Full conversation with Magnus Carlsen: ua-cam.com/video/0ZO28NtkwwQ/v-deo.html Quick note from Lex: The camera on Magnus died 20 minutes in. Most folks still just listen to audio-only version, but here on UA-cam we did our best to still make it interesting to watch & listen by adding image overlays. I mess things up sometimes, like in this case, and it hits me hard when I do. I'm sorry for this. I'm always working hard to improve. I hope you understand. Thank you for your patience and support along the way. I love you all.
What if there is superior chess engine in deep subconcious, and all that blunders are actually brilliant? They don't work, because our concious mind freak out when we lost a pawn and "screw you guys, I'm going home" after loosing queen?
Not any human, tho. It will assign a value to each one, and will sacrifice pawns to protect the overall plan much more willingly than rooks, queens, or even bishops.
Originally, the queen moved like the king. Chess underwent gradual gameplay adjustments over centuries, whereas StarCraft's underwent rapid play testing due to being a commercial product. The most modern chess innovation has come with online play, where time pressure is needed as a counter to cheaters. (Chess clocks were merely a means to prevent games from going too long.)
If there was no bishop or knight, there would just be pawns, rooks and queens. Rooks and queens move in a pretty simplistic way. The geometric variety of chess comes from the knights and bishops. The knights go in an L and its path can't be obstructed by another piece. Bishops are really interesting because they are each constrained to one color.
Knights are unique pieces. I love the spherical pattern that knights cover in space as well. Once you get two of them into the middle the can smother a lot of play together with good pawn usage.
What Magnus said about shogi is true, but only until you exchange a bunch of pieces, because shogi is like "bughouse" - you can return the pieces you captured from your opponent back into play and then it suddenly becomes clear why they are so weak. You can drop them back on almost any free square on the board so this just balances it out beautifully as opposed to how it would be if they were also really powerful on top of this. I highly recommend that any chess fan try learning shogi. It's an amazing game in its own right.
@@MaLLinz289 I am not sure if bughouse was directly inspired by shogi or not, but this mechanic had a few centuries of evolution in shogi so it works much much better in it than it does in bughouse. Chess pieces are just too powerful for this mechanic and bughouse is just someone's little experiment with an already well formed game. I wholeheartedly recommend that you give shogi a shot. It's similar enough to chess for chess-fans to enjoy, but it's different enough that it gives a very different experience when playing.
Even as a casual player such as myself, I've always been taken aback by how much ego plays a role in chess games, especially when it comes to losing pieces. Interesting how AlphaZero highlights this.
There is ego in chess but not about losing pieces specifically. Every beginner approaches the game with caution and is afraid of losing material balance, as, often, having less material leads to losing. Once you get more confident, you become less materialistic and start to see the profitable sacrifices and stop responding to every capture threat and look for a higher threat as a solution. Doesn't matter if you lose your queen if the move you make means mate in 2 or 3 moves...
I think the point of this video is the opposite from the point you got. Humans don't hold onto pieces for ego, but a lack of calculating ability (finding eventual compensation). The egotistical thing to do would be to make sacrifices. Chess players get boners over bold sacrifices.
It isn't ego, its inability to play many lines many moves ahead and see how efficient a sacrifice is. I would also group castling.....when humans play they pretty much always castle whereas bots often dont castle and still win
I think theres a large fear aspect in castling. Humans seeing the indeterminate future think that castling can only be a good thing because of their fear of a decisive embarrassing loss due to the center opening up. The computer is unafraid of this and plays for the advantage regardless of position. I wouldnt say that humans castle because theyre incapable of calculating. Its more of a misunderstanding of the game and an aversion to risk embarrassment
For the note on balance, there might be asymmetry between pieces, but because each player gets the same set, balancing each piece is not as big of a deal as balancing factions like in Starcraft. And in terms of design, it still is hundreds or thousands of hours of playtesting that tell you about balance that chess got naturally. Although because of the turn-based nature of chess, no piece can be super overpowering or the first-player advantage would skyrocket. It would be interesting to see how much chess strategy would change if the rules were adjusted so that each player could have the same number of turns regardless (e.g., if mated, be allowed to play one more move and if they can also mate it would be a draw)
Totally agree.....starcraft balance is very complex because there are 3 races and any change will impact say pvt differently than pvz; and now the game is mature enough and blizzard has not patched broodwar in 20 years so balance is done through maps not tweeks to the game. But with chess the biggest "balance" due to turn based vs real time. I guess balance would mean under perfect play it is a draw. But even that may not be practical balance since maybe quantum computers with near infinite elo (or even quantum computer who has solved chess) may be a draw (or even black win) but in human play white has advantage. Similar problem to starcraft where in the very top terran is arguably OP but at basically any level except top 10-20 players in the world protoss is consensus OP
9 dan professional go player Micheal Redmond has mentioned that AlphaGO will make larger sacrifices and trades than humans. In Go at least part of this may be that humans have future plans for pieces when played, and attachments to groups that influence their play. The ability to change plans and not being attached to groups is a major difference between weaker and stronger human players also.
@@mattskov2917 Thanks. I was just watching another Micheal Redmond review of an AlphaGO vs AlphaGO game just now. Some other points about AlphaGo vs human differences are, people have memories of why they played moves and previous calculation of positions in the game being played. AlphaGo has no memory of the game it is playing, it calculates its next move only from the current state of the board. AlphaGo does have a "memory" of every previous game it played but not the current game.
I think the thing about the "tension between the bishop and the knight" is that they're opposites in many ways: each bishop can control only one color, while a knight can control both, but which color switches whenever it moves; a bishop is a long-range piece that can be blunted by pawns along a diagonal, while a knight is a short-range piece that's often used to blockade pawns while still controlling squares beyond it; bishops prefer open boards with fewer pawns on it, while knights prefer closed boards with more pawns on it. The reason bishops are usually a bit more valuable is because most games end up (either soon or eventually) as being more open, so bishops are able to control more squares more quickly, but they're really close in value and that miniscule difference only matters at super-grandmaster level where they have to exploit every tiny edge.
