Great insights! I've always disliked Chess as a metaphor for warfare so I'm glad to see you poke at that. As you pointed out, there's only a middle game that has variability and most of that ends up being psychological warfare, not chess strategy per se.
I was amazed to learn that Chess Masters will burn as many calories in a match as a marathoner...because they're doing that rote computational analysis in their brain like a computer.
But in warfare I can change the rules... or not play by them... I like chess to a degree but it was too ridged for me
There is a lot more creativity in chess than most realize. In fact watching high level chess engines fight against each other leads to some incredible positions. Between humans, once the game departs from opening theory, amazing things can happen on the board.
I think there are more constraints than you allow for and less creativity in war than you think might be possible (that would be a long argument to have, wouldn’t it! lol), but If regular chess is too rigid, you would probably like Kriegspiel!
There is a certain value in rigidity in war. We might call that discipline but that discipline unlocks a whole other level of capability. I think of the maturation of warfare from Mele to Mass to Maneuver. Mele might seem to be the most flexible but it's the least scalable.
There's a thousand layers to the application of creativity and constraints where, like discipline, constraints breed creativity and creativity can flex constraints. There's a lot to pick on with that for sure.
I think it would lose a lot of its magic on the computer, but it would be relatively easy to program. It’s also underrated as a spectator sport. It is incredibly exciting for spectators who can see everything that’s happening and wondering if the players will find good moves.
I remember a series of short science fiction stories that featured humans and Martians working together in exploring space. The Martians were very intelligent as a species but had not developed the physical abilities humans have to manipulate their environment. However, they did have telekinetic abilities and the ability to influence human senses, especially sight, so that we could see only what they wanted us to see - sometimes hiding things and other times convincing us see something that wasn’t there. The Martians loved the human game of chess and played it whenever they could while using their abilities to present to human players whatever board situation they wanted. That was my introduction to chess variations.
“I think you could build a dominant Kriegspiel chess engine, but not to the level of a 700 point ELO advantage.”
Yes, the computers will probably end up beating the best humans at Kriegspiel (if they are not already). But the *difference* in performance will likely be narrower. This narrower differential is because of the uncertainty.
I just feel pretty confident that you would be able to beat humans. AI doesn’t actually know the future. “The best it can do (at this point) is see patterns and make probabilistic guesses” humans can’t predict the future either, so we have to use the information available to us and make educated guesses
I had never heard of Kriegspiel chess! I find it quite terrifying. But I don't buy your analysis of AI capabilities.
There are AIs already that can be trained to play a game (Go being one of the first done this way) by making random moves, at first. Over thousands or millions of iterations, it gets very very good at the game. This is how I would train a system to play Kreigspiel chess. Pit 2 AIs against each other and just let them go. Eventually, they will reach the level of beating a human every time.
The good news (if you can call it that) in war is that you can only train the machine on simulations. The real thing doesn't happen often enough for the machine to build a workable model.
Just to clarify, your position is that a Kriegspiel engine will eventually be able to obtain the equivalent of a greater than 700 ELO rating over the best human players?
Sure anything is possible. I’m just saying, based on what I know about how chess engines work and the work that was done on building Kriegspiel engines about a decade ago, it seems unlikely that Kriegspiel engines will achieve the same level of dominance that traditional chess engines have achieved. A Kriegspiel AI might eventually beat the best humans, but the *level* difference between the AI and the best humans will simply be narrower.
What you write is interesting, but this "But how would AI fair on a Kriegspiel board where it does not have perfect information? I would submit to you that you could never build a chess engine that is as dominant over humans at Kriegspiel as it is at regular chess." is likely wrong.
NEVER is a mighty powerful claim. I am confident even under imperfect information chess engines would consistentl beat even grandmasters.
The reasons machines can consistently outperform humans are: PERFECT memory, FAST access. Imperfect information changes neither. The computer knows exactly where all your pieces are at start, & minimax STILL works for that reason. Moving pawns to take the center board is correct whether I know or do not know your positions. That is, the correct strategy wins EVEN FASTER if you deviate from it.
Examples may help. I am white boldly move to take center but clever you opens with ... a knights paw. Cool! I have center board!
There are basically only six "good" opening moves, in chess. They are still good even if opponent makes bad moves, even if information is imperfect.
Fischer thought you could get out of this sort of problem by varying the start positions of non-pawns randomly.
If you wish to play chess with less than perfect information simply build a set like stratego. This is reasonably accurate since its semi perfect. You know Where the enemy is but not What he has. Then you neeWhat you write is interesting, but this "But how would AI fair on a Kriegspiel board where it does not have perfect information? I would submit to you that you could never build a chess engine that is as dominant over humans at Kriegspiel as it is at regular chess." is likely wrong.
