Predicting Human Players' Moves

     With the long holiday weekend now over it's time to get back to chess. With my neck of the woods seeing a surge of Covid, our usually large family get together didn't happen this year and everybody had a quiet meal at home with only their immediate family attending. 

     Also, we had the last good weather until next spring. The weather forecast for today is crappy turning crappier. The day dawned with an all day drizzling rain that will be changing to a predicted 6-10 inches of snow tonight. 

This is ugly

     I had a nice post already to go, but accidentally deleted it and don't feel like trying to reconstruct it, so today I thought I would post a link to an interesting (and scary) article from a paper I just read titled Aligning Superhuman AI with Human Behavior: Chess as a Model System. Download pdf of the paper HERE.

     There is a program called Maia, a version of Leela, which has the goal of having the engine play not necessarily the best move, but human-like moves over 50 percent of the time. According to Maia's website the program is learning to predict moves made by online human players that are based on the players' ratings. 

     Predicting situations where humans err is the main goal behind this joint project of the University of Toronto, Cornell University and Microsoft Research, but it can potentially make a big difference to the statistical evidence tools to detect cheaters. 

     Engines have already destroyed correspondence chess for all but a handful of players dedicated to doing very deep research on powerful computers. I was just wondering what's in the future for the game when engines can predict a GMs move half the time?

Happy Thanksgiving!

     Wishing all readers who celebrate it a Happy Thanksgiving. If you don't celebrate the day have a great rest of the week. 

     By the way, if your are interested in the real Thanksgiving story you can read my post from last year HERE.

Was Arthur Bisguier right?

     In online games one of my favorite openings is the Four Pawns Attack against the King's Indian; in fact, sometimes I go overboard and play the Five Pawns Attack where I play g2-g4 (after h2-h3 if necessary) right in the opening. Seriously, there is no such thing as the Five Pawns Attack; I made it up. I just really like the Four Pawns Attack which was played in the following game. 

     Many times I've run into online opponents willing, even eager, to sacrifice B and N on f7 (or f2) for a R and P. Apparently their thinking is that in doing so they are playing tactically and justify the "sacrifice" because two minor pieces are worth about six Pawns and so are a Rook and a Pawn and that makes it an even trade, plus they are weakening the enemy King’s position. Is such a trade a good idea? 

     According to GM Lev Zaitsev in endgames a Rook is better especially if the side with the Rook also has a passed Pawn, but in a complicated middle game it is much easier to create an attack with the two minor pieces. 

     GM Mark Dvoretsky wrote that in the endgame a Rook is frequently stronger than two minor pieces. It happens when the Rook penetrates into the opponent's camp and wins some material or when there is an opportunity to create a passed Pawn.   

     So, generally speaking we can conclude that in the middle game two minor pieces are better and in the ending a Rook is better. 

     On the other hand Tal confessed his favorite sacrifice was to give up two minor pieces for a Rook. i.e. he preferred the Rook! He claimed this paradox is valid in an endgame especially when the Rook fights against a Bishop and a Knight that are not cooperating very well. The paradox is true in a middle game, too, providing that a Rook has an open file, or better yet, files. 

     For most of us the most accurate evaluation of the situation comes from GM Arthur Bisguier who believed that against a weaker opponent he could win with either side. Thus, the answer as to which is to be preferred can be determined by answering the question, "Am I the stronger player?" 

     Case in point...in the following game which was played online just before Christmas last year I opened with the Four Pawns Attack and we ended up in an ending where I had a R+3Ps vs. a B+N+3Ps. As it turned out, Bisguier was right...the stronger player won.

