Practical chess

School is for me already longtime ago but I suspect teaching hasn't changed much since. When a new theme is studied then first the teacher will explain the theory and next this theory is practiced via solving exercises. The same is valid for chess. If we use the method of steps then first the rules of the game or tactical patterns like double attack/ skewers/ pins... are explained before the student is asked to make exercises based on the new learning material.

The Dutch author Willy Hendriks was with his book Move First Think Later which I discussed on this blog see I knew it, one of the first writers whom successfully diverted from this established rule. He started first with the exercises and only next he gave the explanation. It is a risky concept but it worked as you didn't really need it (assuming you are already familiar with the basics) to solve the exercises. The book is however not a standard textbook but rather a guide for the teacher in which old and new methods of teaching are discussed and analyzed.

The sequel of Willy : On the Origin of Good Moves see for my review old wine in new skins part 3, uses the same approach. Also here we find the exercises before the text and again you can solve those exercises perfectly without this text assuming again you are not a novice anymore at chess. However while in the first book it still felt weird, nowadays this reversed approach is already considered fully acceptable. The great success of the first book of course convinced other writers to copy this model. It is something which I remarked again when I bought the book Think Like a Machine of the Israelian authors Noam A. Manella and Zeev Zohar.

Naturally the catchy title was too much for a computerfan. I already work since the early beginning of my chess-career with chessprograms (1990). So you don't need to convince me that engines can teach you a few things about chess. Beside nowadays a lot of new programs are created which makes it very hard to be up to date about everything. Also there are almost no good books about those developments so I thought this book could be filling a gap in the market.

On the other hand I also realized that I took a gamble with this book. The title is very provocative. I agree that we can learn from the machines but thinking like a machine is impossible for a human. A computer doesn't think but calculates per second from million of positions an evaluation. Anyway a sharp title always sells better than a soft one so you can never judge a book based on its cover.

Unfortunately sometimes you also get what is told by the title. My previous reviews were each time (very) positive but this time I can't do that. The book was for me one of the most frustrating experiences ever. I considered several times to put the book away forever while reading and I would've done that if I hadn't paid 30 euro for only 250 pages. The suffering starts immediately at the exercises so which are put before the explanations. "Happy solving" writes the author just at the start of it but then starts the horror. I spent countless hours trying to solve them but I only managed a few times to get close to the hidden ideas. What was going on? As a fidemaster I should be able to solve much more, no? Maybe it is related to the long period of inactivity due to the corona-crisis. Eventually I gave up completely disillusioned and started to check the solutions.

That was the moment when the monkey came out of the sleeve. Only a few you can solve without using an engine. Old analysis of Fischer, Kasparov, Tal, Petrosian,... published in their famous books, were destroyed by using analysis of Stockfish 10. Giri, Carlsen, Svidler, Caruana, Liren... are completely ridiculed by showing from each of them a long list of errors often made even in standard games. Further we also get a number of positions which occurred in games between Stockfish and Alphazero. To conclude we get a small collection of high level studies but which you need to solve not like in a solving competition but as you would be thinking in a tournament which is much harder as you aren't allowed to move the pieces to try something out.

The book is full of fantastic, fabulous and extremely complicated analysis created by the engine. It is a very nice collection which definitely makes it worth presented in a book-format but I don't understand why the authors present this as exercises. I am sure that even +2700 rated players will miss many things. Nonetheless the authors are convinced that each of the solutions are not too difficult for a human to find. We just need to use better our brains and think more creatively and deeper. The proof of their thesis is based on a number of splendid annotated games published in the book in which e.g. Carlsen manages despite some very complex positions to find almost each time the move recommended by the engine.

I was shocked after reading so much nonsense. Who are those authors? Why are they so critical about human players and which results did they already achieve with their revolutionary techniques? Well don't be surprised as both have no fide-rating. I only found from Zeev Zohar a chess.com account in which he has a current blitzrating of 1967.

The analysis in the book are excellent and there are some exercises which are solvable but I don't understand why a leading publisher like Quality Chess agreed to this format of the book. It would've been much better just to present it as a nice collection of positions/ games in which the engine has found some brilliant ideas and illustrate how big the gap is already between human and computer.

Anyway I didn't find any serious tips in the book to play better chess. The practical side of chess was completely neglected and that is a shame of course as I hoped to see a bridge between the computer and humans. On this blog I've written several articles about how I work with a computer but there exists no final answer about what is the best method for every player. Often a human coach is still the best adviser. A nice example of this I recently read on chesspub:  "White has the advantage but it is very hard to make the right decisions of which piece has to be exchanged when." This was last confirmed in my most recent game of the Belgian interclub covering that exact opening.

[Event "Interclub Deurne - KGSRL"] [Site "?"] [Date "2020"] [Round "?"] [White "Brabo"] [Black "Ponnet, H."] [Result "1/2-1/2"] [ECO "B06"] [WhiteElo "2296"] [BlackElo "2288"] [PlyCount "25"] [EventDate "2020"] 1. e4 g6 2. d4 Bg7 3. Nf3 d6 4. Nc3 a6 5. Be2 b5 6. O-O Bb7 7. Re1 Nd7 8. a4 b4 9. Na2 c5 10. e5 $2 {(In my notes of 2017 I had dxc5 but I couldn't remember it during the game as I hadn't prepared for this opening. At chesspub the openingsection of "1.e4 others" warned the reader that this type of position isn't easy to play for white as you need to make the right decisions when to exchange which piece.)} (10. dxc5 $1 dxc5 11. c3 Ngf6 $5 12. e5 Ne4 $5 13. cxb4 O-O $5 $16) 10... dxe5 11. dxc5 Nxc5 $2 {(Hendrik played this move quickly but this brings new problems to black. It seems black also can't find the right moves easily.)} (11... e4 $1 12. c6 Bxc6 13. Nd4 {(First capturing at b4 and only after Bb7 playing Nd4 is a possible transposition.)} 13... Bb7 14. Nxb4 Qc7 $13) 12. Qxd8+ Rxd8 13. Nxb4 {(The engine shows white is better with a pawn but I couldn't demonstrate that in the game.)} 1/2-1/2

