[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
computer-go: proof
At 09:51 AM 9/25/00 +1200, you wrote:
>This discussion has been interesting, but I feel is not developing very well.
>
>At 05:36 PM 24/09/00 +0100, Vincent Diepeveen wrote:
>>If Kramnik has a good position, he'll win from god.
>
>I think you are probably right, but Chess is a deep enough game that the
>possibility exists that human understanding is at a lower level than we
>believe. I am a very weak chess player but i can still reasonably well
>understand the meaning behind professionals' moves. However humans cannot
>read deeply enough to really consider all the possible plans.
We can however extrapolete the performance of strong players.
if i'm a pawn up i kill anyone.
I kill computers too at blitz which are higher rated, just because
i know them.
let's directly start with proof.
Hossa (2592) vs. mid (2378) --- 2000.09.24 11:43:26
Rated blitz match, initial time: 5 minutes, increment: 3 seconds
Move Hossa mid
---- ---------------- ----------------
1. d4 (0:00) Nf6 (0:06)
2. c4 (0:00) g6 (0:02)
3. Nc3 (0:00) d5 (0:03)
4. Qb3 (0:00) dxc4 (0:52)
5. Qxc4 (0:00) Bg7 (0:07)
6. e4 (0:00) O-O (1:11)
7. Nf3 (0:00) a6 (0:02)
8. Bd3 (0:00) b5 (0:14)
9. Qb4 (0:42) Nc6 (0:11)
10. Qc5 (0:22) Bb7 (0:05)
11. Be3 (0:24) h6 (0:20)
12. Ne2 (0:30) Nd7 (0:04)
13. Qc3 (0:54) Nb6 (0:06)
14. O-O (0:11) e6 (0:29)
15. Qd2 (0:01) Kh7 (0:17)
16. Rac1 (0:39) Qd7 (0:08)
17. d5 (0:14) exd5 (0:17)
18. exd5 (0:33) Nxd5 (0:14)
19. Rfd1 (0:01) Rad8 (0:11)
20. Bc5 (0:10) Rfe8 (0:08)
21. Ng3 (0:01) Ne5 (0:21)
22. Nxe5 (0:15) Bxe5 (0:02)
23. Qc2 (0:09) c6 (0:20)
24. Be2 (0:15) Qe6 (0:03)
25. Bf3 (0:11) Nf4 (0:06)
26. b4 (0:10) Rxd1+ (0:02)
27. Rxd1 (0:05) Nd5 (0:05)
28. Re1 (0:08) Qc8 (0:06)
29. Qc1 (0:08) Bg7 (0:02)
30. Rxe8 (0:06) Qxe8 (0:01)
31. Ne4 (0:02) Bc8 (0:04)
32. Nd6 (0:05) Qe6 (0:01)
33. Qd1 (0:04) Bd7 (0:01)
34. Qd2 (0:04) Qe5 (0:02)
35. Nxf7 (0:04) Qa1+ (0:08)
36. Qd1 (0:04) Qxa2 (0:02)
37. Nd8 (0:01) Qa1 (0:03)
38. Qxa1 (0:04) Bxa1 (0:02)
39. Nf7 (0:00) Bc3 (0:05)
40. Bxd5 (0:07) cxd5 (0:00)
41. Kf1 (0:00) Bf5 (0:03)
42. Ke2 (0:07) Bc2 (0:02)
43. Bd6 (0:00) d4 (0:02)
44. Ne5 (0:00) Kg7 (0:03)
45. Bc5 (0:07) Kf6 (0:01)
46. f4 (0:00) Kf5 (0:02)
47. Kf3 (0:04) d3 (0:02)
48. g4+ (0:03) Ke6 (0:02)
49. Nxd3 (0:01) Bxd3 (0:01)
50. Ke3 (0:01) Bc2 (0:02)
51. h4 (0:01) Bf6 (0:03)
52. h5 (0:04) gxh5 (0:02)
53. gxh5 (0:01) Kf5 (0:01)
54. Bf8 (0:04) Bd1 (0:03)
55. Bxh6 (0:03) Bxh5 (0:01)
56. Bf8 (0:02) Bf7 (0:01)
57. Bd6 (0:04) Bc4 (0:01)
58. Kf3 (0:04) Bc3 (0:02)
59. Bc5 (0:03) Bd2 (0:03)
60. Bd6 (0:04) Bd5+ (0:01)
61. Ke2 (0:03) Bxf4 (0:02)
62. Be7 (0:01) Ke4 (0:02)
63. Bc5 (0:04) Bc4+ (0:01)
64. Ke1 (0:03) Kd3 (0:02)
65. Kf2 (0:03) Bd2 (0:02)
66. Kf3 (0:03) Bc3 (0:01)
67. Kf4 (0:03) Bd5 (0:02)
68. Bd6 (0:03) Kc4 (0:01)
69. Bc7 (0:02) Bxb4 (0:01)
70. Ke3 (0:02)
{White resigns} 0-1
So i get a pawn from the computer with 17.d5?
