AlphaZero has plenty of ideas in the the Schliemann. It introduces a big sharp novelty with Black (involving queenside castling and a march of the rook's pawn towards the white king!) With White it improves on a game of the Czech super-GM David Navara in the 4.d3 line which seems to put the 6...d6 line under a cloud.
Thank you GM Sadler and WIM Regan for enlightenng us in such a clear way that even a rank amateur such as myself can learn from your explanations of what is happening on the board!
The Schliemann (Jaenisch) is a really interesting opening variation! I became aware of it many years ago when a friend of mine used to play it all the time against me. Back in the pre-Internet days, I often wondered what was the best way to handle it. I think that 4) d3 seems to be the safest move. I used to try 4) Nc3 but it led to really wild positions as analyzed in an old book by Shamkovich & Schiller- black's king ends up on c5 but survives unscathed if you can believe it. I would like to point out: if white does try 4) Nc3, black does not have to play fxe, instad: 4.. Nd4! leads to very complicated play, but its exciting chess, never dull.
It's really not a bad opening at all! There have been some interesting games recently in the TCEC Season 17 Superfinal in the 4.Nc3 fe lines: www.tcec-chess.com/archive.html?season=17&div=sf&game=23 and www.tcec-chess.com/archive.html?season=17&div=sf&game=24. Seemed pretty fine for Black! I do think as well that the 4.d3 lines are best: it's much more difficult there for Black to create good counterplay. By the way, there's a also a very good book on the Schliemann by Junior Tay www.amazon.co.uk/Schliemann-Defence-Move/dp/1781943265. Best Wishes, Matthew
Yes indeed! The very best example of that is the game "Rook's Pawn Symphony" which is looked at here - th-cam.com/video/buVH99eoXCQ/w-d-xo.html - and also in "Game Changer". With Black, AlphaZero gets its rook's pawns to a3 and h2! Best Wishes, Matthew
I wonder how AlphaZero's play would change if it were trained to treat a draw as a loss. Could it convert more draws into wins in exchange for more losses? Would make AlphaZero more exciting to watch.
I noticed in the 2018 paper that AlphaZero had a tremendous score with White in the position after 3... f5. This actually surprised me a little, given Black's reasonable results in correspondence chess, as well as the relatively open and tactical nature of the position which would normally favour Stockfish. With only these two games in the downloadable file it's hard to know where this is coming from. Do you guys know if AlphaZero has discovered anything juicy in the 6... Bc5 mainline (did Stockfish ever even play 6... Bc5?), or do you attribute AlphaZero's success here to something else?
Hi Garry, thanks for your comment. I haven't seen any other Schliemann games than these I'm afraid. It wouldn't surprise me if Stockfish never tried 6...Bc5, but I don't know any more than the released games. It's funny, I never rated 4.d3 that much (a personal lapse: I had a bad experience with White when I was very young and that seems to have stuck in my head for 30 years!) but after seeing this game and reading Junior Tay's excellent book, it does seem like a good way of taking the edge off Black's attacking play. Situations in which the activity of the opponent's pieces is restricted are the ones that AlphaZero handles really well. But indeed, would have been interesting to have seen some games with 6...Bc5 as well.
@@GameChangerChessInteresting to hear. As a Schliemann player I'm certainly inclined to agree that the 4. d3 line can be a good option to reduce Black to some pretty miserly defence - the endgame that often arises after 6... Bc5 7. Qd3 d6 8. Qc4 is especially depressing, and 7... Nd4 seems downright bad for Black if White is well prepared. You make a good point about how this line would suit AlphaZero's style, and I would be interested to see if it went for 7. Qd3 instead of what I'd always regarded as the more objectively critical line in 7. Bxc6. I'm also glad to hear that it's not just patzers like me who reject opening lines simply because I've misplayed them!
Yes it's always difficult to know whether to stick to a line you lost in or play something different! Nowadays, with so much opening theory at your fingertips, I think it's worth having a bit of variety in your openings: stops you getting bored with the same stuff and keeps your opponents on their toes. In one of our previous books (Chess for Life) we looked at the Russian / Dutch player Sergei Tiviakov and how he managed to inject variety in his opening repertoire despite playing exclusively 1.e4 d5 2.ed Qd5 3.Nc3 Qd6 against 1.e4 for 7 or 8 years (against very strong opposition). It was really impressive: the number of move orders he found after 4.d4: playing 4...c6, playing 4...Nf6, 4...g6... putting the light-squared bishop on f5 or g4. From the same basic opening, he had 7 or 8 different opening schemes at his disposal. That's seems like the best approach to me (certainly for amateurs like me!) Have one basic system (which reduces the amount of theory of sidelines from the opponent you need to learn) but exploit all the variety within it. Best Wishes Matthew
These posts are brilliant. Deeply interesting. I look forward to getting a copy of the book. What I challenge though is the significance of AlphaZero with respect to the development of AI. For example, in passing Matthew Sadler (naturally) praises the quality of a particular "idea" AZ has in a position. This represents a category error,. For all the brilliant chess that has come out of AZ, it does not have any consciousness, it has no reflective ability to realise it is playing chess, and has no delight in "move ordering" its opponent. In short AZ is totally bereft of ideas. If that is a criticism - and I do not think it is anything more than a statement of fact, it does not take away from the quality of chess it tumbles out. And for the avoidance of any doubt Matthew Sadler and Natasha Reagan deserve our deep gratitude for presenting this incredible phenomenon - even if it is testament to hardware rather than software.
