World Computer Chess Championship (the human side)
Feng-Hsiung Hsu
Edmonton, Canada. 24 teams from 9 nations participated in the event.
The human Canadian Champion and one of the final 8 in the current cycle of
Human World Championship Candidate Matches, Grandmaster Kevin Spraggett was
present commentating on the games along with IM David Levy, IM Mike Valvo,
IM Danny Kopec, IM Julio Kaplan and the co-author of Chess 4.X, David
Slate. A large crowd of about 300 were on site during most of the games.
The 1988 ACM North American Computer Chess Championship signified the
coming of the commericial micros, but this time the academic machines
had the final say. The 4th seeded Fidelity lost to both Deep Thought
and Hitech to tie for 6th in the final standing. The 5th seeded Mephisto
was upset by Waycool from Caltech, and out of the running by its round 4
draw with Novag. The 7th seeded Bebe was able to finish with a clear
2nd place ahead of both Cray Blitz (2nd seed) and Hitech (3rd seed) by
the luck of the pairing. Hitech lost the 3rd place to Cray Blitz on
tie-break, but it could have easily turned the other way.
Prior to the event, we thought that Deep Thought had less than 30% chance
of winning the title, given the small number of rounds. Lady Luck smiled
on us this time, and Deep Thought finished a full point ahead of the field.
Deep Thought was well prepared as usual: We tried to increase the number
of the processors from 2 to 6, but could not get the software debugged in
time--the hardware was delayed by manufacturing defects of the printed
circuit boards and there was simply not enough time to get the software
going. A bug that essentially adds a random number to the positional
score was located in the evaluation code after the first two games.
The program had to be recompiled and we had to pull a new set of evaluation
parameters out of thin air. DT played two of the strangest games in
its career, against Cray Blitz and Hitech in the last two rounds, possibly
as a consequence of this new set. The opening book was as brain-damaged
as ever. Ken Thompson loaned us part of Belle's book for the preparation
of the last round game to avoid book traps that might be in waiting. DT
was out of the book early in the last round nonetheless, but at least no
book trap was sprung.
The exhibition match draw between DT and Spraggett was mentioned on Candian
National TV News with the final comment ``It's scary''. However, the
commentator can be consoled by the fact that Spraggett won easily an informal
speed game match with DT after the last round championship game. From
our own observations, it appears that computers may not have that much of
an advantage in speed games. A human GM knows lots of patterns that are
too deep for today's computers to recognize in speed games.
I believe that Mike Valvo or Ken Thompson will post the full game records.
However, I will give you here the two games DT played against Cray Blitz and
Hitech, with some comments about what happened on site. Peter Jansen
provided some of his own analyses based on the log.
--Hsu
------------------------------------------------------------------------
Deep Thought - Cray Blitz (WCCC 89, round 4, 5/30/89)
1. d4 Nf6 2. Nc3 d5 3. Bg5 Bf5 4. e3
This move gets both players out of book. Cray Blitz's next move is
probably not the best, since it allows DT to obtain a significant
advantage in space.
4. ... Ne4 5. Nxe4 Bxe4 6. f3 Bf5 7. c4 c6 8. Qb3!?
DT cheerfully accepts the loss of castling right to obtain a strong
lead in development.
8. ... Qa5+ 9. Kd1
At this point, Robert Hyatt of Cray Blitz commented that this move will
either get an ! or a ?, depending on the outcome of the game.
9. ... Bc8 10. Bf4 Nd7 11. cxd5 Qxd5?
After this dubious move, black's queen gets chased all over the board.
12. Bc4 Qf5 13. g4 Qf6 14. g5 Qf5 15. Ne2 e6 16. e4 Qg6 17. h4 h6 18. gxh6
We (the operators) were looking at the possibilities after 18. h5 Qh7
19. g6! with advantage for white. DT's move prepares a piece sacrifice
which, however, turns out not to be playable.
18. ... gxh6 19. Kc2 b5 20. h5 Qf6 21. Bd3
At the last moment, DT decides not to go for 21. Bxb5 after all.
21. ... e5 22. dxe5 Nxe5 23. Qc3! Bd6 24. a3
Black's position is much more difficult than it would appear. Black's
main worry is the double pin on his Ne5, made worse by the fact that
Rh8 is hanging. Cray Blitz tries to untie this knot, but doesn't find
the best plan.
24. ... Bd7? (better 24. ... a5) 25. Bg3!
The threat 26. f4 and 27. e5 now becomes extremely annoying. The best
chance was 25. ... Bc7. After Black's next move, which locks up the rook
on a8, both black rooks become objects of attack. The game is over.
25. ... Bb8? 26. f4 Ng4 27. e5 Qe6 28. Nd4 Qd5 29. Rae1 Rg8 30. Bh7!
Forces the rook back to the long diagonal.
30. ... Rg7 31. Be4 Qc4
The 10th and last move of black's wandering queen. The exchange of
queens comes too late to help: white has a forced combination to win
a rook, after which the rest is "a matter of technique".
