Recently in TIME Magazine, in a summary titled "A history of
interactivity", Kasparov's 1997 defeat by Deep Blue was named
alongside events like the Gutenberg Bible, the first photograph, the
first sound movie, and the invention of the World Wide Web.
In the chess world, we should know better.
This was the position after 91.Kxb8 in the game NewRival1816
- Faile06 in a computer tournament in 2001. According to the rule,
the game should have been stopped as a draw here, but this rule apparently
wasn't implemented in either program, and they continued to use their
brute force against each other, evaluating billions of positions per second
until, at move 493, a referee intervened and declared it a draw.
493 moves is a new world record, but there had
already been a few computer games of over 300 moves, most of them endgames
with Bishops of opposite colors between programs that did not know the
50-move rule.
This position is from Nimzo - Fritz, London 2000, the decisive game
in the playoffs of the blitz World Championship for microcomputers. There
followed: 13...h6 A mysterious move. 14.Bd3 h5 15.Nd5 O-O-O??
16.Nb4??
How is such a thing possible between computers? This
is a case of the schoolboy who prefers copying his neighbour's mistakes to
thinking of a good answer himself. Both programs used the same openings
book, with a faulted game where Black, instead of 13...h6, had really
played 13...Qc5.
In Ant - XiniX, Open Dutch Computer Championship 2000, one might
wonder how Black could have managed to lose, especially when it was his
move, and he could have mated with Qff2. However, there followed:
1...Qe6+ 2.Kf4 Qd2+ Again, not Qcf5 mate 3.Kf3 Qb3+ Not Qf5
mate. 4.Ke4 Qe2+ Not Qbd5 mate. This went on for almost thirty more
moves. Black could always mate in one move, but he never did
because the programmer had made a little mistake that made XiniX prefer a
long mate over a mate in 1. 5.Kf5 Qf7+ 6.Kg5 Qxg7+ 7.Kf4 Kb5 8.Kf5 Qg8
9.g4 hxg4 10.h5 Qge6+ 11.Kg5 g3 12.Kh4 g2 13.Kg3 g1Q+ With three
Queens, there are only more mates in 1 that he must be careful not to
give. 14.Kh4 Ka5 15.h6 Qf6+ 16.Kh3 Qf4 17.h7 Qf5+ 18.Kh4 Qd4+ 19.Kg3
Qc3+ 20.Kh4 Qfe4+ 21.Kg5 Qcd2+ 22.Kf6 Qh4+ 23.Kg7 Qe5+ 24.Kg8 Qg4+ 25.Kf8
Qf6+ 26.Ke8 Qh8+ 27.Ke7 Qhg7+ 28.Ke8 Qe3+ 29.Kd8 Kb6 30.h8Q Qed4+ 31.Ke8
Qde4+ 32.Kd8 Q4g5+ 33.Kc8 and here, poor XiniX even overstepped the
time limit: 1-0.
But the most interesting demonstration that computer chess isn't chess,
comes from a 43-year old German: Eduard Nemeth who, despite being only a
2100 player, consistently beats the world's strongest commercial chess
programs with a unique anti-computer style.
"Anti-computer style", as we know it, is to
set up a closed position, usually of the hedgehog type, and then to do
nothing, and to watch the computer also doing nothing until its lack of
time causes it to do something wrong. Nemeth's anti-computer style
consists of unsound sacrifices and the zaniest gambits you've ever seen.
One of his tactics is the classic sacrifice of a
Bishop on g5 (or g4) in exchange for an open h-file against the King.
Shredder 5 - Nemeth, 30' rapid, 2000
(NB: in all of these games, Nemeth used a P600 computer with 256 Mb RAM,
and 64-128 Mb hash tables.)
1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 f6 5.d4 d6 6.c3 h5 7.Be3 Bg4 8.O-O Be7
9.d5 b5 10.Bc2 Na7 11.a4 f5 12.h3 f4 13.Bxa7 Rxa7 14.hxg4 Probably the
losing move already. 14...hxg4 15.Nfd2 Nh6 16.axb5 Bf8 17.g3 Qg5
18.Ba4 Qh5 19.Nf3 (see diagram) 19...Nf5 20.Nh4 Nxh4
21.bxa6+ Ke7 22.Re1 Nf5 23.Kf1 f3 and mate.
Fritz 6 - Nemeth, 5' blitz, 2000
1.d4 h5 Ready to capture to g4 and open a deadly file.
2.e4 e5 Nemeth likes to humiliate his computers.
3.dxe5 Nc6 4.Nf3 Be7 5.Nc3 a6 6.Bc4 Nh6 7.O-O d6 8.exd6 Bxd6 9.Bd5 Bg4
10.Bxc6+ bxc6 11.h3 Qe7 12.hxg4 Powerless against taking the bait.
