Thursday, June 27, 2019

Computers achieve autonomy part 2

The recently finished top-tournament Norway Chess Altibox got very mixed reviews to say the least. The local hero, worldchampion Carlsen won as the organizers hoped for but quality and quantity was below the usual standard. Some players didn't hide their intentions to bypass the classical chess immediately for the armageddons. Unfortunately many of those armageddons were poorly played containing many blunders.

If the organizers had the intent to counter the death of draws in chess then they clearly missed the goal. Besides didn't Carlsen prove in the latest months that it is still possible to win against the best players? The cure made things only worse. In correspondence-chess however things look more grim at the highest level. I already wrote about that 4 years ago in computers achieve autonomy part 1 and it only became worse since.
The 30th WC-final is still ongoing but currently the draw-percentage is 93% with 95% of the games played. A change of rules for the wc-finale is needed or it makes no sense to organize it anymore.

However as often we see that changes don't happen despite everybody is aware about the problem. Big changes in history happen mostly only when extreme situations are occurring. Also for most questions history has already a solution but we easily forget. The death of draws is nothing new. Already in 1900 so 120 years ago a solution was defined for the many draws in checkers (on 64 squares). Initially players were forced to play an opening of which the first 2 moves were selected in advance by lottery. Later from 1934 onward this was extended to the first 3 moves. As every opening must be played with both colors, nobody was favored.

This concept also exists for chess. You have the voluntary thematic-tournaments which often are organized specially for a certain festivity like I explained in my article a mini-thematic tournament but nowadays it is best known from computer-chess.  By the way the imposed openings were initially not used in computer-chess to counter the number of draws but rather to get more variety in the games. Definitely old chess-programs had the terrible habit to always play the same line see chesskids but also the brandnew Lc0 does the same see my comment at the bonusfinal from March 2019 between Lc0 and Stockfish.

Only after the superfinal of season 8 in which 89 draws out of 100 played games occurred, people realized that it is not enough to variate openings to get an interesting match. Since then more attention was given to the choice of the openings so we could see more decisive games. This is not something easy to achieve. Some complex openings for humans were easily neutralized by the best engines. On the other hand you don't want to select openings in which the win/ loss is already defined from the start. So one color should not get a too large advantage which makes the other color without a chance.