That makes no sense. Pieces and pawns aren't given away for nothing. The point of the sacrifice isn't that the piece being sacrificed is useless. What is usually meant by a sacrifice is an exchange of material where one side loses on point count---a rook (5 points) for a knight (3.25 points), say. The sacrificing player reasons that he lost the exchange on points but gained space or time or an attack or long-term positional advantage.
@@josephpeeler5434 ultimately it's about checkmate,so if we accomplish it ny giving away as many pieces in a good combination,that is then a good sacrifice
@@sunset1394 But it isn't about "giving away" pieces. The sacrifice might be a minor for 2 pawns or a rook for a bishop. If there is a net point loss in the exchange sac, then the player sacrificing must get either space, time, attack or positional advantage in return. The sac is an exchange. It is not a give away.
For anybody interested, "Game Changer: Alphazero's Groundbreaking Chess Strategies and the Promise of AI" is an excellent read on the subject. It's from 2019 but was when AZ really tore into SF. It offers good insights and walks through some of the immortal games in that series, as well as explaining some of the novelties behind AZ.
I think what people are more interested in is sacrifices that give you sure shot advantages (at least with our present knowledge of chess). Tal's sacrifices, most of them were not like that. They are still great though because to do better would be extremely difficult for humans.
Especially coming from someone out of their depth on a subject. It would have been interesting to have someone like Ilya Sutskever on the opposite end of the table drill down into that response from Magnus.
If it comes from calculation, it's not really creative, is it? I do think a neural net can be creative if the act of the computer coming up with the move is not calculation based. If stockfish has just brute forced all the possibilities from saccing a piece and that gives a winning advantage, it's not really creativity IMO. It's the outcome of an optimization heuristic. The original programmers were creative though.
@@AbcXyz-rn2lzI guess because of his controversial statements, though I'm sure he's not the only great chess player in history to have said or believed unpopular/unusual things.
3) About the queen in chess: Salomon, an humain, also an former chess champion in my nowadays country, said that I lose half of my strength in chess when the queen are not on the chessboard :-)
The 'knight vs bishop' is an big aspect in chess, but it's far from the basis of the game being fun. Just look at all of the variety in end games that don't feature knights or bishops.
No, that's easy. Creativity (in the non-artistic sense) involves generating new explanations and creating new knowledge from those explanations. AlphaGo and AlphaZero don't generate any explanations, they simply optimize for winning a game with simple, known rules. That's why no chess or Go players have gotten better from playing/studying these systems (actually, some have gotten worse). When an AI can create new explanations that actually add to our knowledge, then we can consider it to be exhibiting the kind of creativity we actually care about.
@@vinesthemonkey it isn't - there have been just as many players that have decreased their ELO after playing AIs than have improved - no better than chance. But regardless, it still isn't the kind of creativity we actually care about, even if you think these computational statistics programs are creative. Creating new explanations is what matters.
I feel like Magnus is missing the point of weak pieces in Shogi. The power, complexity, and strategy comes from dropping pieces which the game is balanced around. Being able to put a captured weaker piece anywhere you want is more "powerful" (and interesting) than a queen in my opinion.
It's very interesting comment on the appeal of chess and he's probably right, however on the overall strategy, perhaps Shogi is superior as you suggest? I think I agree with both those considerations.
@@inafridge8573 In my experience each game tends to go a certain way in close games (broadly speaking of course) Chess: much faster arrival to middle game with good excitement that cools down into a drawish endgame with fewer and fewer pieces. Shogi: very slow maneuvering opening, but once middle game is in full swing it's usually a very sharp race to mate. This is due to clearing space but material always increasing after trades. So it's kind of picking where you want the "explosiveness" to be concentrated.
The notion that bishops and knights are relatively equal is true. It is also true that, in any given position, the value of all pieces on the board are completely fluid. There’s certainly a subset of positions where a bishop or rook could be evaluated as being more valuable than a queen.
That is almost impossible. As in, if you were to replace a given queen with a rook or bishop, the only way your position is better is if you were about to have some sort of exotic stalemate
5:47 - Chess is always a "mirror matchup", both sides get the exact same units. Balancing a mirror matchup is much simpler than when the sides have different pieces. Starcraft Brood War was released in 1999 and is still extremely balanced between the three races. But the balance is mostly achieved by map design. How wide are the chokepoints, how far apart are the bases, how large are the open areas, how easy is it to reposition armies, where/how can you build your production centers, which expansions can be harrassed easily by which sources... These things matter far more for balance than only how much hitpoints and damage the units have. You don't just play the same map all the time. So I think it's difficult to compare chess to Brood War in terms of balance. Also there has been additions made to chess, like castling. And unless I remember wrong, didn't the queen use to move like the king at some point in time? En Passant might also be an added rule, when people realized passed pawns were something really strong.
4:32 Magnus playing off Lex’s comment here was cute, but also considerate. Just a little reminder of like “hey I’m a noob at games I haven’t dedicated my career to same as you”
Than it plays it goes for max entropy, trying to make opponent make a mistake. As soon as it sees its winning, it changes it's style. Trying to limit opponent move choices.
@@XeroAIX Read carefully I said: "The neural net self-training AI blows the old school AI out of the water.". Note I said the *old school AI*. SF started using NN tech with Stockfish 12.
What I read is that the Queen's origin was the "Minister" from Chaturanga, which could only move 1 square diagonally like a really limited bishop. But I also read that they are unsure about how some pieces were meant to move as a bunch of info go lost to history and thus have different interpretations. So maybe one was the queen moving like the king. Heck, maybe people then were just making their own variants like people do now and it's just hard to find the original rules for the pieces because they all intermingles with the same pieces given different powers. It's pretty interesting history though!
