4 ms·

I'm personally a bit more interested in computer chess, but find it very interesting how engines are used for preparation in human chess. You might think that

I'm personally a bit more interested in computer chess, but find it very interesting how engines are used for preparation in human chess. You might think that engines make opening preparation trivial: just play what the engine recommends. However, that misses a critical point. To simplify a bit, engines (whether neural network or classical engines) always play with the implicit assumption that they play against the best response of an equally strong opponent. But in human chess, the strategy is not just to make objectively good moves, but also to give your opponent a chance to make mistakes. For example, Stockfish's search doesn't distinguish between a move where the opponent has an obvious answer, and a move where you need very deep calculation to find the correct response. Mathematically, both moves are equally good. Strong human players would, of course, always prefer the line that's more "difficult" to play for the opponent. Therefore, in opening preparation, the goal is to find moves that are both: * objectively "good enough" (that is, they don't immediately lose if the opponent happens to know or find a good answer) * difficult/confusing enough for a world championship contender to give you at least the chance to gain an advantage Finding such lines with engine help is to my understanding an important part of the work of seconds. Not only does that take a lot of time, you also need to be very good yourself to make realistic judgements about what a top player might or might not find over the board.

dmurray 5y ago
> To simplify a bit, engines (whether neural network or classical engines) always play with the implicit assumption that they play against the best response of an equally strong opponent. This is a bit of a simplification! Neural net engines generally use Monte Carlo Tree Search over a pure minimax/alpha-beta search. MCTS rewards moves that lead to wins in all explored variations, not just the best play for each player. Your point is still true though: computers still aren't able to reliably find "this will be the most testing line in a practical game against a human opponent". They're too far ahead of us: the MCTS leads to better results against computers of similar strength, but doesn't accurately predict where humans will go wrong. Humans (with computer help) are still better at that.
- codeflo 5y ago
  I considered clarifying that, but it's a bit subtle. First of all, neural networks are typically trained from (millions of) self-play games. So you have the "equally strong opponent" assumption right there in the weights of the network. Second, when actually playing, the variations that are explored with high probability in MCTS are (recursively) informed by the same neural networks that are applied at the root. This means that there's no extra reward for moves that lead to complications in lines that only a "stupid human" would play. In effect, neither alpha-beta search nor Monte-Carlo tree search have any concept of "tricking" a weaker opponent, admittedly for very different reasons and more probalistically in the MCTS case. That's what I meant with "the implicit assumption that they play against the best response of an equally strong opponent" -- yes it's a simplification, but directionally true I think.
- mam4 5y ago
  What's the difference ? if you win against best play you win against less good play too no ?
  - dmurray 5y ago
    If you win, yes. But perhaps you can find two moves that both draw against best play (overwhelmingly likely in most chess positions you will prepare, though even computers aren't good enough to tell you this for certain). The important thing then is to pick the one that wins against less good play, or perhaps that you won't lose when you inevitably show some less good play.
me_me_me 5y ago
> the strategy is not just to make objectively good moves, but also to give your opponent a chance to make mistakes. That's not true in classical chess, making sup-par move in classical in order to throw off your opponent is a loosing strategy. That's why rapid and bullet formats are much more interesting, because those moves are viable. In those formats aggressive play is considered advantageous. In classical you will bounce off a hard wall defence with disadvantage. Current games are described as most accurate games in history of classical chess. Those guys are like machines in this time format.
- codeflo 5y ago
  I said "not just", not "not". It's a matter of degree. Picking a top three engine move instead of the top one, when it's only 0.01 worse, isn't a losing strategy. Even at this level, and especially if it takes a human opponent a lot of time to counter the move accurately. And from the analyses I've seen, that's precisely what seems to be happening in this world championship.
  - me_me_me 5y ago
    hmmm... I am not sure if the two concepts really square. The thing is they don't see the exal table, and their candidate moves don't eval positions in 0.1 increments. So they cannot be sure if their weaker move will result in 0.1 or 1.0 advantage down the line. Maybe you are referring to playing off book continuations of known openings? Those kind of fit your comment. Though, those moves are heavily prepared before the games and advantage there is results of pre-game strategy, rather than a tactical decision to throw off/confuse the opponent. Anyway I think we are splitting hair at this stage
    - codeflo 5y ago
      We’re talking about the same thing I think: the context was pre-game preparation, where the teams of the contenders can (and according to every top GM commentary I’ve seen, extensively do) use engines to look for “off book” but good moves that pose some problems.
pedrocr 5y ago
I've been wondering how much they automate that kind of search. Does the team of seconds now have some programmers taking Stockfish and altering it to search for lines where you always have multiple options but your opponent is on a tight path for example? That kind of work may be easier to hide than the grandmaster seconds which are drawn from a very small pool of possibilities.
- MauranKilom 5y ago
  Here is someone doing that on a non-GM level: https://medium.com/applied-data-science/how-to-calculate-the-trappiest-openings-in-chess-2bc24a6345f7 https://medium.com/applied-data-science/how-to-calculate-the...