Yall are way ahead of the game. I have looked at Leela's self play games and found out that "she" prefers Schliemann Defense vs Ruy Lopez with Berlin Defense being her second choice. docs.google.com/spreadsheets/d/1MzFrF5Ho5sfM2Z86J0qmieH1K6FJmFBvSDjFfmm8WBM/edit#gid=0 This site shows all the self play data of openings for Leela in "her" first roughly 600 iterations. Later the Neural networks play openings like the English but if you look at total games played e4 is Leela's first choice and like Kasparov's career c5 is Leela's primary choice to defend it. I wish yall would focus on Leela vs alpha0 because alpha0 didn't self-play the sicilian as much and lost most of it's games in sicilian positions by stockfish. Leela is confirming chess tradition that e4 is "best by test" and c5 is the best response to it. Interestingly enough Leela avoids the Ruy Lopez as white and plays the 4 knights game Scotch variation. Unfortunately I don't know what "she" prefers after e4 e5 Nf3 Nc6 Nc3 Nf6 d4 The ECO codes don't differentiate between it and the Belgrade Gambit. The idea that a neural network prefers Schliemann defense to the Ruy Lopez means that humanity has a lot to learn. The Schliemann defense is one of the openings that ranks high for Leela that chess history doesn't.
Well a difficult line turned into an easy game for Black, and a clear advantage achieved in the White game... you can't ask for much more from your openings! Best Wishes, Matthew
AlphaZero has plenty of ideas in the the Schliemann. It introduces a big sharp novelty with Black (involving queenside castling and a march of the rook's pawn towards the white king!) With White it improves on a game of the Czech super-GM David Navara in the 4.d3 line which seems to put the 6...d6 line under a cloud.
Thank you GM Sadler and WIM Regan for enlightenng us in such a clear way that even a rank amateur such as myself can learn from your explanations of what is happening on the board!
Thanks Philip!
The Schliemann (Jaenisch) is a really interesting opening variation! I became aware of it many years ago when a friend of mine used to play it all the time against me. Back in the pre-Internet days, I often wondered what was the best way to handle it. I think that 4) d3 seems to be the safest move. I used to try 4) Nc3 but it led to really wild positions as analyzed in an old book by Shamkovich & Schiller- black's king ends up on c5 but survives unscathed if you can believe it. I would like to point out: if white does try 4) Nc3, black does not have to play fxe, instad: 4.. Nd4! leads to very complicated play, but its exciting chess, never dull.
It's really not a bad opening at all! There have been some interesting games recently in the TCEC Season 17 Superfinal in the 4.Nc3 fe lines: www.tcec-chess.com/archive.html?season=17&div=sf&game=23 and www.tcec-chess.com/archive.html?season=17&div=sf&game=24. Seemed pretty fine for Black! I do think as well that the 4.d3 lines are best: it's much more difficult there for Black to create good counterplay. By the way, there's a also a very good book on the Schliemann by Junior Tay www.amazon.co.uk/Schliemann-Defence-Move/dp/1781943265. Best Wishes, Matthew
God Alpha loves to push those Rook Pawns...Its like Alpha starts to laugh evilly when those guys get into motion...
Yes indeed! The very best example of that is the game "Rook's Pawn Symphony" which is looked at here - th-cam.com/video/buVH99eoXCQ/w-d-xo.html - and also in "Game Changer". With Black, AlphaZero gets its rook's pawns to a3 and h2! Best Wishes, Matthew
I wonder how AlphaZero's play would change if it were trained to treat a draw as a loss. Could it convert more draws into wins in exchange for more losses? Would make AlphaZero more exciting to watch.
Even more exciting ;-) It's an interesting idea!
Amazing explanation!
Glad you liked it!
I noticed in the 2018 paper that AlphaZero had a tremendous score with White in the position after 3... f5. This actually surprised me a little, given Black's reasonable results in correspondence chess, as well as the relatively open and tactical nature of the position which would normally favour Stockfish. With only these two games in the downloadable file it's hard to know where this is coming from. Do you guys know if AlphaZero has discovered anything juicy in the 6... Bc5 mainline (did Stockfish ever even play 6... Bc5?), or do you attribute AlphaZero's success here to something else?