I do not know a lot about Shogi but it seems like it is quite good in its own ways, having concepts similar to chess, but different enough to make it its own game. Many game players like to compare chess with go, when really they are just comparing how much they like chess with how much how they like go.
Shogi is hilarious. When you take your opponents piece you can air drop it anywhere *as your own piece*!. Makes for very powerful attacks and much more elaborate defences needing to be set up in advance.
its not surprising ai would be so easy to sacrifice pieces. sacrificing things to people is a thing that we naturally don't want to do. yeah we do still do it if there is benefit to it but an ai doesn't have emotions...yet...even if you realize its a game and don't consciously think about losing pieces as a thing to care about, i bet its still there subconsciously. an ai will just sacrifice because its a thing to do. i think its more interesting that magnus makes it seem like ai just sacrifices things without having any future plan for what happens after because you would think it would only do what's necessary to win so it would have some kind of plan after it sacrifices a piece.
I always described checkers as 'the art of sacrifice'. It was always about creating that one set up where you can sacrifice one for two, and then just going one for one until you've won.
Lex last bit was interesting, it going through its own natural selection. Such a great game, how could we of invented it? From the millions of other games we've selected away from.
It is not sacrificing anything, it just sees more steps ahead than human opponents and makes the optimum moves which involves letting go of material to move the opponents pieces... AI doesn't sacrifice anything it just carries out necessary moves to win.
@@jamesflames6987 you have a point 😌😂👌 but what I was trying to draw attention to was the fact that computer devoid of emotional context doesn't see it like that. Precisely why it baffles the most advanced players we got. Just the fact that we choose to give it such an emotionally charged word "sacrifice" goes to prove my point. Computer looks at it as optimization, maximization, us humans "sacrifice" that stark difference is why AI player surprises human players. Hope it makes sense... Cheers.
@MMMM great point mate, that's really interesting indeed. In fact, when you think about it, how we value chess pieces is ultimately arbitrary and depends on our lack of computational ability. If you can perfectly calculate all the possibilities of a situation, you no longer need to consider that a bishop is more valuable than a horse for the value of a position; a human cannot foresee an indeterminate series of potential blows, and is obliged to say to himself: "well, I don't know where this will lead me, but I know that a bishop is worth more than a knight, or that 'one knight is better than two pionts'. The computer, on the other hand, knows precisely where it is going to lead such and such an exchange, and the one after, and this of course more than a human brain is capable of doing, and it does not have to think in the absolute, as we have to, but always relatively to a well-determined situation. He can therefore evaluate the value of the pieces in a completely contextual way, relative to the current position. And indeed, in certain cases, even a pawn, if it is well positioned and if it fits into a perfectly calculated computer plan, can be worth a Queen, most certainly. You don't need all ouf your peices to win, after all Excuse my poor english ^^
@@Mag-Nuss I am glad the idea resonated with you, in addition to expressing it in terms of chess quite elegantly you have taken a step forward and made a profound statement about "value". Your point about the difficulties of determining value in complex situations is a crucial one that can be generalized to economics and society at large. As humans, we often have a hard time accurately attributing value due to our limited data and computational power compared to the infinite complexity of the real world. One could argue, to deal with this, we have developed cultures, codes of conduct, and social norms to help us navigate situations. So you could look at it as, everyone is playing someone else's winning hand trying to get a winning result while not knowing if the precise strategy will yield any positive outcome at all... So as you said: "Human can't foresee an indeterminate series of events", and I'd add that may be the reason why history is filled with tragedies and men made catastrophes because no one really knows precisely "the value", or the outcome of their next move...
how I agree with you ! The question of the value of things, of the cultural processes of valorization, is probably one of the philosophical problems that fascinates me the most; I could not be more intellectually delighted to find it here, linked to the pregnant question of artificial intelligence applied to chess. And the hypothesis that you propose, this rapprochement that you underline between our inaccuracy in chess and the tragic aspect of our historical political existence is very interesting, I must say that the idea is as beautiful as it is terrible. I'm going to think about it for a long time ^^
They need to get ai to modify the rules of chess to maximize the optimization of making the game the most enjoyable possible game. I wonder if the ai can actually make the game better and communicate the current flaws.
Part of why they seem to sacrifice without "fear" is because they look so far ahead that they are "confident". Note my terms there because engines have no emotions, they're just "there". But us humans are more likely to second guess ourselves because unless something is obvious, we are never sure if a sacrifice will turn out good even if it seems good because there's always a chance of missing something important. One bad move in the game could ruin the whole effort after all. But engines also have no such worry about losing their games as they have no attachment to the effort. So while we would gravitate towards equal trades and hoping for flubs, engines treat every position as something new and feels no obligation to keep it simple if it seems something better. Which I guess we could learn from in a way, but is indeed very difficult to put into practice.
My favorite is when AlphaZero trapped the black Queen to defeat a superior Stockfish program. If we ever design machines with emotions, despair has to be the first.
I'm new at chess and not very good. I don't understand some rules and terminology yet, but depending on the game I also tend to sacrifice pieces to improve overall position and win. I assume I do it subconsciously due to lack of safe moves but for A.I. to do it is interesting.
This is why it is a bit funny when people question whether or not God exist based on all the pain and turmoil in the world. If a God was achieving the optimum outcome for the majority life, then certain sacrifices would baffle the human mind. Human can be quite arrogant
I guess it is pretty subjective. Europeans play chess, while the Japanese play shogi. Magnus would likely not do well if he went to Tokyo to play in a high-level Shogi tournament. On the other hand, Japan came out 107th in the Olympiad. Oof!
Wait until you've dipped tangerine segments into a warm cheese sauce and served them to guests on cocktail sticks. Only then will you know what it means to be really weird.
