Great stuff Daniel ! Why not suggest to Google they consider sponsoring a tournament for the world's top GM's where each game in each round is a different chess variant ? It would be a great promotion for Google and attract millions of viewers across the world. It would open so many eyes, both old and new, to the limitless beauty of chess.
Interesting. Did they try to come up with a variant that gave even more advantage to white? That's really the only way to get a non-drawish outcome with a machine that learns (close to) optimal play.
If they add the variant of no castling to this Torpedo Chess, and also played with Chess 960 format of NBRKQRBN (one of the least drawish variant of Chess 960) there would be lesser draw.
As I understand it the old Indian chess allowed pieces to move one or two squares at the time. Eg bishop (the elephant) could not run from one side to the other side of the board. Neither could the queen. Soldiers (pawns) moved one square at the time and en passant was not needed. It was like a real old fashion ethical battle ! However as the result of all these rule changes the whole game has become faster. The addition of the clock made it even faster. Almost all games in the ‘watch’ section of chess.com are 3 minutes games. If I am lucky I watch a 15 minutes game. They want to win lose draw live or die quickly. They want the end result and not interested in the journey. It took me 10 years to move the pawn one step at the time and not to be taken en passant. now we have en passant all over the board. Having said all that the game that you described forced me to think longer about the situations.
have to say, glad I learned a bit of chess basics like opening and such so I can appreciate the game better and watch stuff like this, because until learning a few basics with dr wolf I knew how to move pieces, but I didn't know what to do because it was just moving things until I got blindsided by a loss.
If you ❤️ my videos do *subscribe* bit.ly/powerplaysubscription and do checkout the *supporting* *options* through Patreon: bit.ly/patreondanielking or through *PayPal* (links in the description)
Wisdom of Nimzowitsch: "A pawn is a criminal which should be kept under lock and key. Mild measures, such as police surveillance, are not sufficient." Carsten - unknown clubplayer
If they add the variant of no castling to this Torpedo Chess, and also played with Chess 960 format of NBRKQRBN (one of the least drawish variant of Chess 960) there would be lesser draw.
No problem with variants. But we still love the traditional chess. The computers can add as much as they want because they have no problems with calculations. Chess at GM level is already quite a very hard task
After Rd8 can black play Bg3+ and if Kxg3 f1=Q or if Kg1 f2+ and then Ba6? Is it illegal to promote the pawn after moving two squares? Anyway very interesting variants and it should be a lot of fun trying them out
These look like great fun! The self-capture variant and this one are simple enough to implement in casual games; while this won’t make it to precessional chess any time soon, probably never, online chess sites would really benefit from incorporating some of these. I’d love to test my skills at this!
Is there a variant of chess ,in wich u must promote the pawn into the piece at start ? Promoting to h1/8 = Rook ,g1/7=Knight aso ...d1/8 and e1/8 would go Queening ...
Interesting - would make outside pawns stronger (relatively) since they can "rook" into a win whereas the b/c/f/g pawns are not winning on their own.Might just end up creating a lot more draws.
What are typical scores for white and black in the variants? If normal chess is, say, an expectation of 0.55:0.45 (white:black) score at the top level (I'm just making that number up), it seems like this torpedo chess would be even more favourable for white, don't ya think?
It's certainly fun and interesting but I don't see how this is more than another cool demonstration of the power of neural networks and a nice occupational therapy for Kramnik. It's not like there aren't already lots of chess variants out there, at least one of which (chess960) is certainly quite good at avoiding the often lamented "draw death" of chess (even though I don't see how a draw is inherently a bad thing). Anyway, I don't see any of these variants making an impact on chess as a whole other than maybe giving us another few weird chess variants on the chess website of our choice. The only exception, possibly, being the stalemate=win rule change but that's another topic with lots of implications of its own.
+Commanber yea I don't understand why they don't play it against Leela. Demonstrate that your approach is better than just normal neural networks like Leela
The motivation for DeepMind is not necessarily to make an impact on chess - the abstract states: "It is non-trivial to design engaging and balanced sets of game rules....[AlphaZero] is a system that can learn near-optimal strategies for any rule set from scratch, without any human supervision." So in a broader sense, this paper hints that neural networks can be used to experiment on potential changes to rules/laws/constraints with real-world applications prior to implementation, using chess variants as a proof-of-concept.
@@eyeofhorus1301 That wouldn't prove anything. Comparing an AI that works on your every day laptop versus an AI that works on Google servers. What's different in their approach to other neural nets btw?
