DEEPMIND, the British artificial intelligence company, has released over 200 games of AlphaZero, the awe-inspiring chess-playing entity that taught itself to play using deep reinforcement learning algorithms.

The results show that AlphaZero is much stronger than conventional chess engines, which are programmed to evaluate positions according to criteria honed by their programmers. In contrast, AlphaZero was given only the basic rules of the game and improved by playing 44 million games against itself in just nine hours and updating its neural networks with the knowledge learned from experience. AlphaZero decisively defeated the best software program, Stockfish. In the games played from the starting position where AlphaZero could select its openings, AlphaZero won +35 = 72 -3. From starting positions used in computer chess competition, the score was AlphaZero +17 = 75 -8.

It is a delight to see how aggressive – swashbuckling, one might even say – AlphaZero’s playing style is. AlphaZero seems to value mobility and activity over everything else and pays far less regard to material gain than conventional programs. In that sense, it seems far more human, which I find rather heartening. As Garry Kasparov commented: “Because AlphaZero programs itself, I would say that its style reflects the truth.”

Back in June, DeepMind co-founder Demis Hassabis, himself a former chess prodigy, allowed Dominic Lawson, Chris Flowers and myself to play AlphaZero as long as the games were kept under wraps until the recently published peer-reviewed scientific paper on AlphaZero appeared in Science. GM David Howell joined as our adviser.

We played two games and were incredibly lucky to draw the first. AlphaZero played the Berlin Defence – we went for the dull symmetrical variation and the first 17 moves followed a game of David’s against Dmitry Andreikin that we had been analysing previously. In the return we were not so lucky:

[Event "Deep Mind"]
[Site "King's Indian"]
[Date "2018.04.23"]
[Round "5"]
[White "AlphaZero"]
[Black "Pein Lawson Flowers Howell"]
[Result "1-0"]
[ECO "E73"]
[Annotator "Malcolm Pein"]
[PlyCount "80"]
[EventDate "2018.04.19"]

1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. Be2 O-O 6. Be3 (6. Nf3 c5 7. d5 e6 8.
Bf4 exd5 9. cxd5 b5 10. e5) 6… c5 7. d5 (7. dxc5 Qa5) 7… e6 (7… b5 8.
cxb5 a6 9. a4 Qa5 10. Qd2 (10. f3 axb5 11. Bxb5 Ba6 12. Bd2) (10. Bd2 axb5 11.
Bxb5 Ba6 12. Ra3 Nbd7 13. Nf3) 10… axb5 11. Bxb5 Qb4 (11… Bd7 12. Be2 Qb4
13. f3 Na6)) 8. Nf3 exd5 9. cxd5 b5 ({I take responsibility for this error. I
had seen that if acceptance of the pawn sacrifice meant that we would get some
control of dark squares but AlphaZero demosgtrated this was of no value} 9…
Bg4 {Is sounder}) (9… Ng4 10. Bf4 Qb6 11. Qb3) 10. e5 $1 (10. Bxb5 Nxe4 11.
Nxe4 Qa5+ 12. Qd2 Qxb5 13. Nxd6 Qxb2) 10… dxe5 (10… Ng4 11. Bg5 Qd7 (11…
Qa5 12. exd6 Bxc3+ 13. bxc3 Qxc3+ 14. Bd2 Qf6 15. O-O) (11… Qb6 12. exd6 b4
13. Ne4 Nd7 (13… Bxb2 14. Nfd2) 14. Nfd2) 12. Bxb5 (12. e6 fxe6 13. dxe6 Qxe6
14. Nxb5 Na6) 12… Qb7) 11. Bxc5 Re8 (11… b4 12. Bxb4 Re8 13. O-O e4 14. Ne1
Ba6 15. d6 Nc6 16. Ba3 Bxe2 17. Qxe2 Nd4 18. Qd1 Nf5) (11… e4 12. Bxf8 exf3
13. Bxf3 Qe8+ 14. Qe2 Qxf8 15. d6 Bg4 16. Bxg4 Nc6 17. O-O Nd4 18. Qd1 Nxg4 19.
Qxg4 b4 20. Nd5 Qxd6) 12. Bxb5 Nbd7 (12… Bd7 13. Be2) 13. Be3 Ng4 {To my
human eye.. so it was a surprise when AlphaZero replied} 14. O-O $1 (14. Bg5
Qb6 15. O-O e4 16. Qe2 Ba6 17. Bxa6 exf3 18. Qxf3 Nxh2 $1 {Was my idea} 19.
Kxh2 Qxa6 {Which isn’t especially good but like many of my errors, seemed like
a good idea at the time}) 14… Nxe3 15. fxe3 Qb6 (15… Rb8 16. Nd2 a6 17. Bc6
Rxb2 18. Nc4 Rb4 19. Nd6 Rf8) (15… e4 16. Nd4) 16. Kh1 $3 ({Completely
unexpected} 16. d6 Qxe3+ (16… e4 17. Nd4 Qxd6)) 16… Qxe3 17. d6 e4 18. Nd5
Qc5 19. Nc7 {Both sides have loose pieces hanging but our loose pieces are
rooks} Rf8 ({We should have gone for Dominic Lawson’s supremely insouciant
proposal, ignore all the hanging rooks, after all only one can be take on any
move!} 19… Bxb2 {Lawson} 20. Nxe8 Qxb5 21. Nc7 Qc5 22. Ng5 $3 {Is the
refutation} Qxg5 23. Qb3 $1 Bf6 (23… Ne5 24. Qxb2) 24. Nxa8 Ba6 25. Nc7 Bxf1
26. Rxf1 e3 27. Re1 Be5 28. Qxe3 Qh4 29. g3 Qb4 {Black can wriggle for a while
longer}) 20. Rc1 Qh5 21. Nd4 Qxd1 22. Rfxd1 Bxd4 (22… Rb8 23. Nc6 {is
overwhelming}) 23. Nxa8 Be3 (23… Bxb2 $1 24. Rc2 (24. Rc7 Nf6) 24… Be5 (
24… Ba3) 25. Rdc1 Bb7 26. Rc7 Bxa8 27. Rxd7) 24. Rc3 Bf4 25. Nc7 Be5 26. Rc4
Rd8 (26… e3 27. Re4 Bxb2 28. Rxe3 {is no better}) 27. Rxe4 Bb7 28. Re2 Bf4
29. Bxd7 Rxd7 30. Re7 Rxd6 31. Rxd6 Bxd6 32. Rd7 Bc5 33. Ne6 Bxg2+ 34. Kxg2
fxe6 35. Kf1 e5 36. Ke2 Bd4 37. b4 h5 38. a4 g5 39. Kf1 g4 40. Kg2 h4 1-0

You must activate JavaScript to enhance chess game visualization.

 and it was time to bend the knee … 

The post Dancing with Algorithms appeared first on Chess.co.uk.

Source: The Week in Chess