My guess is that the best AI (before chess becomes a solved game) to the next highest beatable AI, will win by a flood of sacrifices ending by a only a single move ahead of the other AI with the opening to do so granted by the sacrifices.
1) it’s “funny” to see the computer designer or programmer or even user to take themself for great masters in chess. Personally some start play with their computers against me. That was befor all players in my adopted country got urgent need to go in toilet during the game with their portable phone.
Yup! A bunch actually. That piece is called the "Amazon" and it's basic form is "Amazon Chess" where the queen is just given that knight movement for both sides. It may also be called the "General" in stuff like Giant Chess
@@libertyprime9307 Im talking about ai in general when compared to humans in a competitive game. Ai doesn't have the greed that humans do. Them not losing anymore doesn't mean they aren't trying anymore.
@@drakeusmaximus2459 Even then it really just depends how the engine is programmed. You can tell for example to try for a win even at greater risk of losing, or you can tell it to prioritize not losing even if that means rarely winning, or somewhere in between. The reason they don't lose anymore is because even the more aggressive engines are just so good that they can't be grinded down in difficult endgames the same way humans do. Engines are incredible at endgame because it's such a calculation game at that point.
@Liberty prime he is talking about alphaGo. It will play moves that appear lazy but drastically reduce counterplay. Humans playing Baduk prefer to keep pushing advantages since skill gaps are pretty abusable
Full conversation with Magnus Carlsen: ua-cam.com/video/0ZO28NtkwwQ/v-deo.html
Quick note from Lex: The camera on Magnus died 20 minutes in. Most folks still just listen to audio-only version, but here on UA-cam we did our best to still make it interesting to watch & listen by adding image overlays. I mess things up sometimes, like in this case, and it hits me hard when I do. I'm sorry for this. I'm always working hard to improve. I hope you understand. Thank you for your patience and support along the way. I love you all.
Unacceptable!
@@johnnylove2073 yeah, i demand my money back! Oh, wait...
We all appreciate you Lex, especially when you periodically pretend to be human by making convincing oopsies like this! :-)
I am still going to devour this so don't sweat it!
you did a great job with the overlays, thanks
My willingness to sacrifice pieces also fascinates me sometimes.
Blundering and sacrificing are two different things :)
LOL
What if there is superior chess engine in deep subconcious, and all that blunders are actually brilliant? They don't work, because our concious mind freak out when we lost a pawn and "screw you guys, I'm going home" after loosing queen?
@@josephpeeler5434 It's only a blunder if you give up.
It’s the most fascinating when I do it accidentally
It's quite fascinating how I sacrifice pieces, as well. Unfortunately, the engine considers them blunders everytime
I just bank on the opponent making more blunders, or hope that if they play well in the opening that they will blunder the rest of the game.
AI: "I will sacrifice this human for the greater good."
🧐, yes indeed! 😳
Hhh
They are just pawns anyways
Not any human, tho.
It will assign a value to each one, and will sacrifice pawns to protect the overall plan much more willingly than rooks, queens, or even bishops.
AI: Joseph Stalin is the greatest leader throughout the world's history
Magnus is on point with the queen. The invention of the queen was the last major addition to chess, essentially creating modern chess as we know it.
It gave each side a clear win con
@@inafridge8573 not always a clear win condition though
Does anyone know how many points a queen is actually worth?
@@julianhodgson1961 9
Originally, the queen moved like the king. Chess underwent gradual gameplay adjustments over centuries, whereas StarCraft's underwent rapid play testing due to being a commercial product.
The most modern chess innovation has come with online play, where time pressure is needed as a counter to cheaters. (Chess clocks were merely a means to prevent games from going too long.)
If there was no bishop or knight, there would just be pawns, rooks and queens. Rooks and queens move in a pretty simplistic way. The geometric variety of chess comes from the knights and bishops. The knights go in an L and its path can't be obstructed by another piece. Bishops are really interesting because they are each constrained to one color.
Its weird i knew all of which you said(obviously) but i never really thought about it like that until i read what you wrote
@@drew4176 ya i never really thought about it until i saw this video too
You could say the same thing for pawns, bishops, and queens too.
Thats why I take out knights in rapid as fast as possible. Those things teleport
Knights are unique pieces. I love the spherical pattern that knights cover in space as well. Once you get two of them into the middle the can smother a lot of play together with good pawn usage.
What Magnus said about shogi is true, but only until you exchange a bunch of pieces, because shogi is like "bughouse" - you can return the pieces you captured from your opponent back into play and then it suddenly becomes clear why they are so weak. You can drop them back on almost any free square on the board so this just balances it out beautifully as opposed to how it would be if they were also really powerful on top of this. I highly recommend that any chess fan try learning shogi. It's an amazing game in its own right.
I love bug house! Thanks for telling me that the mechanic originates with Shogi
@@MaLLinz289 I am not sure if bughouse was directly inspired by shogi or not, but this mechanic had a few centuries of evolution in shogi so it works much much better in it than it does in bughouse. Chess pieces are just too powerful for this mechanic and bughouse is just someone's little experiment with an already well formed game.
I wholeheartedly recommend that you give shogi a shot. It's similar enough to chess for chess-fans to enjoy, but it's different enough that it gives a very different experience when playing.
My Japanese father in law taught me Shōgi. It's a wonderful game.
Even as a casual player such as myself, I've always been taken aback by how much ego plays a role in chess games, especially when it comes to losing pieces. Interesting how AlphaZero highlights this.
There is ego in chess but not about losing pieces specifically. Every beginner approaches the game with caution and is afraid of losing material balance, as, often, having less material leads to losing. Once you get more confident, you become less materialistic and start to see the profitable sacrifices and stop responding to every capture threat and look for a higher threat as a solution. Doesn't matter if you lose your queen if the move you make means mate in 2 or 3 moves...
I think the point of this video is the opposite from the point you got.
Humans don't hold onto pieces for ego, but a lack of calculating ability (finding eventual compensation).