Stalemate=win seems very interesting to me. I know in Xiangqi (Chinese chess) stalemating an opponent is a win and putting your opponent in perpetual checks are a loss. I'm not familiar with the effects on top level play, but as a casual player it feels more right to me. I feel the concept that a player having no legal moves, which occurs in both stalemate and checkmate, means a loss just seems right to me. Stalemate would be akin to zugzwang because player is put into a situation where no matter the move they make, they compromise their king. What I am not familiar with draw rates in Xiangqi, but the rules are significantly different enough that changing stalemate rules probably isn't the only difference. Nor do I know what strategies would prevail with these rules. Would a player play even safer because taking a risk is even more disastrous? Would the broader winning rules be incentive for more aggressive play? I personally don't think a draw is bad, but there are definitely casual onlookers who do, which may be hurting the growth of chess. Not all draws are created equal. A draw like the one presented in this video is exciting because it seems like both sides are balanced on a knifes edge. Finding ways to incentivize enterprising play should be the goal. Not just eliminating draws.
@@edmis90 Under equal conditions it would show Alpha still has the most superior approach and hasn't lost its place at top of the chess world. One thing is the reinforcement learning thats heavily incorporated into alpha but not leela
Torpedo Chess and Semi-torpedo implemented as Zillions program: mlwi.magix.net/bg/torpedochess.htm (Its playing strength is average club player, at least.)
In my head it makes sense that more passive openings by white are less risky as you can make up closing in on center any time, so openings like Van't Kruijs or Saragossa (my favorites) are more viable. Am I incorrect? I probably am.
I used standard ChessBase software, but when it came to implementing 'new' moves I just had to set up a new position and go from there. It was time consuming!
This 6. g43 reminds me of a famous old line from the Semi-Slav : 5. e3 Nbd7 6. Qc2 Bd6 7. g4!? Unfortunately that line has turned out really not to work for White, but this Torpedo version could be a lot more promising!
Interesting rule. I think the colmplexity that arises is too much for the human brain. But it is a good alternativ for friendly matches in the pub with some good beers.
16:02 I think Rd8 gets answered by Bg3+ and white actually is in trouble, no? Kxg3 f1=Q Kf8 Ba3+ and it's impossible to stop black queening Kg8 f7+ Kf8 Ba3+ and black is gonna promote again
don't you get it? It's not the which variant of chess is more attacking or interesting and could have less or more draws but strength of the player. And trust me traditional chess has more than enough for extremely intelligent ideas. Whatever varient it is if it is still 90% draws how have you made it interesting. You can only make it interesting by increasing the quality of the player.We love alpha zero because it plays same old chess but better and different.
just checked the paper, nothing much changed from previous results published by AZ about classical chess. white still has advantage in all variants, but torpedo is outright bad for black.
If you would like to look at the complete paper, you can download it from here: arxiv.org/abs/2009.04374
Great stuff Daniel ! Why not suggest to Google they consider sponsoring a tournament for the world's top GM's where each game in each round is a different chess variant ? It would be a great promotion for Google and attract millions of viewers across the world. It would open so many eyes, both old and new, to the limitless beauty of chess.
great suggestion
with Daniel on commentary!
It would be amazing. I'd like to see the strongest GMs playing the Capablanca and/or the S-chess variants.
@@fib1478 I think he means minor variations like pawns back, etc Even 3-checks or king of the hill could be thrown in, just for fun.
I would love to see a tournament of Torpedo Chess + No Castling allowed + NBRKQRBN (the 11th least drawish variant of Chess 960).
This was the only variant with less than 90% draws in one minute per move games, and it was an 89.4% draw rate!
Let me guess: because of the amount of games played, that difference was enough to be significant.
Interesting. Did they try to come up with a variant that gave even more advantage to white? That's really the only way to get a non-drawish outcome with a machine that learns (close to) optimal play.
Was there a variant where the side that couldn't move lost, so lost the stalemate if you like, as imagine that would have a lower draw rate.
@@PSModelling yes
If they add the variant of no castling to this Torpedo Chess, and also played with Chess 960 format of NBRKQRBN (one of the least drawish variant of Chess 960) there would be lesser draw.
When Philidor said "Pawns are the soul of chess" he meant Torpedo chess.
When Alpha Zero is giving extra meaning to your channel , you know it's meant to be
As I understand it the old Indian chess allowed pieces to move one or two squares at the time. Eg bishop (the elephant) could not run from one side to the other side of the board. Neither could the queen. Soldiers (pawns) moved one square at the time and en passant was not needed. It was like a real old fashion ethical battle ! However as the result of all these rule changes the whole game has become faster. The addition of the clock made it even faster. Almost all games in the ‘watch’ section of chess.com are 3 minutes games. If I am lucky I watch a 15 minutes game. They want to win lose draw live or die quickly. They want the end result and not interested in the journey. It took me 10 years to move the pawn one step at the time and not to be taken en passant. now we have en passant all over the board. Having said all that the game that you described forced me to think longer about the situations.