Tartajubow - Opponent

Result: 0-1

Site: Online G10

Date: 2019.12.23

King's Indian Defense, Four Pawns Attack

[...] 1.e4 d6 2.c4 ♘f6 3.♘c3 g6 4.d4 ♗g7 5.♗g5 This is commonly known as the Averbakh System; he played it with Be2. The Bisguier System is coupled with Bd3 and f4. 5...c6 6.f4 O-O

6...h6 7.♗h4 ♘h5 8.♘ge2 a6 9.h3 b5 10.g4 Love this! The game ended in a hard fought draw. Beliavsky,A (2646)-Lopez Martinez,J (2540)/Budva 2009

7.h3 Preventing the annoying ...Bg4 and toying with the idea of g4.

7.♕d2 a6 8.a4 ♕a5 9.♖a3 e5 10.fxe5 dxe5 11.d5 is equal. Sherwin,J-Panno,O/Portoroz 1958

(7.♘f3 ♘bd7 8.♗e2 ♘e8 9.O-O is even. Burkart,R-Struth,E/Germany 1993 ) 7...a6 8.a4 Hindering black's ...b5.

8.♗d3 As in the Bisguier system. 8...b5 9.e5 with equal chances.

8...♘h5 Instead of this crude threat (... Mg3) black should either develop with 8...Nbd7 or perhaps even better play 8... d5 9.♘ge2 f5 This doesn't look very good, but maybe he wanted to hinder g4.

9...♘d7 and I would have played 10.g4 but black could have at least equalized with 10...h6 11.♗h4 ♘xf4 12.♘xf4 g5 13.♗xg5 hxg5 14.♘h5

10.e5 Still thinking of playing g4, but this allows black to get slightly the better of it.

10.exf5 was correct as black is forced to recapture with the P. 10...gxf5 11.g4 and white has a promising position. 11...fxg4 12.hxg4 ♗xg4 13.♕d3 Black needs to watch out for his N. 13...♕e8 14.♗h3 ♗d7 15.♕f3 ♘f6 16.O-O-O with good attacking possibilities.

10.exf5 gxf5 11.g4 ♘f6 is correct. Then after 12.♕c2 d5 white can only claim a slight advantage.

10...dxe5 11.dxe5

11.fxe5 would be a mistake as after 11...h6 12.♗h4 c5 White's center is under attack plus his Ks position is looking a little insecure.

11...♕c7 12.♕c2 h6 13.g4 This turns out to be a mistake. (13.♗h4 g5 14.fxg5 hxg5 15.♗xg5 ♕xe5 16.O-O-O is unclear.) 13...fxg4 Fortunately for me black fails to take advantage of the opportunity to gain the advantage.

13...hxg5 14.gxh5 gxf4 15.♘xf4 ♕xe5+ 16.♘ce2 gxh5 17.♖g1 (17.♘xh5 ♗h6 leaves white in dire straits.) 17...♔h7 with the advantage.

14.♕xg6 ♗f5 A gross blunder. All he had to do was play 14...hxg5 with equal chances. 15.♕xh5 hxg5 16.hxg4 ♗d3 17.♕xg5

17.O-O-O Packs an even bigger punch...the threat is to eliminate the defender of h7. 17...♕d7 18.f5 Cutting off the B. 18...♖d8 19.♖xd3 ♕xd3 20.e6 And black is without any defense against Qf7 mate.

17...♘d7 18.♕xe7 Being three passed Ps up it seems that white has an easy win, but in fact with his K in the center black has sufficient play.

18.O-O-O would have left black helpless against threats on the h-file. 18...♗xe2 19.♗xe2 ♘xe5 (19...♘c5 20.♗d3 ♘xd3+ 21.♖xd3) 20.fxe5 ♕xe5 21.♕g6

18...♖ae8 Komodo informs us that this position is nearly equal! 19.♕h4 Renewing threats on the h-file. 19...♘c5 This does not deal with the threats adequately.

19...♘xe5 20.O-O-O Best. 20...♘g6 21.♕h7+ ♔f7 22.♕h2 ♗xe2 23.♘xe2 ♖h8 with an unclear position.