The egotistical thing to do would be to make sacrifices. Chess players get boners over bold sacrifices.
Same, i mostly played when i was younger so id assume not much ego but i was always willing to do whatever i needed to to win
It isn't ego, its inability to play many lines many moves ahead and see how efficient a sacrifice is. I would also group castling.....when humans play they pretty much always castle whereas bots often dont castle and still win
I think theres a large fear aspect in castling. Humans seeing the indeterminate future think that castling can only be a good thing because of their fear of a decisive embarrassing loss due to the center opening up. The computer is unafraid of this and plays for the advantage regardless of position. I wouldnt say that humans castle because theyre incapable of calculating. Its more of a misunderstanding of the game and an aversion to risk embarrassment
For the note on balance, there might be asymmetry between pieces, but because each player gets the same set, balancing each piece is not as big of a deal as balancing factions like in Starcraft. And in terms of design, it still is hundreds or thousands of hours of playtesting that tell you about balance that chess got naturally.
Although because of the turn-based nature of chess, no piece can be super overpowering or the first-player advantage would skyrocket. It would be interesting to see how much chess strategy would change if the rules were adjusted so that each player could have the same number of turns regardless (e.g., if mated, be allowed to play one more move and if they can also mate it would be a draw)
Totally agree.....starcraft balance is very complex because there are 3 races and any change will impact say pvt differently than pvz; and now the game is mature enough and blizzard has not patched broodwar in 20 years so balance is done through maps not tweeks to the game. But with chess the biggest "balance" due to turn based vs real time. I guess balance would mean under perfect play it is a draw. But even that may not be practical balance since maybe quantum computers with near infinite elo (or even quantum computer who has solved chess) may be a draw (or even black win) but in human play white has advantage. Similar problem to starcraft where in the very top terran is arguably OP but at basically any level except top 10-20 players in the world protoss is consensus OP
The balance actually came from a sort of trial and error across milenia, it didn't came up naturally. But I do get your point.
Apparently the bishops only moved exactly two squares at first.
Which explains the juxtaposition with the knights movement to a degree
9 dan professional go player Micheal Redmond has mentioned that AlphaGO will make larger sacrifices and trades than humans. In Go at least part of this may be that humans have future plans for pieces when played, and attachments to groups that influence their play. The ability to change plans and not being attached to groups is a major difference between weaker and stronger human players also.
Way more insightful than the clip, thank you.
@@mattskov2917 Thanks. I was just watching another
Micheal Redmond review of an AlphaGO vs AlphaGO game just now. Some other points about AlphaGo vs human differences are, people have memories of why they played moves and previous calculation of positions in the game being played. AlphaGo has no memory of the game it is playing, it calculates its next move only from the current state of the board. AlphaGo does have a "memory" of every previous game it played but not the current game.
I think the thing about the "tension between the bishop and the knight" is that they're opposites in many ways: each bishop can control only one color, while a knight can control both, but which color switches whenever it moves; a bishop is a long-range piece that can be blunted by pawns along a diagonal, while a knight is a short-range piece that's often used to blockade pawns while still controlling squares beyond it; bishops prefer open boards with fewer pawns on it, while knights prefer closed boards with more pawns on it. The reason bishops are usually a bit more valuable is because most games end up (either soon or eventually) as being more open, so bishops are able to control more squares more quickly, but they're really close in value and that miniscule difference only matters at super-grandmaster level where they have to exploit every tiny edge.
nicely put
Not every piece might be useable over time, we just don't see it, hence better sacrifice it, make space and waste opponents move.
AI knows Strategery !
That makes no sense. Pieces and pawns aren't given away for nothing. The point of the sacrifice isn't that the piece being sacrificed is useless. What is usually meant by a sacrifice is an exchange of material where one side loses on point count---a rook (5 points) for a knight (3.25 points), say. The sacrificing player reasons that he lost the exchange on points but gained space or time or an attack or long-term positional advantage.
@@josephpeeler5434 ultimately it's about checkmate,so if we accomplish it ny giving away as many pieces in a good combination,that is then a good sacrifice
@@sunset1394 But it isn't about "giving away" pieces. The sacrifice might be a minor for 2 pawns or a rook for a bishop. If there is a net point loss in the exchange sac, then the player sacrificing must get either space, time, attack or positional advantage in return. The sac is an exchange. It is not a give away.
If it could win without sacrificing the piece, it would.
For anybody interested, "Game Changer: Alphazero's Groundbreaking Chess Strategies and the Promise of AI" is an excellent read on the subject.
It's from 2019 but was when AZ really tore into SF. It offers good insights and walks through some of the immortal games in that series, as well as explaining some of the novelties behind AZ.
I'm surprised nobody has mentioned Tal's sacrificing style
Because he sucked
It was beautiful but not always sound
mostly speculative and risky, but super beautiful
I think what people are more interested in is sacrifices that give you sure shot advantages (at least with our present knowledge of chess). Tal's sacrifices, most of them were not like that. They are still great though because to do better would be extremely difficult for humans.
Tal was ahead of his time
a true chess genius
“…Mistake for creativity”. Strong words.
Especially coming from someone out of their depth on a subject. It would have been interesting to have someone like Ilya Sutskever on the opposite end of the table drill down into that response from Magnus.
Shots fired
If it comes from calculation, it's not really creative, is it? I do think a neural net can be creative if the act of the computer coming up with the move is not calculation based.
If stockfish has just brute forced all the possibilities from saccing a piece and that gives a winning advantage, it's not really creativity IMO. It's the outcome of an optimization heuristic.
The original programmers were creative though.
Id love for Magnus to give Go a go. Itd be fascinating to watch all the top GMs go at it in Go.
I think Fischer's version is better, to be honest, as it promotes more creativity/improv. Not sure anyone is playing it though.