There were some f1=Q torpedo move interesting to look out for at some point.
have to say, glad I learned a bit of chess basics like opening and such so I can appreciate the game better and watch stuff like this, because until learning a few basics with dr wolf I knew how to move pieces, but I didn't know what to do because it was just moving things until I got blindsided by a loss.
If you ❤️ my videos do *subscribe* bit.ly/powerplaysubscription and do checkout the *supporting* *options* through Patreon: bit.ly/patreondanielking or through *PayPal* (links in the description)
Torpedo chess, also known as the only chess variant where 1 pawn is worth 1.5 knights
12:11 **BRAIN IMMEDIATELY MELTS**
Wisdom of Nimzowitsch:
"A pawn is a criminal which should be kept under lock and key. Mild measures, such as police surveillance, are not sufficient."
Carsten - unknown clubplayer
Ha! So the square rule is really a rectangle rule in this variant :)
This one is okay, but i still prefer the self capture. Can't wait to see the other ones presented by Daniel.
If they add the variant of no castling to this Torpedo Chess, and also played with Chess 960 format of NBRKQRBN (one of the least drawish variant of Chess 960) there would be lesser draw.
No problem with variants. But we still love the traditional chess. The computers can add as much as they want because they have no problems with calculations. Chess at GM level is already quite a very hard task
After Rd8 can black play Bg3+ and if Kxg3 f1=Q or if Kg1 f2+ and then Ba6? Is it illegal to promote the pawn after moving two squares? Anyway very interesting variants and it should be a lot of fun trying them out
These look like great fun! The self-capture variant and this one are simple enough to implement in casual games; while this won’t make it to precessional chess any time soon, probably never, online chess sites would really benefit from incorporating some of these. I’d love to test my skills at this!
Now I want to see pawns that can move 1,2, or 3 squares on the first move (but not thereafter). The openings might change dramatically.
Scary pawns. The rule of the square no longer works. Lots of endgames are going to be sharper because of this rule.
True, the pawns can now outrun the king.
great video, what about bxb2?
I was thinking the same at 12:02.
I think it makes playing Rd1 more difficult because of Rxa2 but im bad at chess so idk
Is there a variant of chess ,in wich u must promote the pawn into the piece at start ?
Promoting to h1/8 = Rook ,g1/7=Knight aso ...d1/8 and e1/8 would go Queening ...
that's a really cool idea
@@Overclocked3770K So simple that i'm surprised it's not a known version :p
Interesting - would make outside pawns stronger (relatively) since they can "rook" into a win whereas the b/c/f/g pawns are not winning on their own.Might just end up creating a lot more draws.
I’d rather see a standard alpha zero game against a top level programme
You'd need a rewrite of a standard engine to accommodate a rule change, but Alphago can train itself in a few hours.
I still find the "self-capture" to have the most potential to evolve in an actual sub-genre than the other variants shown so far
What are typical scores for white and black in the variants? If normal chess is, say, an expectation of 0.55:0.45 (white:black) score at the top level (I'm just making that number up), it seems like this torpedo chess would be even more favourable for white, don't ya think?
It's certainly fun and interesting but I don't see how this is more than another cool demonstration of the power of neural networks and a nice occupational therapy for Kramnik.
It's not like there aren't already lots of chess variants out there, at least one of which (chess960) is certainly quite good at avoiding the often lamented "draw death" of chess (even though I don't see how a draw is inherently a bad thing).
Anyway, I don't see any of these variants making an impact on chess as a whole other than maybe giving us another few weird chess variants on the chess website of our choice.
The only exception, possibly, being the stalemate=win rule change but that's another topic with lots of implications of its own.
+Commanber yea I don't understand why they don't play it against Leela. Demonstrate that your approach is better than just normal neural networks like Leela
The motivation for DeepMind is not necessarily to make an impact on chess - the abstract states: "It is non-trivial to design engaging and balanced sets of game rules....[AlphaZero] is a system that can learn near-optimal strategies for any rule set from scratch, without any human supervision." So in a broader sense, this paper hints that neural networks can be used to experiment on potential changes to rules/laws/constraints with real-world applications prior to implementation, using chess variants as a proof-of-concept.
@@eyeofhorus1301
That wouldn't prove anything.
Comparing an AI that works on your every day laptop versus an AI that works on Google servers.
What's different in their approach to other neural nets btw?