12...hxg4 13.Bg5 f6 14.e5 Qf7 15.exd6 O-O-O 16.Ne5 fxe5 17.Bxd8
(see diagram) 17...Qh5 18.f3 g3 19.Re1 Ng4 20.Kf1 Qh1+
21.Ke2 Qxg2+ 22.Kd3 Nf2+ 23.Kc4 Nxd1 24.Raxd1 Rxd8 and Black won.
Nemeth - Junior 6.0, 5' blitz, 2001
1.Nc3 d5 2.d3 e5 3.e4 d4 4.Nb1 Nc6 5. g3 Nf6 6.h4 Be7 7.Bh3 O-O 8.b3 Bxh3
9.Nxh3 h6 10.a3 a5 11.Kf1 a4 (see diagram) 12.Bg5 axb3
13.cxb3 hxg5 14.hxg5 Nb4 15.Kg2 Ne8 16.Qh5 f5 17.g6 Nf6 18. Qh4 Rf7
19.Ng5 Kf8 20.Qh8+ Ng8 21.Qxg8+ Kxg8 22.gxf7+ Kf8 23.Rh8 mate
There are other ploys to take advantage of the computer's materialism and
inability to see far away dangers.
Nemeth - Fritz 6, 5' blitz, 2000
1.e4 c5 2.c4 g6 3.Nc3 Bg7 4.h4 Nf6 5.d3 d6 6.Nh3 Nc6 7.Bg5 O-O 8.Be2 h6
9.Be3 Nd4 10.Qd2 h5 (see diagram) 11.Bh6 The right
moment; there is a bait to be taken. 11...Bxh6 12.Qxh6 Nc2+ 13.Kd2 Nxa1
14.Ng5 And Black is powerless. 14...Qb6 15.Kc1 e6 16.g4 hxg4
17.h5 Nb3+ 18.axb3 Qa5 19.Kb1 Qxc3 Finally the mate is within his
horizon. 20.bxc3 and mate in a few moves.
Nemeth - Fritz 6, 10' blitz, 2001
1.e4 c5 2.e5 Nc6 3.f4 g5 My Fritz does this too. 4.Bc4 gxf4
5.d4 cxd4 6.c3 dxc3 7.Bxf7+ Kxf7 8.Qh5+ Ke6 9.Qh3+ Kxe5 10.Bxf4+ Kxf4
(see diagram) 11.Qh5 Qa5 12.Ne2+ Ke4 13.Nbxc3+ and mate
in a few moves.
But perhaps Nemeth's greatest invention is an absurd anti-computer gambit
that I would like to call the Nemeth Gambit. Using it, he beat
five of the world's strongest chess programs in one day.
The characteristic position arises after:
1.e4 c5 2.Na3 Nc6 3.d4 cxd4 4.Bc4 Nf6 5.h3 Nxe4 6.Bxf7+ Kxf7 7.Qh5+ Ke6
8.Qg4+ Kd5 9.c4+ dxc3 10.Be3 (see diagram)
In Nemeth - Hiarcs 7.32, 10' blitz, 2001
there followed: 10...Ne5 11.O-O-O+ Nd2 12.Qe2 Qa5 13.Bxd2 cxd2+
14.Rxd2+ Ke6 15.f4 Kf6 16.fxe5+ Qxe5 17.Qf2+ Kg6 18.Nf3 Qf6 19.Nb5 Qf4
20.g3 Qc4+ (see diagram) 21.Rc2 Qxb5 22.Nh4+ Kh6 23.Rc5 Qxc5+
24.Qxc5 e6 25.Qe3+ g5 26.Rf1 Bg7 27.Nf5+ Kg6 28.Nxg7 and White won.
Nemeth - Junior 6.0, 10' blitz, 2001
went: 10...Ne5 11.O-O-O+ Nd2 12.Qe2 d6 13.Bxd2 cxd2+
14.Rxd2+ Kc6 (see diagram) 15.f4 Nd7 16.Rc2+ Nc5 17.b4 e6 18.bxc5 dxc5 19.Nf3 Qf6
20.Ne5+ Kc7 21.Qe3 a6 22.Rd1 Kb8 23.Nac4 Ka7 24.Nd6 Rb8 25.Qxc5+ Ka8
26.Nec4 Qxf4+ 27.Kb1 Qxc4 28.Qxc4 and White won.
And Nemeth - crafty 18.10, 10' blitz, 2001:
10...e5 11.Rd1+ Nd4 12.Bxd4 exd4 13.Qf5+ Kc6 14.Qxe4+ d5 15.Qxd4 Qa5
16.Rc1 Bb4 17.Nf3 Re8+ 18.Ne5+ Rxe5+ 19.Qxe5 Kb6 (see diagram)
20.O-O cxb2 21.Qxb2 Qxa3
22.Qd4+ Ka6 23.Rb1 Bf8 24.Qxd5 Qc5 25.Qd3+ b5 26.Rfc1 Qf5 27.Rc6+ Ka5
28.Qd8+ Ka4 29.Qd1+ Ka5 30.a4 Bd7 31.Rxb5+ Qxb5 32.axb5 Bxc6 33.bxc6
and White won.