20.♗g2

20.♖d1 was better. 20...♗c2 21.♗g2 (21.♖d2 ♘d3+) 21...♗xd1 22.♗e4 ♗xe2 23.♕h7+ ♔f7 24.♗g6+ ♔e7 25.♕xg7+ and wins.

20...♖xf4 Surprise! It looked too risky to take the R, but that's what I should have done. 21.♗e4 The losing move as they say.

21.♘xf4 ♕xe5+ 22.♘fe2 and I am a R and a P ahead, but Komodo says black is better by a P and a half. How can that be?! 22...♕e3 Threatening to remove the guard from the N on e2 with ... Bxc3+ 23.♗f1 ♘b3 24.♖a3 The only safe square. 24...♘d4 with the nasty threat of ...Nf3+ 25.♖h3 A slaughter ensues. 25...♕c1+ 26.♘d1 ♗xe2 27.♔f2 ♕f4+ 28.♖af3 ♗xf3 29.♔g1 ♗xd1

21...♗xe4 (21...♗xe2 Nails it down 22.♕h7+ ♔f8 23.♔xe2 ♘xe4 24.♘xe4 ♕xe5) 22.♘xe4 ♕xe5 23.♕h7+ ♔f8 24.♘xf4 ♕xe4+

24...♕xf4 is also playable. 25.♖f1 ♖xe4+ 26.♕xe4 ♕xf1+ 27.♔xf1 ♘xe4

25.♕xe4 ♖xe4+ (25...♘xe4 is a blunder. 26.O-O) 26.♔d2

26.♘e2 leads to a really messy position after 26...♗xb2 27.♖a2 ♗f6 28.♖h5 ♖xc4

26...♖xf4 27.♖af1 ♖xf1 28.♖xf1+ Both sides have emerged from the complications unscathed and now a new phase of the game begins. 28...♔e7 29.g5 Played based on the principle that passed Ps must be pushed. 29...♘e4+ (29...♘xa4 30.b4 ♘b6 31.♖e1+ ♔f7 32.♔d3 with about equal chances.)

29...♗xb2 30.a5 ♗e5 31.g6 ♗f6 is also about equal.

31...♘b3+ 32.♔d3 is OK as long as black doesn't get greedy and grab another P. 32...♘xa5 33.♖e1

30.♔e3 Believe it or not this is a mistake. The K needs to stay close to the Ps so 31.Kd3 was correct. 30...♘xg5 This looks like it could be a draw, but it isn't. Every engine there is going back to Fritz 5.32 gives black the edge here. The R has no target so white is going to be on the defensive. 31.b4 ♘e6 32.♔d3

32.♖d1 cuts the K off but after 32...a5 33.bxa5 (33.b5 cxb5 34.cxb5 ♘c5 black is winning.) 33...♘c5 34.♖g1 ♗c3 wins because white loses his Ps.

32...♗e5 33.♖f5 ♔d6 This is a subtle mistake that should have allowed me to equalize. Simply moving the B away was correct. 34.c5+

34.♖f7 would have been considerably better. It can get tricky! 34...b6 35.a5 bxa5 36.c5+ ♘xc5+ (36...♔d5 37.♖d7+ wins) 37.bxc5+ ♔xc5 38.♖f5 is unclear, but a draw looks likely.

34.♖f7 ♘d8 is best. Then 35.c5+ ♔e6 36.♖h7 ♘f7 37.a5 with a likely draw.

34...♔d5 35.♖f7 Now with black's K able to penetrate this is not good...in fact it loses.

35.a5 should hold. 35...♘f4+ 36.♔e3 ♘g6 37.♖g5 ♘e7 38.♔d3 Cutting off black's K from going to c4 after which a draw is probable.

35...♘f4+ 36.♔c2 After this black's K penetrates and ther is no way to save the game.