32. Qxc4 bxc4 33. e6! fxe6 34. Nxe6! Rg8 35. Bg6+ Rxg6 36. hxg6 Bxe6
37. Rxe6+ Kd7 (or Kf8 38. Rh4 and the Knight is lost) 38. g7 Bxf4
39. Rg6 Ne3+ 40. Kc1 Bxg3 41. Rxg3 Rg8 42. Rxe3 Rxg7 43. Rxh6 Kc7
44. Ree6 1-0
After round 4, only Deep Thought and Hitech could win the title.
If it was Hitech's strategy to get DT out of its book early, they
succeeded completely: DT was on its own after 3. Bf4. Given this
game and the game against Cray Blitz, it isn't clear if that was a
wise strategy, though.
The tension, on and off the board, was tangible, and even Spraggett
was not entirely sure what was going on. Hitech's pieces penetrated
deeply into the black position, but could not get back to the kingside
in time to prevent the debacle when DT launched a decisive attack.
The game was over rather suddenly when Hitech failed low down to mate
on move 39 and did not have time to consider alternatives. With both
programs seeing mate (and feeling no need to think long about that),
the game ended in a confused flurry of activity, to great delight (?)
of the spectators.
Hitech - Deep Thought (WCCC 89, round 5, 5/31/89)
1. d5 d5 2. Nf3 Nf6 3. Bf4 e6 4. e3 Nc6 5. Nbd2 Be7 6. h3 o-o
7. Be2 Nh5!?
This move and its follow-up are highly unusual, to say the least.
It seems that black is about to just waste two moves and/or create
weaknesses in its pawn structure, specifically on e5, where white
will be able to position an unassailable knight.
8. Bh2 g6!? 9. O-O f5 10. Be5 Nxe5 11. Nxe5 Nf6 12. c4 c5(!)
Black assures a possibility to open up the position. Neither program
was in a hurry to actually resolve the tension in the center, however,
putting the commentators for a virtually impossible task.
13. Ndf3 Bd6 14. a3 Qc7 15. Rc1
Putting "pressure" on the black queen, but DT simply refuses to worry
about it.
15. ... a5 16. Qb3 b6 17. Qa4
This is probably not such a good square for the queen, which will remain
here in almost complete passivity for the rest of the game.
17. ... Bb7 18. Rc2 Kh8 19. cxd5 (?)
This improves only black's position, who now gets his bishop on the
beautiful strong point d5. Around this time, when asked what he thinks
about the current state of computer chess, Kevin Spraggett said that
quite a few of the games actually reminded him of Grandmaster games.
Hitech-DT in particular made him think of games by Britain's top GM
Nigel Short with the Dutch defense.
19. ... Bxd5 20. Rd1 Rad8 21. Bb5 Ne4 22. Nd7 Rg8 23. Nfe5 Rg7
Most spectators thought that white was better here (as did Hitech,
evaluating the position more than 1/2 pawn to its advantage). For
some reason, neither program considered 24. f3 to be a good idea,
possibly because of the resulting weakness of e3 and g3.
In the following manuevering phase, Hitech doesn't seem to be able
to decide what to do, while DT gradually improves the position of
its pieces and sets up a kingside attack.
24. Rd3 Be7 25. Rd1 h5 26. Rdc1 Bg5 27. Re1 Bh4 28. Rf1 Be7 29. Rfc1
(This is the same position as after 26. Rdc1) g5(!) 30. f3 Nf6
31. Kf1 g4 32. hxg4 hxg4
The open h-file will prove entirely to black's advantage. In addition,
white's pawn position is already decisively weakened. Hitech can no
longer escape the loss of a pawn.
33. f4 Be4 34. Rd2 Nd5 35. Re2 Rh7 36. Ree1 Nxe3+!
A gasp was audible from the spectators when this move was played on the
demo board.
37. Kg1
Several commentators thought that Rxe3 was the best chance, but after
37. ... Rh1+ 38. Ke2 Rxc1 39. Nf7+ Kg8 40. Nxd8 Bxd8 black has too
many threats : cxd4, Qxf4, Rc2, ...
37. ... Nd5 38. Ng6+ Kg7! 39. Nxe7?
White's only hope was 39. Nde5, although then too black must be winning:
39. ... Bd6 or Bf6 followed by 40. ... Bxe5 41. Nxe5 Nxf4
(or if 40. g3 Rh1+ etc.) and black is 2 pawns up in the better
position.
At this point both programs detected mate, and began to make their
(not very human-like) moves at great speed, making at least
for some excitement for the audience.
39. ... Qxf4! 40. Nxf5 exf5 41. Rxe4 Qcx1+ 42. Bf1 fxe4 43. Qb3 Rh1+!
44. Kxh1 Qxf1+ 45. Kh2 Rh8+ 46. Qh3 g3+(!) 47. Kxg3 Qf4 mate.
After a few moments of confusion as to what had actually happened, the
roughly 200 spectators that were still around (a large number for such an
event!) burst out in applause.
--