We also see that it becomes increasingly difficult to select interesting unbalanced openings. Till a couple of years ago it was sufficient just to avoid openings which were tactically refuted. Nowadays we learn by experience that strategically dubious openings often can't be used anymore for a test between the best engines. A nice example of this is what happened with the Grob : 1.g4 in the TCEC super-final of season 12. Both Komodo and Stockfish showed a large advantage for black (-0,9 till -1,46) after already the first white move and both succeeded to convert this advantage into a win.
[Event "TCEC Season 12 - Superfinal"] [Site ""] [Date "2018.06.26"] [Round "23.2"] [White "Komodo 12.1.1"] [Black "Stockfish 180614"] [Result "0-1"] [ECO "A00"] [WhiteElo "3475"] [BlackElo "3519"] [PlyCount "91"] [CurrentPosition "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"] 1. g4 d5 {(-1,46 shows Stockfish already. So to some extend the game is decided after 1 move. Komodo is as usual more moderate with only -0,90.)} 2. h3 (2. e3 {(Stockfish chose in the inverse-game for e3.)} 2... h5 3. g5 e5 4. d4 Bg4 5. f3 Bf5 6. h4 exd4 7. exd4 Bd6 8. Ne2 Ne7 9. Nbc3 O-O 10. Bf4 Nbc6 11. Qd2 Nb4 12. Bxd6 Qxd6 13. Nb5 Nxc2+ 14. Qxc2 Qb4+ 15. Qc3 Qxb5 16. Ng3 Qa4 17. Nxh5 c5 18. dxc5 d4 19. Qc4 Qd7 20. Ng3 b5 21. Qxb5 Qc7 22. Nxf5 Nxf5 23. O-O-O Rab8 24. Qd3 Qxc5+ 25. Kb1 Ne3 26. Rd2 Qa5 27. Bh3 Nd5 28. Ka1 Nf4 29. Qc2 d3 30. Qc3 Qd5 31. Bg4 Rb5 32. Qc7 Qd4 33. Qc3 Qe3 34. Rdd1 Qb6 35. g6 Rc5 36. gxf7+ Rxf7 37. Qb3 Rb5 38. Qa3 Ra5 39. Qc3 Rc7 40. Be6+ Nxe6 41. Qxd3 Nc5 42. Qd5+ Rf7 43. Qd8+ Qxd8 44. Rxd8+ Kh7 45. Rhd1 Ne6 46. R8d6 Nf8 47. R6d4 Raf5 48. Rh1 Rxf3 49. Rdd1 Rf2 50. Kb1 Ng6 51. Rd5 Kh6 52. h5 Rf1+ 53. Rxf1 Rxf1+ 54. Kc2 Rf2+ 55. Kb3 Nf4 56. Rf5 Nh3 57. Rc5 Rf7 58. Kc3 {(Stockfish - Komodo: 0 - 1)}) 2... h5 3. gxh5 e5 4. c3 Qh4 5. d3 Nc6 6. Nf3 Qxh5 7. Nbd2 Nge7 8. b4 a6 9. a4 g5 10. Rg1 f6 11. Ba3 Ng6 12. Qb1 Nf4 13. b5 Ne7 14. c4 a5 15. Qc2 Qf7 16. b6 cxb6 17. Qb2 Nc6 18. Bxf8 Kxf8 19. cxd5 Nb4 20. Nxe5 fxe5 21. Qxe5 Rg8 22. Rc1 Nfxd5 23. Rxg5 Rxg5 24. Qxg5 Bd7 25. Nc4 Bxa4 26. Kd2 Bc6 27. Ra1 a4 28. e4 Qf4+ 29. Qxf4+ Nxf4 30. Nxb6 Rd8 31. Nxa4 Nfxd3 32. Kc3 Bxe4 33. Nb6 Rd6 34. Nc4 Rd7 35. f3 Bh7 36. Bxd3 Nxd3 37. Ra8+ Kg7 38. Rb8 Nc5 39. Ne3 Rf7 40. Ng4 Rxf3+ 41. Kc4 Rf5 42. Kd4 Rh5 43. Nf2 Bg8 44. Re8 Be6 45. Re7+ Kg6 46. Rc7 0-1
Only 1 move was played and in a higher sense the game was decided already. We are again a step closer to the apocalypse of chess. However how relevant is that for us? At chesspub some members thought this new information would only be useful for the worldclass-players. I objected as I proofed it a couple of months ago on my mediocre level. In 2012 I wrote on my blog about the Czech-defense not having found a clear anti-dote. Today engines have got much stronger and are able to got much further unraveling the puzzle. This new acquired knowledge I was able to implement perfectly against maybe the biggest expert in Belgium of the Czech defense, Frederic Verduyn in our game played during the Belgian interclubs.
[Event "Interclub Deurne - KBSK"] [Site "?"] [Date "2019"] [Round "?"] [White "Brabo"] [Black "Verduyn, F."] [Result "1-0"] [ECO "B07"] [WhiteElo "2309"] [BlackElo "2150"] [PlyCount "75"] [CurrentPosition "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"] 1. e4 d6 2. d4 Nf6 3. Nc3 c6 4. Nf3 Bg4 5. h3 {(Frederic is probably the biggest fan in Belgium of this opening so obviously I was prepared for it. Previously I tried Be3 and Be2 in standard-games.)} 5... Bh5 6. Bd3 {(Frederic only prepared for the more popular Qe2 although I had written about Bd3 on my blog. Apparently there are still chessplayers not checking my blog.)} 6... e6 7. g4 Bg6 8. g5 Nfd7 9. Nh4 {(During the preparation I had discovered an interesting alternative: h4. In the end I decided to stick with Nh4 as I found the resulting lines more easy to play.)} 9... d5 10. Nxg6 hxg6 11. Bf4 {(I had prepared this novelty at home. Still known is Qg4 with also some advantage for white.)} 11... Qa5?! {(My analysis continued with Bb4 and white sacrifices the h-pawn with 0-0 just like I played in the game.)} 12. O-O dxe4 13. Nxe4 Na6 14. c3!? {(After the game Frederic recommended a3. Both are good for a white advantage.)} 14... e5?! {(This is very risky with the king still in the center. 0-0-0 is better but doesn't bring much joy either for black.)} 15. Bg3 O-O-O 16. Qg4 Kb8 17. b4 Qd5 18. Bxa6 bxa6 19. dxe5 Qe6 20. Rfd1 {(I sacrifice the h-pawn to get more activity. It is not the best but my advantage is so big that it doesn't matter anymore.)} 20... Rxh3 21. Qxe6 fxe6 22. Nd6 Nb6 23. Nf7 Rxd1+ 24. Rxd1 Be7 25. Nd8 Bxd8 26. Rxd8+ Kc7 27. Rd6 Rh5 28. Rxe6 Rxg5 29. Re7+ Kd8 30. Rxa7 Nd5 31. Rxa6 Nxc3 32. Kf1 Rf5 33. Rxc6 Ne4 34. Kg2 Nxg3 35. Kxg3 Rxe5 36. Rxg6 Re2 37. a4 Re4 38. Rg4 1-0
After the game Frederic maybe gave me the nicest compliment by confessing to me that I am the first one to let him doubt about the soundness of the opening after having hundred(s) of standard games with it played. Personally I always enjoy winning more when I can do it on my opponent's favorite territory. I also remember till today after the British Senior International Master John Anderson resigned against me (see our correspondence-game published in the article using databases part 2) that he told me, I was the first one to defeat him in his favorite line.

Now I guess some people will not agree that I call the strategic dubious openings as refuted. Can you ever talk about a refutation when a human can't formulate a clear winning-plan? Except against the best players of the world, there will always remain practical chances. On the other hand who wants to play voluntarily with such handicap in standard games?

As a surprise-weapon strategic dubious openings will still be used. However engines will get better in autonomously refuting them so I do expect in the next years a decline of their popularity even in games played between amateurs. It is also the reason why I wrote last month in the article chess position trainer part 2 not to start playing the Dutch defense.


No comments:

Post a Comment