An engine shows a huge advantage for white but can't tell you that this position isn't easy to play for humans. Fortunately we do finally see some changes. Last month Franky Nolf wrote in a reaction on my blog about DecodeChess and it is coincidence that the same tool is also presented in "Think Like a Machine" as an encouraging AI-system (also Israeli). It is just a start-product but there is definitely a lot of demand for qualitative and affordable coaching in chess.

Also more and more engines now start to work with winning-percentages instead of evaluations based on material. Leela was one of the very first engines which clearly showed the benefits of it. Others have followed. This brings me to the new Stockfish NNUE which could be soon again a major step forwards. This new engine is a mix of the old Stockfish and new networks used by Leela. I did recently a test with it and Stockfish NNUE socred 6 extra wins in 100 games compared to Stockfish 11 against the same Leela and with the same set of openings. Stockfish 11 - Leela (strongest net which I have tested) : 48 - 52; Stockfish NNUE (version of  29th of July) - Leela (strongest net which I have tested) : 51- 49.

On the other hand we should not blindly trust those winning-percentages shown by this new generation of engines. It is nice to see who has the best practical chances in an equal position but don't expect this will fully match with the results based on games played between humans. Experience is often a much bigger dominator. Also some winning-percentages shown by the engines are pretty useless especially when a win demands one side to find some very concrete moves. An engine doesn't take into account how difficult it can be for humans as is shown nicely in below game see position at move 13.

[Event "Rated Blitz game"] [Site "https://lichess.org/LVTDvPM6"] [Date "2020"] [Round "?"] [White "Brabo"] [Black "Ernesto_Aponte"] [Result "1/2-1/2"] [ECO "B03"] [WhiteElo "2398"] [BlackElo "2263"] [PlyCount "100"] [EventDate "2020"] [EventType "blitz"] [TimeControl "180"] 1. e4 Nf6 2. e5 Nd5 3. d4 d6 4. c4 Nb6 5. f4 {(An Aljechinplayer loves to play against the four-pawns-system but I play it already for 20 years so I often know the lines better than black does.)} 5... Bf5 6. Nc3 dxe5 7. fxe5 e6 8. Be3 Nc6 9. Be2 Nb4 10. Rc1 c5 11. Nf3 cxd4 12. Nxd4 Bg6 13. a3 {(Black's opening is a fiasco but as often it is not always easy especially in blitz to find the refutation. The new generation of engines like Leela work with winning-percentages but you can't fully rely on them. The winning-percentages of games played between engines are often very different from the winning percentages played between humans. This position is a nice example of it. The engine finds a clear win but this is very hard to discover for a human.)} (13. h4 h6 14. c5 Nd7 15. Qa4 Bxc5 16. Nxe6 Nc2+ 17. Rxc2 fxe6 18. Bxc5 Bxc2 19. Bd6 Bxa4 20. Bh5+ g6 21. Bxg6# {(Even with an extra hour on my clock for move 13 then still I don't think that I would be able to find this line.)}) (13. c5 Nd7 14. Qa4 $4 Bxc5 15. Nxe6 Nc2+ 16. Rxc2 $6 fxe6 17. Bxc5 Bxc2 18. Bd6 $6 Bxa4 19. Bh5+ g6 20. Bxg6+ hxg6 $19 {(So this is why white first provokes h6 with h4.)}) 13... Nc6 14. Nxc6 bxc6 15. O-O Be7 16. Qe1 O-O 17. Qg3 Bh4 18. Qf3 Bg5 19. Rfd1 Bxe3+ 20. Qxe3 Qh4 21. g3 Nxc4 22. Qc5 Qg5 23. Bxc4 Rab8 24. b4 a5 25. Kf2 axb4 26. axb4 Bh5 27. Rd6 Qxc1 28. Bd3 Qd2+ 29. Ne2 Qxb4 30. Qxc6 Rfc8 31. Qd7 Bxe2 32. Bxe2 Qc5+ 33. Kg2 Qxe5 34. Rd2 h6 35. Bf3 Rd8 36. Qxd8+ Rxd8 37. Rxd8+ Kh7 38. Rd2 g5 39. g4 Kg6 40. h3 f5 41. Re2 Qd6 42. Kf2 Qd4+ 43. Kg2 Kf6 44. Re1 f4 45. Re2 e5 46. Re4 Qd2+ 47. Re2 Qd4 48. Re4 Qd2+ 49. Re2 Qd3 50. Re4 Qd2+ {Normal} 1/2-1/2

The engine screams that the win is clear for white but I will never find without external help the h4-line and I am not yet considering the many sidelines. That is just one example but you meet such things continuously if you analyze your games with a computer.

Players always try to find practical solutions (shortcuts) for their problems (openings/ strategy/...) There are many books available which can help you with it. Some are very good but others are less useful. I believe "Think Like a Machine" belongs to the latter category. We will still have to wait till AI can give us answers.

Brabo