Did Bobby Fischer write this? 😂
FYI don't mention him around the community... It's Kool to say he wasn't good
@@AbcXyz-rn2lzI guess because of his controversial statements, though I'm sure he's not the only great chess player in history to have said or believed unpopular/unusual things.
There's plenty of people playing Chess 960 AKA Fischer Random. Not as many, but plenty
Yes, I think that is true. A paper I read indicated: Constant opening position and castling were negative to chess variety.
Every blunder is a gambit
3) About the queen in chess: Salomon, an humain, also an former chess champion in my nowadays country, said that I lose half of my strength in chess when the queen are not on the chessboard :-)
The 'knight vs bishop' is an big aspect in chess, but it's far from the basis of the game being fun. Just look at all of the variety in end games that don't feature knights or bishops.
Lex and Magnus definitely vibrate on the same frequency. I really have a lot of admiration for both of them.
How many hertz I wonder
i disagree
I really appreciate the extra effort put into this episode as a whole and this clip in particular after the technical issue.
my favourate thing about chess is that its a game of no rng but where every game is different
It’s so complicated the damn human players are the RNG generators!
If AlphaGo isn't creative, it sure makes me wonder what would it take for a machine to exhibit creativity...
I would argue sacrificing pieces in top level games like it does is extremely creative in its own right.
No, that's easy. Creativity (in the non-artistic sense) involves generating new explanations and creating new knowledge from those explanations. AlphaGo and AlphaZero don't generate any explanations, they simply optimize for winning a game with simple, known rules. That's why no chess or Go players have gotten better from playing/studying these systems (actually, some have gotten worse). When an AI can create new explanations that actually add to our knowledge, then we can consider it to be exhibiting the kind of creativity we actually care about.
top chess players have definitely learned from engines. the assertion that no chess player improves from engine games is flat out wrong
@@vinesthemonkey it isn't - there have been just as many players that have decreased their ELO after playing AIs than have improved - no better than chance. But regardless, it still isn't the kind of creativity we actually care about, even if you think these computational statistics programs are creative. Creating new explanations is what matters.
@Pollen Applebee I wish I could be so confident just pulling stuff out of my ass
The balance is because both sides have the same pieces which make the same movements. The only imbalance in chess is that white goes first
True enough but probably only for elite players. I feel that for mere mortals like myself it really doesn't matter that much.
Why there is no clip of a video but just the photo ?
There was a technical issue with the camera recording
When I go solo camping or ice fishing or anything dangerous/high-stakes I always have a checklist. That would prevent things like the battery issue.
I feel like Magnus is missing the point of weak pieces in Shogi. The power, complexity, and strategy comes from dropping pieces which the game is balanced around. Being able to put a captured weaker piece anywhere you want is more "powerful" (and interesting) than a queen in my opinion.
It's very interesting comment on the appeal of chess and he's probably right, however on the overall strategy, perhaps Shogi is superior as you suggest? I think I agree with both those considerations.
@@commentarytalk1446 If 2 rooks and bishops and queen were in shogi the game would end in checkmate very quickly I think.
Yeah. Suppose the chess with its explosiveness is just more Magnus's cup of tea.
He said he is noob in shogi
@@inafridge8573 In my experience each game tends to go a certain way in close games (broadly speaking of course)
Chess: much faster arrival to middle game with good excitement that cools down into a drawish endgame with fewer and fewer pieces.
Shogi: very slow maneuvering opening, but once middle game is in full swing it's usually a very sharp race to mate. This is due to clearing space but material always increasing after trades.
So it's kind of picking where you want the "explosiveness" to be concentrated.
The notion that bishops and knights are relatively equal is true. It is also true that, in any given position, the value of all pieces on the board are completely fluid. There’s certainly a subset of positions where a bishop or rook could be evaluated as being more valuable than a queen.
That is almost impossible. As in, if you were to replace a given queen with a rook or bishop, the only way your position is better is if you were about to have some sort of exotic stalemate
Which openings are based around saccing a pawn to get the bishop pair.
5:47 - Chess is always a "mirror matchup", both sides get the exact same units. Balancing a mirror matchup is much simpler than when the sides have different pieces.
Starcraft Brood War was released in 1999 and is still extremely balanced between the three races. But the balance is mostly achieved by map design. How wide are the chokepoints, how far apart are the bases, how large are the open areas, how easy is it to reposition armies, where/how can you build your production centers, which expansions can be harrassed easily by which sources... These things matter far more for balance than only how much hitpoints and damage the units have. You don't just play the same map all the time.
So I think it's difficult to compare chess to Brood War in terms of balance. Also there has been additions made to chess, like castling. And unless I remember wrong, didn't the queen use to move like the king at some point in time? En Passant might also be an added rule, when people realized passed pawns were something really strong.
The queen used to be very weak but didn't move like the King.
how chess rules evolved
4:32 Magnus playing off Lex’s comment here was cute, but also considerate. Just a little reminder of like “hey I’m a noob at games I haven’t dedicated my career to same as you”
I've been waiting for this conversation.
Love it
Than it plays it goes for max entropy, trying to make opponent make a mistake. As soon as it sees its winning, it changes it's style. Trying to limit opponent move choices.
Isn’t alpha zero already a weak engine compared to stockfish 15
With 300,000 games played across 50 years, I still mistakenly "sacrifice" (or you could call it a blunder) all sorts of pieces...
Why on your clips don't you show the actual video from the convo?
for me its always, at the moment, man vs man. its man that produces and programs the machines.
Explicitly coding AI behavior is passe now. The neural net self-training AI blows the old school AI out of the water.
Eh? A self learning program that is better at the most fundamental aspects of the game, memory and calculation?
@@dannygjk No, AlphaZero loses against Stockfish.
@@XeroAIX Read carefully I said:
"The neural net self-training AI blows the old school AI out of the water.". Note I said the *old school AI*. SF started using NN tech with Stockfish 12.