Hi Garry, thanks for your comment. I haven't seen any other Schliemann games than these I'm afraid. It wouldn't surprise me if Stockfish never tried 6...Bc5, but I don't know any more than the released games. It's funny, I never rated 4.d3 that much (a personal lapse: I had a bad experience with White when I was very young and that seems to have stuck in my head for 30 years!) but after seeing this game and reading Junior Tay's excellent book, it does seem like a good way of taking the edge off Black's attacking play. Situations in which the activity of the opponent's pieces is restricted are the ones that AlphaZero handles really well. But indeed, would have been interesting to have seen some games with 6...Bc5 as well.
@@GameChangerChessInteresting to hear. As a Schliemann player I'm certainly inclined to agree that the 4. d3 line can be a good option to reduce Black to some pretty miserly defence - the endgame that often arises after 6... Bc5 7. Qd3 d6 8. Qc4 is especially depressing, and 7... Nd4 seems downright bad for Black if White is well prepared. You make a good point about how this line would suit AlphaZero's style, and I would be interested to see if it went for 7. Qd3 instead of what I'd always regarded as the more objectively critical line in 7. Bxc6.
I'm also glad to hear that it's not just patzers like me who reject opening lines simply because I've misplayed them!
Yes it's always difficult to know whether to stick to a line you lost in or play something different! Nowadays, with so much opening theory at your fingertips, I think it's worth having a bit of variety in your openings: stops you getting bored with the same stuff and keeps your opponents on their toes. In one of our previous books (Chess for Life) we looked at the Russian / Dutch player Sergei Tiviakov and how he managed to inject variety in his opening repertoire despite playing exclusively 1.e4 d5 2.ed Qd5 3.Nc3 Qd6 against 1.e4 for 7 or 8 years (against very strong opposition). It was really impressive: the number of move orders he found after 4.d4: playing 4...c6, playing 4...Nf6, 4...g6... putting the light-squared bishop on f5 or g4. From the same basic opening, he had 7 or 8 different opening schemes at his disposal. That's seems like the best approach to me (certainly for amateurs like me!) Have one basic system (which reduces the amount of theory of sidelines from the opponent you need to learn) but exploit all the variety within it. Best Wishes Matthew
These posts are brilliant. Deeply interesting. I look forward to getting a copy of the book. What I challenge though is the significance of AlphaZero with respect to the development of AI. For example, in passing Matthew Sadler (naturally) praises the quality of a particular "idea" AZ has in a position. This represents a category error,. For all the brilliant chess that has come out of AZ, it does not have any consciousness, it has no reflective ability to realise it is playing chess, and has no delight in "move ordering" its opponent. In short AZ is totally bereft of ideas. If that is a criticism - and I do not think it is anything more than a statement of fact, it does not take away from the quality of chess it tumbles out. And for the avoidance of any doubt Matthew Sadler and Natasha Reagan deserve our deep gratitude for presenting this incredible phenomenon - even if it is testament to hardware rather than software.
Depends on your definition of idea.
Yall are way ahead of the game. I have looked at Leela's self play games and found out that "she" prefers Schliemann Defense vs Ruy Lopez with Berlin Defense being her second choice. docs.google.com/spreadsheets/d/1MzFrF5Ho5sfM2Z86J0qmieH1K6FJmFBvSDjFfmm8WBM/edit#gid=0 This site shows all the self play data of openings for Leela in "her" first roughly 600 iterations. Later the Neural networks play openings like the English but if you look at total games played e4 is Leela's first choice and like Kasparov's career c5 is Leela's primary choice to defend it. I wish yall would focus on Leela vs alpha0 because alpha0 didn't self-play the sicilian as much and lost most of it's games in sicilian positions by stockfish. Leela is confirming chess tradition that e4 is "best by test" and c5 is the best response to it. Interestingly enough Leela avoids the Ruy Lopez as white and plays the 4 knights game Scotch variation. Unfortunately I don't know what "she" prefers after e4 e5 Nf3 Nc6 Nc3 Nf6 d4 The ECO codes don't differentiate between it and the Belgrade Gambit. The idea that a neural network prefers Schliemann defense to the Ruy Lopez means that humanity has a lot to learn. The Schliemann defense is one of the openings that ranks high for Leela that chess history doesn't.
Well, great novelties. Both games ended draw....
Well a difficult line turned into an easy game for Black, and a clear advantage achieved in the White game... you can't ask for much more from your openings! Best Wishes, Matthew
@@GameChangerChess Indeed but it would have been interesting to see what AlphaZero plays against ... Bc5.