36.♔e3 is no help. 36...♔c4 37.♖xb7 ♘d3 38.♖b6 ♗d4+ 39.♔d2 ♘xb4 40.♖b7 ♗xc5 wins

36...♔c4 37.♖xb7 ♘d5 38.♖a7

38.b5 Technically this is no good either, but practically it was worth a couple of exclamation marks. 38...cxb5 39.axb5 axb5 and black's b-Pawn advances. However, all is not lost! 40.♖xb5 ♔xb5 and I could have made black prove he could mate with a B and a N!. Black mates in 28. Here is how: 41.♔d3 ♘c3 42.c6 ♔xc6 43.♔e3 ♔d5 44.♔f3 ♘e4 45.♔e3 ♘g3 46.♔f3 ♔d4 47.♔f2 ♔e4 48.♔g2 ♔e3 49.♔g1 ♔f3 50.♔h2 ♗d4 51.♔h3 ♗g1 52.♔h4 ♘e4 53.♔h5 ♗d4 54.♔h4 ♗e5 55.♔h3 ♘g5+ 56.♔h4 ♔f4 57.♔h5 ♔f5 58.♔h4 ♗f4 59.♔h5 ♗g3 60.♔h6 ♘e6 61.♔h5 ♘g7+ 62.♔h6 ♔f6 63.♔h7 ♘f5 64.♔h8 ♗d6 65.♔g8 ♔g6 66.♔h8 ♗c5 67.♔g8 ♘h6+ 68.♔h8 ♗d4#

38...♘xb4+ 39.♔d2 ♔xc5 I could resign here, but we were both getting short of time. 40.♖e7 ♗d6 41.♖e4 ♘d5 42.♖g4 ♔b6 43.♖e4 ♗b4+ 44.♔c2 c5 45.♔b3 ♔a5 46.♖e5 ♘b6 47.♖e4 ♘xa4 48.♖c4 White exceeded the time limit in this lost position although it's not at all easy for black to demonstrate the win. Some engine analysis showed that the game could easily have gone on for another 40-50 moves.

Powered by Aquarium

Tal's Famous Queen Sacrifice

 

     In 1958 the World Student Chess Championship was held from 5th to 20th of July at the spa Zlatni Piasatsi (Golden Sands) near Varna, Bulgaria on the Black Sea coast. Sixteen teams, a record number, participated. 

     It was a particuarly strong event because some of the students, Tal, Filip, Panno, Olafsson and Sanguineti had played in the Interzonal at Portoroz where Tal and Olafsson qualified for the candidates' tournament. Also playing were Spassky and 8 IMs. 

     The winning Soviet team consisted of Mikhail Tal, Boris Spassky, Bukhuti Gurgenidze, Aivars Gipslis, Alexander Nikitin and Yury Nikolaevsky. 

     The US team was William Lombardy, Edmar Mednis, Anthony Saidy, Arthur Feuerstein and Robert Sobel. 

     The day before the famous Bobotsov vs. Tal game, after Tal had finished his regular game he played speed chess against Nikola Padevsky and in all the games Tal played black. The games all were the same variation in which Tal sacrificed his Q for two minor pieces with mixed results. 

     Bobotsov was among the spectators and one version of the story is that Bobotsov said to Tal that he would never risk it in a serious game. Tal's reply was that he thought it was good and the two of them agreed to play it the next day.

     According to Jiri Vesely in the book White and Black Memories, when Bobotsov and Tal met the next day the opening moves, the same variation Tal had been experimenting with against Padevsky, were blitzed out as Bobotsov was anxious to find out what Tal had prepared instead of the Q sacrifice. Without flinching Tal played the Q-sac and the surprised Bobotsov, looking a bit embarrassed, took the Q and then lost quickly.

     It's quite possible that Tal knew of the game played about a year and a half earlier in the 1956 semi-finals of the 24th USSR Championship at Kharkov between Zamikhovdsky and Nezhmetdinov. That game didn't attract much attention, but it had featured the same Q sacrifice in a slightly different position and had ended in a draw.