@@dannygjkMy point was more that Stockfish still uses primarily bruteforcing on high depths. Whereas AlphaZero is truly self-trained AI.
Ive heard somewhere that the queen was once only able to move like the king. Does anyone know anything about this?
What I read is that the Queen's origin was the "Minister" from Chaturanga, which could only move 1 square diagonally like a really limited bishop. But I also read that they are unsure about how some pieces were meant to move as a bunch of info go lost to history and thus have different interpretations. So maybe one was the queen moving like the king. Heck, maybe people then were just making their own variants like people do now and it's just hard to find the original rules for the pieces because they all intermingles with the same pieces given different powers. It's pretty interesting history though!
I beat a chess computer during a power cut.
Magnus!!! Holy shit I gotta watch this
I wouldn’t want an AI commanding soldiers.
I do not know a lot about Shogi but it seems like it is quite good in its own ways, having concepts similar to chess, but different enough to make it its own game.
Many game players like to compare chess with go, when really they are just comparing how much they like chess with how much how they like go.
Shogi is hilarious. When you take your opponents piece you can air drop it anywhere *as your own piece*!. Makes for very powerful attacks and much more elaborate defences needing to be set up in advance.
@@jamesflames6987 Maybe we could combine to two into one game.
Cheggi or Shoess
@@comic4relief “Cheggi” is essentially Crazyhouse. It’s very, *very* tactical and sacrificial.
It's amazing how they can talk without moving their lips
its not surprising ai would be so easy to sacrifice pieces. sacrificing things to people is a thing that we naturally don't want to do. yeah we do still do it if there is benefit to it but an ai doesn't have emotions...yet...even if you realize its a game and don't consciously think about losing pieces as a thing to care about, i bet its still there subconsciously. an ai will just sacrifice because its a thing to do. i think its more interesting that magnus makes it seem like ai just sacrifices things without having any future plan for what happens after because you would think it would only do what's necessary to win so it would have some kind of plan after it sacrifices a piece.
I have tried Shogi and found it 5 times more complicated than the regular chess. Very interesting game though
omg does lex play sc2
Is the Bishop really better than the Knight?
Sacrificing pieces = Alpha
Taking pieces = Beta
I always described checkers as 'the art of sacrifice'. It was always about creating that one set up where you can sacrifice one for two, and then just going one for one until you've won.
In a way that you can mistake for creativity...interesting
Imo this is why algos in the stock market sacrifice by creating micro dips to be able to gain an advantage afterwards
Really? I didn't know that
Lex last bit was interesting, it going through its own natural selection. Such a great game, how could we of invented it? From the millions of other games we've selected away from.
It is not sacrificing anything, it just sees more steps ahead than human opponents and makes the optimum moves which involves letting go of material to move the opponents pieces...
AI doesn't sacrifice anything it just carries out necessary moves to win.
Letting your piece be taken is called "sacrificing".
@@jamesflames6987 you have a point 😌😂👌 but what I was trying to draw attention to was the fact that computer devoid of emotional context doesn't see it like that. Precisely why it baffles the most advanced players we got.
Just the fact that we choose to give it such an emotionally charged word "sacrifice" goes to prove my point.
Computer looks at it as optimization, maximization, us humans "sacrifice" that stark difference is why AI player surprises human players.
Hope it makes sense... Cheers.
@MMMM great point mate, that's really interesting indeed. In fact, when you think about it, how we value chess pieces is ultimately arbitrary and depends on our lack of computational ability. If you can perfectly calculate all the possibilities of a situation, you no longer need to consider that a bishop is more valuable than a horse for the value of a position; a human cannot foresee an indeterminate series of potential blows, and is obliged to say to himself: "well, I don't know where this will lead me, but I know that a bishop is worth more than a knight, or that 'one knight is better than two pionts'.
The computer, on the other hand, knows precisely where it is going to lead such and such an exchange, and the one after, and this of course more than a human brain is capable of doing, and it does not have to think in the absolute, as we have to, but always relatively to a well-determined situation. He can therefore evaluate the value of the pieces in a completely contextual way, relative to the current position. And indeed, in certain cases, even a pawn, if it is well positioned and if it fits into a perfectly calculated computer plan, can be worth a Queen, most certainly. You don't need all ouf your peices to win, after all
Excuse my poor english ^^
@@Mag-Nuss I am glad the idea resonated with you, in addition to expressing it in terms of chess quite elegantly you have taken a step forward and made a profound statement about "value".
Your point about the difficulties of determining value in complex situations is a crucial one that can be generalized to economics and society at large. As humans, we often have a hard time accurately attributing value due to our limited data and computational power compared to the infinite complexity of the real world. One could argue, to deal with this, we have developed cultures, codes of conduct, and social norms to help us navigate situations. So you could look at it as, everyone is playing someone else's winning hand trying to get a winning result while not knowing if the precise strategy will yield any positive outcome at all... So as you said: "Human can't foresee an indeterminate series of events", and I'd add that may be the reason why history is filled with tragedies and men made catastrophes because no one really knows precisely "the value", or the outcome of their next move...
how I agree with you ! The question of the value of things, of the cultural processes of valorization, is probably one of the philosophical problems that fascinates me the most; I could not be more intellectually delighted to find it here, linked to the pregnant question of artificial intelligence applied to chess.
And the hypothesis that you propose, this rapprochement that you underline between our inaccuracy in chess and the tragic aspect of our historical political existence is very interesting, I must say that the idea is as beautiful as it is terrible.
I'm going to think about it for a long time ^^
What is up with the still frames? Super awkward. Edit: got it. Links to original have it. Moving on. Thanks Lex.
what was it like talking to magnus carlsen (I spelled his name right this time) compared to gothamchess?
Of all the sacrifices, to sacrifice your scarce youth to push pieces around on a board is an odd one to justify.
