A.I. we can learn from • page 1/2 • General Chess Discussion • lichess.org

We can learn a lot from A.I., Leela, Alpha0 already (more than from Stockfish) bcs AI is a little more human like and not just materialistic.

But the problem is that we can't play being 5 pawns down like Alpha0 bcs we won't find all the perfect only moves to make it work.

So I thought about: wouldn't it be very helpful to make A.I. play like a grandmaster? To make it gather experience in a way a human could? So it can show us playable ideas for humans and not for engines? To show us playable sacrifices for humans and not for engines. For example to show us motifs like the greek gift that apply in many situations and not just one. And to show us sacrifices that work out after 5 moves not after 40 moves. So we can learn in a way from Lc0 as it was the most experienced player of the world and not the most experienced engine of the world.

Also it might be cool if there were a few different styles that Lc0 could apply. For example having to play just with knights or just with bishops or having to play defensive or only positional or only aggressive.

I find it so sad that most of the effort of engines is always to have the best engine and not to help humans adapt new ideas.

We can learn a lot from A.I., Leela, Alpha0 already (more than from Stockfish) bcs AI is a little more human like and not just materialistic. But the problem is that we can't play being 5 pawns down like Alpha0 bcs we won't find all the perfect only moves to make it work. So I thought about: wouldn't it be very helpful to make A.I. play like a grandmaster? To make it gather experience in a way a human could? So it can show us playable ideas for humans and not for engines? To show us playable sacrifices for humans and not for engines. For example to show us motifs like the greek gift that apply in many situations and not just one. And to show us sacrifices that work out after 5 moves not after 40 moves. So we can learn in a way from Lc0 as it was the most experienced player of the world and not the most experienced engine of the world. Also it might be cool if there were a few different styles that Lc0 could apply. For example having to play just with knights or just with bishops or having to play defensive or only positional or only aggressive. I find it so sad that most of the effort of engines is always to have the best engine and not to help humans adapt new ideas.

nynomatica

#2

Hmmm... so interesting... just an idea what if an AI like lc0 is thaugh chess with only ones own games. Like a digital clone. I guess it will do the same moves as the player. And the make instances that predict advancement by feeding ti games of grandmaster. If there is a improvement start learning those games. That way i think an AI could be a amazing coach.

tpr edited

#3

Of course you can learn from Lc0 or AlphaZero: study their games and learn. Many grandmasters come up with concepts they first saw in games of AlphaZero / Lc0.

You can even learn from their method: playing many many games against itself, a method also practiced and recommended by Carlsen.

Of course you can learn from Lc0 or AlphaZero: study their games and learn. Many grandmasters come up with concepts they first saw in games of AlphaZero / Lc0. You can even learn from their method: playing many many games against itself, a method also practiced and recommended by Carlsen.

BlindfoldedNewbie edited

#4

I think its the opposite u may get bad habits playing the engine, I gave too much importance to Stockfish in lichess in the past, u wont find pawns races for example.

TheRider edited

#5

Good point, for newbies like me the best advice I heard was even if after the opening you're struggling to see an obvious 'line' always then look for a 'pawn push' to exploit. (Opponent under pressure is not factored in to AI)
I am talking proper 'newbies' like me

Good point, for newbies like me the best advice I heard was even if after the opening you're struggling to see an obvious 'line' always then look for a 'pawn push' to exploit. (Opponent under pressure is not factored in to AI) I am talking proper 'newbies' like me

BlindfoldedNewbie edited

#6

I understand Stockfish as a powerful machine that it is only well programmed at its top level, or the last two.
I do not remember the numbers, Stockfish does not make sense under the 2200 barrier, Its a tool to analyze, compose, check some line but it can backfire
A good analysis its an analysis of your thought process and the opponent, and thats interesting because usually
I think its the analysis of the position but chess its not that objective.
To give a tip or example, u wont practise the same kind of forks, attacking or defending them facing them..
Stockfish does not play things like sacking two minor
pieces for a rook and a pawn on f7.
In short I think its counterproductive and a bad habit of a chess player, I dont know how to put it but maybe u will lack later defensive skill or another usual skill.
It absorbs you a lot and its misleading, its misleading
to look at the evaluation and do not look at the pawn structures whether they are open or Closed.
So I do not recommend it at all, (as an opponent)
I only use it for testing, for example puzzles, if I want to create a checkmate puzzle to see if there is a hole