dmurray 5y ago

> To simplify a bit, engines (whether neural network or classical engines) always play with the implicit assumption that they play against the best response of an equally strong opponent. This is a bit of a simplification! Neural net engines generally use Monte Carlo Tree Search over a pure minimax/alpha-beta search. MCTS rewards moves that lead to wins in all explored variations, not just the best play for each player. Your point is still true though: computers still aren't able to reliably find "this will be the most testing line in a practical game against a human opponent". They're too far ahead of us: the MCTS leads to better results against computers of similar strength, but doesn't accurately predict where humans will go wrong. Humans (with computer help) are still better at that.

codeflo 5y ago
I considered clarifying that, but it's a bit subtle. First of all, neural networks are typically trained from (millions of) self-play games. So you have the "equally strong opponent" assumption right there in the weights of the network. Second, when actually playing, the variations that are explored with high probability in MCTS are (recursively) informed by the same neural networks that are applied at the root. This means that there's no extra reward for moves that lead to complications in lines that only a "stupid human" would play. In effect, neither alpha-beta search nor Monte-Carlo tree search have any concept of "tricking" a weaker opponent, admittedly for very different reasons and more probalistically in the MCTS case. That's what I meant with "the implicit assumption that they play against the best response of an equally strong opponent" -- yes it's a simplification, but directionally true I think.
mam4 5y ago
What's the difference ? if you win against best play you win against less good play too no ?
- dmurray 5y ago
  If you win, yes. But perhaps you can find two moves that both draw against best play (overwhelmingly likely in most chess positions you will prepare, though even computers aren't good enough to tell you this for certain). The important thing then is to pick the one that wins against less good play, or perhaps that you won't lose when you inevitably show some less good play.

me_me_me 5y ago

> the strategy is not just to make objectively good moves, but also to give your opponent a chance to make mistakes. That's not true in classical chess, making sup-par move in classical in order to throw off your opponent is a loosing strategy. That's why rapid and bullet formats are much more interesting, because those moves are viable. In those formats aggressive play is considered advantageous. In classical you will bounce off a hard wall defence with disadvantage. Current games are described as most accurate games in history of classical chess. Those guys are like machines in this time format.

codeflo 5y ago
I said "not just", not "not". It's a matter of degree. Picking a top three engine move instead of the top one, when it's only 0.01 worse, isn't a losing strategy. Even at this level, and especially if it takes a human opponent a lot of time to counter the move accurately. And from the analyses I've seen, that's precisely what seems to be happening in this world championship.
- me_me_me 5y ago
  hmmm... I am not sure if the two concepts really square. The thing is they don't see the exal table, and their candidate moves don't eval positions in 0.1 increments. So they cannot be sure if their weaker move will result in 0.1 or 1.0 advantage down the line. Maybe you are referring to playing off book continuations of known openings? Those kind of fit your comment. Though, those moves are heavily prepared before the games and advantage there is results of pre-game strategy, rather than a tactical decision to throw off/confuse the opponent. Anyway I think we are splitting hair at this stage
  - codeflo 5y ago
    We’re talking about the same thing I think: the context was pre-game preparation, where the teams of the contenders can (and according to every top GM commentary I’ve seen, extensively do) use engines to look for “off book” but good moves that pose some problems.

pedrocr 5y ago

I've been wondering how much they automate that kind of search. Does the team of seconds now have some programmers taking Stockfish and altering it to search for lines where you always have multiple options but your opponent is on a tight path for example? That kind of work may be easier to hide than the grandmaster seconds which are drawn from a very small pool of possibilities.

MauranKilom 5y ago
Here is someone doing that on a non-GM level: https://medium.com/applied-data-science/how-to-calculate-the-trappiest-openings-in-chess-2bc24a6345f7 https://medium.com/applied-data-science/how-to-calculate-the...