Stalemate=win seems very interesting to me. I know in Xiangqi (Chinese chess) stalemating an opponent is a win and putting your opponent in perpetual checks are a loss. I'm not familiar with the effects on top level play, but as a casual player it feels more right to me. I feel the concept that a player having no legal moves, which occurs in both stalemate and checkmate, means a loss just seems right to me. Stalemate would be akin to zugzwang because player is put into a situation where no matter the move they make, they compromise their king.
What I am not familiar with draw rates in Xiangqi, but the rules are significantly different enough that changing stalemate rules probably isn't the only difference. Nor do I know what strategies would prevail with these rules. Would a player play even safer because taking a risk is even more disastrous? Would the broader winning rules be incentive for more aggressive play? I personally don't think a draw is bad, but there are definitely casual onlookers who do, which may be hurting the growth of chess. Not all draws are created equal. A draw like the one presented in this video is exciting because it seems like both sides are balanced on a knifes edge. Finding ways to incentivize enterprising play should be the goal. Not just eliminating draws.
@@edmis90 Under equal conditions it would show Alpha still has the most superior approach and hasn't lost its place at top of the chess world. One thing is the reinforcement learning thats heavily incorporated into alpha but not leela
Torpedo Chess and Semi-torpedo implemented as Zillions program: mlwi.magix.net/bg/torpedochess.htm
(Its playing strength is average club player, at least.)
Why at around 11 minutes in, didn't white take on b2 with the bishop? That would have avoided giving up the rook
Perhaps after white plays Bxb2 then:
Ra7, Be5 Rxc7, Bxc7 Rxc7
In my head it makes sense that more passive openings by white are less risky as you can make up closing in on center any time, so openings like Van't Kruijs or Saragossa (my favorites) are more viable. Am I incorrect? I probably am.
Just allow a pawn to move one square in any direction. That would make for some interesting games.
L. Legan's "Aviation Chess" (1913), played in Paris during World War I: mlwi.magix.net/bg/aviationchess.htm
I'm so excited by these variants and the return of Alpha Zero. Thanks so much for going over these games!!!
Are you using a modified version of Chessbase? Or position setup maybe?
I used standard ChessBase software, but when it came to implementing 'new' moves I just had to set up a new position and go from there. It was time consuming!
@@PowerPlayChess I bet. Not just in Chessbase but also the video editing.
This 6. g43 reminds me of a famous old line from the Semi-Slav : 5. e3 Nbd7 6. Qc2 Bd6 7. g4!? Unfortunately that line has turned out really not to work for White, but this Torpedo version could be a lot more promising!
Alphazero very very very strong💪
Interesting variant, although I would prefer a rule that pawns on the 6th rank could only advance 1 square instead of optionally 2.
Is it legal to jump the pawn two squares twice? The H pawn first jumps to h4 then it jumps again to h6. Is this a legal move in this torpedo chess?
yes pawns can always move two squares in torpedo
Never mind this comment. My english is poor. 😂😂
L'Italia s'è desta thanks.
Awesome game!!
I wonder what the comparative value of a pawn is in this game. Losing a bishop for a pawn might not be such a bad deal.
Interesting rule. I think the colmplexity that arises is too much for the human brain. But it is a good alternativ for friendly matches in the pub with some good beers.
5:50 zwischenschach = zwischenzug that is a check.
why not bishop takes b2 and maintains the pin?
It's clear that Alpha zero could not make anymore progress in Normal chess.
So, how about a 10x10 board instead?
16:02 I think Rd8 gets answered by Bg3+ and white actually is in trouble, no?
Kxg3 f1=Q
Kf8 Ba3+ and it's impossible to stop black queening
Kg8 f7+
Kf8 Ba3+ and black is gonna promote again
Does this variant increase the nominal advantage the white side has from the get-go?!
don't you get it? It's not the which variant of chess is more attacking or interesting and could have less or more draws but strength of the player. And trust me traditional chess has more than enough for extremely intelligent ideas. Whatever varient it is if it is still 90% draws how have you made it interesting. You can only make it interesting by increasing the quality of the player.We love alpha zero because it plays same old chess but better and different.
how many games they published playing torpedo chess? looks like overwhelming advantage for white, only two results possible, draw or white winning.
just checked the paper, nothing much changed from previous results published by AZ about classical chess. white still has advantage in all variants, but torpedo is outright bad for black.
That was amazingly fun.
En passant is so confusing in this variant!
Much better than std chess
Pawn can reach d5 in 2 moves anyways
Interesting!
also chess960 / Fischer Random Chess?
Something needs to be done about chess. It's too drawish.
For God's sake don't show Ginger GM!