Now the computer has tremendeous pressure, and it's KNOWN
that computers play real strong when they have open files
and can attack. I have many chances to lose the game tactical
somehow, yet i chanceless win the game with a pawn up, even
though many humans will see huge advantage for white short after
the sack because of the much compensation.
So i bet the computer just because i have a single pawn up in a position
which further is dead lost. Only the fact that i can cover the piece on
d5 with the pawn on c6 means i can 'keep' the position. Something
chessplayers know. Yet usually computer wins similar positions still
because humans make errors. This was a blitz game, again favouring
computer bigtime.
Now let's extrapolate this, sure the strongest human
player with 9 stones up in go will win easily against
a perfect database.
NO PROBLEM.
>Anyway whatever we say about human vs god is unprovable. In Tictactoe or
all evidence so far points in the opposite direction. If extrapolation
is legal, just like scientists were allowed to extrapolate single
unimportant games, then consider the above example.
I'm fighting the computer on a terrain where it is worth way more as
2600, more like 3000+ in blitz at that type of positions. It is very
hard for it to make errors. In fact it did NOT make errors after d5?
move. Yet this single speculative move lost the game.
>perhaps even draughts we can be sure about what perfect play looks like,
>but in more complicated games we are only guessing. The professionals have
>developed a set of heuristics that work pretty well against other humans.
it's not about playing perfect. it's about beating the best.
i still think the discussion is done wrong here. it's not interesting
to play perfect. it's interesting to win always from the strongest
player. that's 750 rating points more as the best human at the time
you play.
>>Will a professional go player *ever* lose when he starts with 9
>>stones up and playing for his life?
>
>The basic answer is that we don't know. However it seems unlikely.
all extrapolation obviously says the opposite. We know it. the human wins.
the advantage is just TOO large.
>>ratings in chess don't compare very well to GO. Apart from that
>>there are many ratings in chess too.
>
>Ratings in go vary dramatically according to the ability of the players too.
We're not busy with amateurs like me, not to mention my pathetic go
level.
>>In a post on the rating list one says 100 rating points are worth
>>a stone.
>
>This is a very rough approximation and certainly doesn't apply among
>professionals. It might be true that 100 rating points correspond to one
>rank difference, but about 3-4 rank differences make up one stone among
>professionals.
Oh kasparov is very easy to rate. 9 dan :)
>>If mr. sugimoto with 2030 plays Kasparov who has:
>>
>> 4100018 Kasparov, Gary g RUS 2849 35 13.04.63
>>
>>So if 100 points would be 1 stone, then Kasparov would play with 8 stones
>>behind against any 2049- rated person on the above list.
>As above If Kasparov is 9 dan then Sugimoto would be about 1 dan and would
>need about 3 stones handicap (which seems about right).
Is 3 stones not a bit much for professional players as a difference?
>>Kasparov is 9 dan professional, Kramnik too.
>>Then big gap and a lot of 8 dans on the list.
>
>I guess you could relate Chess pro grades to go by saying that Kasparov is
>9 dan, The weakest GM would be about 5 dan, and The weakest IM would be
>about 1 dan professionals.
>If a pro 9 dan plays a pro 5 dan then the 9 dan wins more than 90%. 2
>stones handicap tips the balance slightly in black's favour.
>A 1 dan beats a 9 dan about 1 game per decade. The difference in handicap
>stones is about 3.
>We can assume that the top Chess players and the top Go players are
>approximately equal in how good they are at the game. It doesn't follow
>that they are equally close to perfect play.
i doubt they're equally strong. Most go players don't even annotate their
game. In chess everything gets annotated and analysed. Kasparov
is probably better in chess as any go player ever was in go.
That's not because kasparov itself is better, but basically also because
he had better training, and better examples and always had all games of
his opponents to prepare against their style.
Apart from that kasparov's real strong weapen which puts him
at this high rating is the fact that he has a superb openings advantage
always, which is quite hard in go, as there are so many possible
openings in go, where in chess one CAN analyze ones opponents openings
play in depth!
So normally spoken kramnik wipes out kasparov everywhere. over the board
kramnik is a level better as kasparov for sure. Yet kasparov combines
his high level play with a superb openingspreparement, something
completely impossible in go (as far as i know).
>Barry Phease
>mailto:barryp@xxxxxxxxxxxxxxxxx
>http://www.es.co.nz/~barryp