Against Deep Fritz, 10' blitz, 2001 it was Nemeth who deviated
(first nine moves as above): 10.Bf4 e5 11.O-O-O+ Nd4 (see
diagram) 12.Rxd4+ Kxd4 13.Nf3+ Kd5 14.Rd1+ Kc6 15.Nxe5+ Kb6
16.Nec4+ Kc5 17.Be3+ Kc6 18.Qxe4+ d5 19.Ne5+ Kc7 20.Nb5+ Kb8 21.Rxd5 Bd6
22.Nxd6 and White won.
Finally, in Nemeth - Chess Tiger 14.0, 10' blitz, 2001, White
showed yet another form of his gambit. After 1.e4 c5 2.Na3 Nc6 3.d4
cxd4 4.Bc4 Nf6 5.f4 (instead of h3) Black again took the bait:
5...Nxe4 6.Bxf7+ Kxf7 7.Qh5+ Kf6 8.Qh4+ g5 He doesn't want the
draw - or to give back the piece. 9.fxg5+ Ke5 10.Qf4+ Kd5 11.c4+ dxc3
12.Be3 Ne5 13.O-O-O+ Nd2 14.Qd4+ Ke6 15.Qxc3 Ne4 16.Qb3+ Kf5 17.g4+ Kg6
18.Qd5 Qc7+ 19.Kb1 Nxg5 20.Nb5 Qb8 21.Bxg5 e6 22.Qe4+ Kxg5 23.Nf3+ Nxf3
24.Qxf3 h5 25.gxh5 Qe5 (see diagram) 26.Nd6 Kh6 27.Nf7+
and White won.
Considering it unfair that Rebel is always left
out of this kind of survey, I played the Nemeth Gambit against it, too.
TK - Rebel Century, 5' blitz, 28 July 2001
1.e4 c5 2.Na3 Nc6 3.d4 cxd4 4.Bc4 Nf6 5.h3 Nxe4 6.Bxf7+ Kxf7
7.Qh5+ Ke6 8.Qg4+ Kd5 9.c4+ dxc3 10.Be3 Ne5 11.O-O-O+ Nd2 12.Qe2 Ke6
13.Bd4 cxb2+ 14.Bxb2 Qc7+ 15.Kxd2 Qc5 16.Nf3 d6 17.Bxe5 dxe5 18.Ng5+ Kd7
19.Qf3 Qa5+ 20.Ke2+ Kc7 21.Rd5 Qb6 22.Rc1+ Kb8 23.Rxc8+ Kxc8 24.Qc3+ Kb8
25.Qxe5+ Kc8 26.Rc5+ Kd7 27.Qd5+ Ke8 28.Qf7+ Kd7 29.Rd5+ Kc7 30.Ne6+ Kc6
31.Qf3 Qxf2+ 32.Kxf2 and mate in a few moves.
It is possible that Nemeth (like me) went through some trial-and-error
before he won these games, but that is not the point; the point is that
it is possible to beat the strongest chess programs in the world with the
Nemeth Gambit, or with 1.d4 h5 2.e4 e5.
Of course, computers also find good moves
that a human would never think of, and they're a great help in analysis,
but what I want to demonstrate is that they don't "play chess". As the
above games show, they have passed the capturing test, but they're a long
way from passing the Turing test. In spite of all
the blustering about Kramnik who will "defend humanity's honor against the
computer" (Der Spiegel) and "gain revenge for the human race"
(The Telegraph), his upcoming match against some Fritz has, apart
from the money involved, no more significance than a match between a cat
and a book for the greatest weight.
As a defender of humanity's honor, I prefer Nemeth.
With thanks to Adolfo Bormida, Bertrand Delafargue, Gian-Carlo Pascutto and Eduard Nemeth.
PS 30 July: One reader asked me if Mr. Nemeth could do
this at regular time controls, too. Here's an example.
Nemeth - Genius 5 (P100, 15MB), 40/120', Stuttgart 1998
1.h4 e5 2.e4 Nf6 3.b4 It seems you can play just about anything
as long as you
have that deadly pawn on h4. 3...Bxb4 4.Nc3 Bxc3 5.dxc3 O-O 6.Bg5 d6
7.Nh3 h6 8.Bd3 and, Black being a computer, White is already winning.
8...hxg5 9.hxg5 Bg4 10.f3 Bxh3 11.Rxh3 Nfd7 12.f4 g6 13.Qg4 Kg7
14.O-O-O Rh8 15.Rdh1 Rxh3 16.Qxh3 Kf8 17.Bc4 Qe8 18.f5 Ke7 19.f6+ Kd8
20.Qh7 d5 21.Bxd5 c6 22.Bxf7 Qf8 23.Rd1 Kc8 24.Be6 Qd8 25.Qxg6 Kc7
26.Qf5 Qf8 27.Bxd7 and White won.
© Tim Krabbé, 2001