Milko Bobotsov - Mikhail Tal

Result: 0-1

Site: Student Team Championship Varna

Date: 1958

King's Indian: Saemisch

[...] 1.d4 ♘f6 2.c4 g6 3.♘c3 ♗g7 4.e4 d6 5.f3 O-O 6.♘ge2 c5 7.♗e3 ♘bd7 Nowadays you won't see this continuation very often because at d7 the N does not take part in the fight for the square d4 and it blocks his B. (7...♘c6 8.d5 ♘e5 9.♘g3 e6 is the modern way.) 8.♕d2 a6 9.O-O-O ♕a5 10.♔b1 b5 11.♘d5

11.dxc5 was played in Zamikhovdsky - Nezhmetdinov, 24th USSR Championshipsemi-finals, Kharkov 1956. 11...dxc5 12.♘d5 ♘xd5 13.♕xa5

13.cxd5 was tried in van der Sterren-DeJong, Wijk aan Zee 1990 which also leads to equality after either 13...Qxd2 or 13...Qc7 as in the game.

13...♘xe3 14.♖c1 Nezhmetdinov was of the opinion that white's most dangerous line was now to play 14.Rxd7, but either way the chances would be about even.

11...♘xd5 12.♕xa5 ♘xe3 13.♖c1

13.♖d3 is probably more dangerous to black. 13...♘xc4 14.♕e1 ♘db6 15.♕c1 b4 16.♖d1 ♗d7 17.♘f4 ♘xb2 18.♔xb2 ♗h6 19.♔a1 ♗a4 20.dxc5 White is better and soon won in Bu Xiangzhi (2565)-Selin,O (2327)/St Petersburg 2000

13...♘xc4 14.♖xc4

14.♕c7 is not at all good. 14...♖b8 15.dxc5 ♗xb2 16.♘c3 ♖b7 17.♕c6 ♗xc1 18.♔xc1 ♘xc5 19.♗xc4 bxc4 and black soon won. Aflalo,S (2105)-Rouffignac,T (2278)/Pau FRA 2015

14...bxc4 This is a critical position as white's next move will determine whether he stays equal or slips into an inferior position. 15.♘c1 After this passive retreat black gains the initiative.

This is the best move. After 15.♘f4 ♗xd4 16.♗xc4 ♖b8 17.b3 Black has only a slight advantage. Adnan,A (2124)-Roeder,M (2426)/Abud Dhabi 1999

15.dxc5 is definitely inferior as the following game shows: 15...♘xc5 16.♘c3 ♗e6 17.♗e2 ♖ab8 and black went on to win. Zakharchenko,A (2316)-Videki,S (2424)/Zalakaros 2003

15...♖b8 16.♗xc4 ♘b6 (was even better. 16...♗xd4 17.♘b3 ♗e5) 17.♗b3 Whether white takes the P or not black gets a dangerous position. Bobotsov probably avoided 17.Bxa6 because it gives black another open file on the Q-side. 17...♗xd4 18.♕d2 the position is going down the drain

18.♘e2 is met by 18...♗xb2 19.♔c2 (19.♔xb2 ♘c4+) 19...c4 20.♗a4 ♘xa4 21.♕xa4 ♗f6 with a winning position.

18...♗g7 This allows the c-Pawn to advance. White is now dead lost. 19.♘e2 c4 20.♗c2 c3 A nifty little finishing touch. 21.♕d3 (21.bxc3 ♘c4+ wins the Q.) 21...cxb2 22.♘d4 ♗d7 23.♖d1 ♖fc8 24.♗b3 (24.♕xa6 loses: 24...♗xd4 25.♖xd4 ♖xc2 26.♔xc2 ♘d5 27.♖d1 ♘b4+) 24...♘a4 25.♗xa4 ♗xa4 26.♘b3 ♖c3 27.♕xa6 ♗xb3 28.axb3 ♖bc8 29.♕a3 ♖c1+ 30.♖xc1 ♖xc1+ It's mate next move so Bobotsov resigned.

Powered by Aquarium