Chess AI • page 2/3 • General Chess Discussion • lichess.org

Enlightenmania2022

Well, chess was more interesting before computers, in my opinion ...
So, If I were to choose a chess-time to play chess, it would be the before-pc time .... but as Hikaru says: it is what it is!
The fact that openings are almost a closed cased due to computers, is sad, in a way ....
But I wouldn't say AI destroyed chess though.

Well, chess was more interesting before computers, in my opinion ... So, If I were to choose a chess-time to play chess, it would be the before-pc time .... but as Hikaru says: it is what it is! The fact that openings are almost a closed cased due to computers, is sad, in a way .... But I wouldn't say AI destroyed chess though.

Toscani

#12

To truly advance, Chess AI needs to introduce novelty. However, its training on curated data inherently introduces bias. While chess players learn from AI, some merely memorize lines, which isn't genuine learning. True learning involves the ability to transpose lines strategically. In contrast, memorization leaves a player stuck until an opponent deviates with a novel move.

The common AI practice of pruning moves and assuming the opponent's best response introduces another form of bias. What is considered the 'best' move is relative. If an opponent doesn't play optimally, the assumed 'best' move might not have been the most advantageous, potentially missing opportunities to exploit the opponent's suboptimal play. Even master-level players make mistakes, so engines should be designed to anticipate and capitalize on such deviations. By solely focusing on the 'best' response, engines miss numerous chances for early checkmates.

Therefore, I envision a powerful synergy in Grandmaster championships through human-AI collaboration. Engines could manage tactical calculations, operating under the assumption of optimal opponent responses. Simultaneously, humans could contribute strategic thinking, considering a wider range of possible opponent moves.

To truly advance, Chess AI needs to introduce novelty. However, its training on curated data inherently introduces bias. While chess players learn from AI, some merely memorize lines, which isn't genuine learning. True learning involves the ability to transpose lines strategically. In contrast, memorization leaves a player stuck until an opponent deviates with a novel move. The common AI practice of pruning moves and assuming the opponent's best response introduces another form of bias. What is considered the 'best' move is relative. If an opponent doesn't play optimally, the assumed 'best' move might not have been the most advantageous, potentially missing opportunities to exploit the opponent's suboptimal play. Even master-level players make mistakes, so engines should be designed to anticipate and capitalize on such deviations. By solely focusing on the 'best' response, engines miss numerous chances for early checkmates. Therefore, I envision a powerful synergy in Grandmaster championships through human-AI collaboration. Engines could manage tactical calculations, operating under the assumption of optimal opponent responses. Simultaneously, humans could contribute strategic thinking, considering a wider range of possible opponent moves.

#13

corvusmellori

#14

@Toscani said in #12:

Therefore, I envision a powerful synergy in Grandmaster championships through human-AI collaboration. Engines could manage tactical calculations, operating under the assumption of optimal opponent responses. Simultaneously, humans could contribute strategic thinking, considering a wider range of possible opponent moves.

This is done - it's sometimes called centaur chess. It's also very relevant in high level correspondence games (NOT on lichess, which does not allow engines). The world correspondence chess championship allows engine use and it is still surprising to me that the humans are at all relevant in that competition.

@Toscani said in #12: > Therefore, I envision a powerful synergy in Grandmaster championships through human-AI collaboration. Engines could manage tactical calculations, operating under the assumption of optimal opponent responses. Simultaneously, humans could contribute strategic thinking, considering a wider range of possible opponent moves. This is done - it's sometimes called centaur chess. It's also very relevant in high level correspondence games (NOT on lichess, which does not allow engines). The world correspondence chess championship allows engine use and it is still surprising to me that the humans are at all relevant in that competition.

mrbasso

#15

WCCC is pointless now since all games end in a draw and if a game is decided it's due to a wrong move transmitted accidentally.

mkubecek

#16

@mrbasso said in #15:

WCCC is pointless now since all games end in a draw and if a game is decided it's due to a wrong move transmitted accidentally.
...or when one of the players dies: https://www.iccf.com/event?id=100104

@mrbasso said in #15: > WCCC is pointless now since all games end in a draw and if a game is decided it's due to a wrong move transmitted accidentally. ...or when one of the players dies: https://www.iccf.com/event?id=100104

Toscani

edited

#17

Elevate the experience of playing chess to a new level of sophistication and enjoyment. Think of the difference between OTB, Online and AI integration. Merge them together, and build a more refined, modernised and enjoyable futuristic game experience.

Giocare Placchi Libro Aperto - Play Placchi Open Book.

Placchi would be the modern name for the change that chess gets like an doing an open book exam.