I understand Stockfish as a powerful machine that it is only well programmed at its top level, or the last two. I do not remember the numbers, Stockfish does not make sense under the 2200 barrier, Its a tool to analyze, compose, check some line but it can backfire A good analysis its an analysis of your thought process and the opponent, and thats interesting because usually I think its the analysis of the position but chess its not that objective. To give a tip or example, u wont practise the same kind of forks, attacking or defending them facing them.. Stockfish does not play things like sacking two minor pieces for a rook and a pawn on f7. In short I think its counterproductive and a bad habit of a chess player, I dont know how to put it but maybe u will lack later defensive skill or another usual skill. It absorbs you a lot and its misleading, its misleading to look at the evaluation and do not look at the pawn structures whether they are open or Closed. So I do not recommend it at all, (as an opponent) I only use it for testing, for example puzzles, if I want to create a checkmate puzzle to see if there is a hole

BlindfoldedNewbie

#7

I can not talk yet about the rest of the engines
but I will love to play them sporadically
Hopefully they added another one on lichess
In the offtopic section I made a question about the game of
Go, alphazero seems to have a lot of hype in the last years
and its suitable for more than game
But speaking from my experience, I can not recall exactly
how they work whether its nodes or tree with the book,
Its good the experience or for programmers but not that
much for chess, u may enjoy the technical aspect of it.
I think they focused in creating powerful machines engines
I played long time ago like 10 years ago a game with a gameboy, no with a psp playstation
I prefer they advance the middle level engines instead
of lottery blundering, not much clue anyway just Stockfish
experience but Ive seen in the internet more than 5 to download there seems to be some chess apps (that requires
another thread).

I can not talk yet about the rest of the engines but I will love to play them sporadically Hopefully they added another one on lichess In the offtopic section I made a question about the game of Go, alphazero seems to have a lot of hype in the last years and its suitable for more than game But speaking from my experience, I can not recall exactly how they work whether its nodes or tree with the book, Its good the experience or for programmers but not that much for chess, u may enjoy the technical aspect of it. I think they focused in creating powerful machines engines I played long time ago like 10 years ago a game with a gameboy, no with a psp playstation I prefer they advance the middle level engines instead of lottery blundering, not much clue anyway just Stockfish experience but Ive seen in the internet more than 5 to download there seems to be some chess apps (that requires another thread).

chessypiano

#8

@BlindfoldedNewbie I agree. But I wasn't talking about stockfish but about Lc0 actually. And Lc0 might sacrife things in an usual way. The problem is most of them cannot be imitated by humans bcs it's calculated too deeply and too precise.

I wish there was a Lc0 that would also calculate with probabilities but always choose a path that is either 1) understandable for a human or 2) doesn't need you to find many only moves in a row to play a human can't find (for example we can find forced checkmate moves but not forced positional moves)

@BlindfoldedNewbie I agree. But I wasn't talking about stockfish but about Lc0 actually. And Lc0 might sacrife things in an usual way. The problem is most of them cannot be imitated by humans bcs it's calculated too deeply and too precise. I wish there was a Lc0 that would also calculate with probabilities but always choose a path that is either 1) understandable for a human or 2) doesn't need you to find many only moves in a row to play a human can't find (for example we can find forced checkmate moves but not forced positional moves)

BlindfoldedNewbie

#9

Cool I ll try to find Lco
Reminds me of Starwars R2po

Cool I ll try to find Lco Reminds me of Starwars R2po

KING_IMPERIAL_CENTO

#10

Stockfish ( Lichess ) level 8 MAX 3000 ELO , is a wonderful engine istructor
learn errors
learn new defence
ecc ecc

this is a GREAT resource for study

Stockfish ( Lichess ) level 8 MAX 3000 ELO , is a wonderful engine istructor learn errors learn new defence ecc ecc this is a GREAT resource for study