They need to get ai to modify the rules of chess to maximize the optimization of making the game the most enjoyable possible game. I wonder if the ai can actually make the game better and communicate the current flaws.
Fast forward to today and buddy sac'ed the rook out the opening for nothing in return and still won. Went 7 and 3 doing it.
Weird to use a thumbnail of something that literally isn't a sacrifice lol
The meaning of the word sacrifice is lost on you.
I hope to get to an elo level where the bishops are better than the horseys
From every elo it’s the same, bishops are better at the start, both pieces are neutral in the mid game and horseys are better at end games.
@@Glock7eventeen it’s easier to take a queen or a rook with horsey than with a bishop at lower elos, forks aren’t as easily seen
Part of why they seem to sacrifice without "fear" is because they look so far ahead that they are "confident". Note my terms there because engines have no emotions, they're just "there". But us humans are more likely to second guess ourselves because unless something is obvious, we are never sure if a sacrifice will turn out good even if it seems good because there's always a chance of missing something important. One bad move in the game could ruin the whole effort after all. But engines also have no such worry about losing their games as they have no attachment to the effort. So while we would gravitate towards equal trades and hoping for flubs, engines treat every position as something new and feels no obligation to keep it simple if it seems something better. Which I guess we could learn from in a way, but is indeed very difficult to put into practice.
Sacrificing a pawn for a bishop pair???
Give a knight and a pawn for a bishop.
My favorite is when AlphaZero trapped the black Queen to defeat a superior Stockfish program. If we ever design machines with emotions, despair has to be the first.
Stock fish wasn’t superior… in 100 games Alpha won about 40, and the rest were draws. It lost zero games to stockfish 8
Alpha zero uses hans code
I'm new at chess and not very good. I don't understand some rules and terminology yet, but depending on the game I also tend to sacrifice pieces to improve overall position and win. I assume I do it subconsciously due to lack of safe moves but for A.I. to do it is interesting.
This is why it is a bit funny when people question whether or not God exist based on all the pain and turmoil in the world. If a God was achieving the optimum outcome for the majority life, then certain sacrifices would baffle the human mind. Human can be quite arrogant
your the one and only man that I love in my life.take care of yourself.
I guess it is pretty subjective. Europeans play chess, while the Japanese play shogi. Magnus would likely not do well if he went to Tokyo to play in a high-level Shogi tournament. On the other hand, Japan came out 107th in the Olympiad. Oof!
Oh man I would love to see Japan get in to chess. I bet they would be a real contender on the scene.
Very interesting stuff
This comments section is really weird lol
Wait until you've dipped tangerine segments into a warm cheese sauce and served them to guests on cocktail sticks. Only then will you know what it means to be really weird.
2724, man i got so much more respect for this set haha.
My guess is that the best AI (before chess becomes a solved game) to the next highest beatable AI, will win by a flood of sacrifices ending by a only a single move ahead of the other AI with the opening to do so granted by the sacrifices.
1) it’s “funny” to see the computer designer or programmer or even user to take themself for great masters in chess. Personally some start play with their computers against me. That was befor all players in my adopted country got urgent need to go in toilet during the game with their portable phone.
Demis actually does play chess lol, or at least according to the alphazero documentary
Alpha zero game changer
Shogi ftw.
Sacrifice is paramount in good Shogi play .
Inspiration words by magnus!!
at no point did magnus say the quote in the title. click bait. you stole my time for a nickel from yt.
Is there a version of chess where the queen can move in L shapes also? I'm just thinking of the utter chaos that would bring lol.
Yup! A bunch actually. That piece is called the "Amazon" and it's basic form is "Amazon Chess" where the queen is just given that knight movement for both sides. It may also be called the "General" in stuff like Giant Chess
The Queen is Boss!
Lex likes to interview people who are superior to himself.
That is how you evolve.
@@johanw2267 chicken and egg problem?
that's how it's supposed to be... why would anyone want to interview someone beneath them?
No video?
Hans is the only human closest to AI in intuitively sacking pieces. Genius
Willingness? An AI with its own will. Now that's true horror
Sounds like the USGOV🇺🇸 Call 9/11😳
The point of chess is to get a pawn down to the other side. Sacrificing a couple pawns to completely control the board is definitely worth it
Why can’t I see them talking
camera died
I want to see, please, AlphaGo playing a chess idiot like me. 🐈⬛
Two beeshups, vat els?
AI: Joseph Stalin is the greatest leader in the world's history
Ahh yes, AI is willing to "sacrifice pieces" to reach an ultimate goal.
Ah yes, let's reword the top comment to farm likes and feel good about ourselves!
Magnus sacs like alpha zero these days
Some Skynet-level sh*t. Not scary. Not at all 😟
Demis is better
Sacrificing the queen first unpredictably is how you beat AZ👌
99% chance to win by 1 point vs a 90% chance to win by 50, the ai will take higher chance of winning, a human wants more points.
Nah chess engines just simply don't lose anymore. The only outcomes against the best ones are lose/draw.
@@libertyprime9307 Im talking about ai in general when compared to humans in a competitive game. Ai doesn't have the greed that humans do. Them not losing anymore doesn't mean they aren't trying anymore.
@@drakeusmaximus2459 Even then it really just depends how the engine is programmed.
You can tell for example to try for a win even at greater risk of losing, or you can tell it to prioritize not losing even if that means rarely winning, or somewhere in between.
The reason they don't lose anymore is because even the more aggressive engines are just so good that they can't be grinded down in difficult endgames the same way humans do. Engines are incredible at endgame because it's such a calculation game at that point.
@Liberty prime he is talking about alphaGo. It will play moves that appear lazy but drastically reduce counterplay. Humans playing Baduk prefer to keep pushing advantages since skill gaps are pretty abusable
If you knew ju jitsu you don’t need to learn chess - Rogan
Rogan connect everything to bjj
Rogan's monkey brain excuse for not becoming good at chess lol