Instead of just showing legal moves, "Placchi" could subtly color-code them based on a quick AI evaluation (e.g., green for generally good, yellow for potentially risky, red for likely bad). This provides immediate feedback without explicitly stating the best move.

The player remains in control of their strategy and decision-making. Placchi only responds to the player's intended plan. This shifts the engine interaction from reactive analysis to proactive assistance for players.

Example Scenario:

Player Input: Premoves their knight to a forward attacking square, indicating a kingside attack.
"Placchi" Response: "Based on your plan to attack the kingside, here are a few potential continuations:
(Line 1 - Green/Yellow): This sequence puts immediate pressure on the f7 pawn, but your knight on [forward square] could become a target.
(Line 2 - Yellow): This develops your rook to support the attack, but it gives your opponent more time to consolidate their defense.
(Line 3 - Yellow/Red): This aggressive pawn push could open lines quickly, but it weakens your pawn structure."

Elevate the experience of playing chess to a new level of sophistication and enjoyment. Think of the difference between OTB, Online and AI integration. Merge them together, and build a more refined, modernised and enjoyable futuristic game experience. Giocare Placchi Libro Aperto - Play Placchi Open Book. Placchi would be the modern name for the change that chess gets like an doing an open book exam. Instead of just showing legal moves, "Placchi" could subtly color-code them based on a quick AI evaluation (e.g., green for generally good, yellow for potentially risky, red for likely bad). This provides immediate feedback without explicitly stating the best move. The player remains in control of their strategy and decision-making. Placchi only responds to the player's intended plan. This shifts the engine interaction from reactive analysis to proactive assistance for players. Example Scenario: Player Input: Premoves their knight to a forward attacking square, indicating a kingside attack. "Placchi" Response: "Based on your plan to attack the kingside, here are a few potential continuations: (Line 1 - Green/Yellow): This sequence puts immediate pressure on the f7 pawn, but your knight on [forward square] could become a target. (Line 2 - Yellow): This develops your rook to support the attack, but it gives your opponent more time to consolidate their defense. (Line 3 - Yellow/Red): This aggressive pawn push could open lines quickly, but it weakens your pawn structure."

petri999

#18

@corvusmellori said in #14:

This is done - it's sometimes called centaur chess. It's also very relevant in high level correspondence games (NOT on lichess, which does not allow engines). The world correspondence chess championship allows engine use and it is still surprising to me that the humans are at all relevant in that competition.
They were relevant but not anymore. has not been a while draw rate well in excess of 90% kinda tells that if you lose a game you just did not have enough computing power

@Toscani said in #12:

The common AI practice of pruning moves and assuming the opponent's best response introduces another form of bias. What is considered the 'best' move is relative.

Not really. Going for less than best and assuming that opponent makes an mistake is strategy viable only if skill levels are quite different. Between truly strong player it merely creates a weakness that can be exploited.

@corvusmellori said in #14: > This is done - it's sometimes called centaur chess. It's also very relevant in high level correspondence games (NOT on lichess, which does not allow engines). The world correspondence chess championship allows engine use and it is still surprising to me that the humans are at all relevant in that competition. They were relevant but not anymore. has not been a while draw rate well in excess of 90% kinda tells that if you lose a game you just did not have enough computing power @Toscani said in #12: > The common AI practice of pruning moves and assuming the opponent's best response introduces another form of bias. What is considered the 'best' move is relative. Not really. Going for less than best and assuming that opponent makes an mistake is strategy viable only if skill levels are quite different. Between truly strong player it merely creates a weakness that can be exploited.

Toscani

edited

#19

Thinking about the future of chess with "Placchi," the aim is to elevate the experience through deeper understanding, similar to the benefits of an "open book" exam. You're in charge of your moves, but the AI acts as a guide, showing potential consequences and encouraging exploration. The feedback is realistic, reflecting chess's complexity, rather than just pointing to a single "best" move. It's about learning and enjoying the process of strategic thought.

In an open book exam teachers don't point to the page to give you the answer. So neither should the engine.

"Placchi" would be a means of enhancing strategic thinking, not replacing it like a chess engine cheat sheet.

Thinking about the future of chess with "Placchi," the aim is to elevate the experience through deeper understanding, similar to the benefits of an "open book" exam. You're in charge of your moves, but the AI acts as a guide, showing potential consequences and encouraging exploration. The feedback is realistic, reflecting chess's complexity, rather than just pointing to a single "best" move. It's about learning and enjoying the process of strategic thought. In an open book exam teachers don't point to the page to give you the answer. So neither should the engine. "Placchi" would be a means of enhancing strategic thinking, not replacing it like a chess engine cheat sheet.

Hagredion

#20

I'm puzzled how could AI even destroy chess?