round four
Not at all. First it was Karjakin’s turn to discover that having a winning
advantage is not sufficient against a very strong program. One must also be
able to convert such an advantage into the full point! Programs in general
are often amazingly resilient and on this day Fritz was no exception. It fought
and fought, compelling the 14-year-old Grandmaster to try to prove that Black’s
king was fatally exposed or that White could transpose to an endgame in which
his two connected passed pawns on the queen side would force a promotion. Once
or twice Karjakin faltered, finding perhaps the second best move when only
the best would do, and that was enough to allow Fritz’ king to escape.
Suddenly it was White whose king was seen to be fatally exposed and Fritz quickly
and with almost contemptuous ease turned the tables.
Ponomariov vs Hydra (operated by Chrilly Donninger) in round four
Meanwhile Ponomariov’s position also started to go downhill rapidly.
I did not have the opportunity to discuss the game with him but I suspect that,
in one line of his analysis, he had overlooked a possibility for White based
on Rxh7+, that appeared to lead to at least a draw by bringing the white queen
and the other rook to the h-file. (Programs do not make such oversights.) By
the time he realised his error, Ponomariov was in serious trouble, from which
he was unable to extricate himself, and the end was horrible to behold. This
was Hydra’s second slice of good fortune in the event, having been lost
against Topalov for much of the game in round 2.
Veselin Topalov (right) came very close to victory against Junior, operated
by Amir Ban, in round four
That left Topalov, who entered the endgame a pawn up but was unable to retain
control over the position. Junior infiltrated with a rook and even created
a menacing looking passed d-pawn before the Grandmaster decided to content
himself with a draw by repetition. What had looked, at one time, to be a 3-0
prospect for the Grandmasters, had turned into another bad day for humankind,
a dramatic 2½ - ½ disaster.
What have we learned from this crushing defeat of the Grandmaster team? I
do not believe that the 8½ - 3½ total score is a true reflection
of the relative strengths of the two teams. Instead I feel that some of this
disparity is due to insufficient preparation on the part of the Grandmasters.
Before the start of the event I had expected that all of the GMs would have
analysed two of the programs in great depth (Hydra is not available for purchase
so no advance preparation was possible for Chrilly Donninger’s monster).
But whatever the level of preparation of team GM it did not show itself to
good effect in most of the games, although Topalov appeared to have a much
better understanding of how computers play chess than did either of his team-mates.
Veselin Topalov losing his only game (to Fritz, operated by Frans Morsch)
in round three
Above all, we learned in Bilbao how difficult it is for a GM to play the best
moves often enough to win against a top program. When two GMs play each other,
each can expect his opponent to make some inaccuracies during a game, so that
one mistake will often be balanced by a subsequent mistake from the opponent.
Programs are not like that. The best programs play at a consistently high level,
so much so that, after a program leaves its openings book, only very rarely
will it make a move that gives much away. A human player, even though he might
be able to play 2750 level chess for 95% of his moves, is somewhat more fragile,
and for the remaining 5% of the time he will often play the second best or
third best move when only the best will do. It is moments like that that frequently
determine the outcome of a GM vs computer game – give a beast a single
chance and it will pounce.
A crowd of between 10 and 40 spectators watched the games
Even making allowances for inadequate preparation on the part of some of the
human players, it is clear that the computer team was significantly stronger
than the GMs. This first Man vs Machine Team Championship has indicated a level
at which we can witness a highly dramatic contest, but with the average strength
of the GMs at this level the contest will inevitably be one-sided. For the
next event in this series, the average strength of the human team will need
to be raised.
The progress of the games on a big plasma screen. In the bottom right panel
is Leontxo Garcia doing commentary for a TV production on the event.
What do the results in Bilbao tell us about the relative strengths of the
leading Chess programs? Of the individual results of the computer team, those
of Hydra and Fritz stand out. Hydra’s recent crushing victory over Shredder
and its score of 3½ out of 4 in Bilbao raises the question of what would
happen in a match between Hydra and Junior. Even with Junior on its best form,
which was certainly not the case in Bilbao, I suspect that the sheer computing
power of Hydra would give it the better chances.
The TV production facilities in a van outside the venue
Fritz also made the magnificent score of 3½, including very convincing
wins against Karjakin in round one and Topaolv in round three, and it did so
running on a laptop! As for Junior, I was reminded of something that Gligoric
had said to me many years ago. “When one plays a lot one must expect
some better than normal results and some worse than normal results.”
This is certainly true of computer programs because a program is a “living”
entity, in the sense that it is continually developing. It is not always possible
for a programming team to test their baby thoroughly, every time they make
“improvements” to the software. Sometimes it happens that an apparent
improvement gives rise to an unfortunate side-effect, one that only appears
when the program is tested in play against a very strong opponent and in particular
types of position. Junior is great program and was simply off form in Bilbao.
The public, with David Levy and TD Jaap van den Herik on the right
The Bilbao organizers have inaugurated an extremely interesting event. The
idea for this format was proposed by the International Computer Games Association
(ICGA) and appears to have been a success. We can look forward for the 2nd
World Computer Team Championship in the certain knowledge that most of the
games will be full of drama.
We welcome your opinions on the subject of man vs machine and the team
championship in Bilbao. Click on the feedback link on the left of this page
and send us your comments. Please use "Bilbao 2004" as the subject
line to facilitate the sorting of our email.
Pictures by Frederic Friedel
Links