NEVER is a mighty powerful claim. I am confident even under imperfect information chess engines would consistentl beat even grandmasters.
The reasons machines can consistently outperform humans are: PERFECT memory, FAST access. Imperfect information changes neither. The computer knows exactly where all your pieces are at start, & minimax STILL works for that reason. Moving pawns to take the center board is correct whether I know or do not know your positions. That is, the correct strategy wins EVEN FASTER if you deviate from it.
Examples may help. I am white boldly move to take center but clever you opens with ... a knights paw. Cool! I have center board!
There are basically only six "good" opening moves, in chess. They are still good even if opponent makes bad moves, even if information is imperfect.
Fischer thought you could get out of this sort of problem by varying the start positions of non-pawns randomly.
If you wish to play chess with less than perfect information simply build a set like stratego. This is reasonably accurate since its semi perfect. You know Where the enemy is but not What he has. No referee then.
Your account of war is very likely inaccurate, but there is too little there to accurately account for. First, the problems (domain) is vast, second you haven't defined (your quotes mean what?) "solved".
Your account of war is very likely inaccurate, but there is too little there to accurately account for. First, the problems (domain) is vast, second you haven't defined (your quotes mean what?) "solved".
Thank you for taking the time to leave such a lengthy comment, Eric! I think once you’ve played a few games of Kriegspiel you might see things from a different perspective. You did not refute the central point of my argument: the computer relies on decision trees to make the strongest moves, and it cannot generate those decision trees if it does not know where the pieces are. The lack of perfect information changes the game substantially.
instead of using a known payoff for each node in the decision tree you just use a probability range. even if we decide that i don't get to know which piece killed me i can still infer which ones likely do. i can also infer positions and pieces from trying possible prohibited moves. imperfect information does not mean no information.
Correct, you repeated the same point I made when I said, “For example, if you attempt to make a move and the referee announces that the move is illegal for 3-reasons, the computer may be able to, over time, build an idea of where an opponent’s pieces are. If you discover an enemy piece in an unexpected location, the computer algorithm could make a very good guess about which piece it is, how it got there, and where other pieces are as a result.”
Great insights! I've always disliked Chess as a metaphor for warfare so I'm glad to see you poke at that. As you pointed out, there's only a middle game that has variability and most of that ends up being psychological warfare, not chess strategy per se.
I was amazed to learn that Chess Masters will burn as many calories in a match as a marathoner...because they're doing that rote computational analysis in their brain like a computer.
But in warfare I can change the rules... or not play by them... I like chess to a degree but it was too ridged for me
There is a lot more creativity in chess than most realize. In fact watching high level chess engines fight against each other leads to some incredible positions. Between humans, once the game departs from opening theory, amazing things can happen on the board.
I think there are more constraints than you allow for and less creativity in war than you think might be possible (that would be a long argument to have, wouldn’t it! lol), but If regular chess is too rigid, you would probably like Kriegspiel!
There is a certain value in rigidity in war. We might call that discipline but that discipline unlocks a whole other level of capability. I think of the maturation of warfare from Mele to Mass to Maneuver. Mele might seem to be the most flexible but it's the least scalable.
There's a thousand layers to the application of creativity and constraints where, like discipline, constraints breed creativity and creativity can flex constraints. There's a lot to pick on with that for sure.
Well said
I've always been surprised I can't find Kriegspiel on lichess, it seems perfect for online gaming. Great piece!
I think it would lose a lot of its magic on the computer, but it would be relatively easy to program. It’s also underrated as a spectator sport. It is incredibly exciting for spectators who can see everything that’s happening and wondering if the players will find good moves.
Thanks for commenting!
Ah, spectators, awesome point
Leave it to the Germans to take something that is already complicated and make it even more complicated...
lol!
I remember a series of short science fiction stories that featured humans and Martians working together in exploring space. The Martians were very intelligent as a species but had not developed the physical abilities humans have to manipulate their environment. However, they did have telekinetic abilities and the ability to influence human senses, especially sight, so that we could see only what they wanted us to see - sometimes hiding things and other times convincing us see something that wasn’t there. The Martians loved the human game of chess and played it whenever they could while using their abilities to present to human players whatever board situation they wanted. That was my introduction to chess variations.
Seems like a really cool story! What was the name?
https://www.goodreads.com/en/book/show/41093
Actually it’s possible to train ai’s to beat professionals in games with fog of war like StarCraft 2 done by deepmind https://deepmind.google/discover/blog/alphastar-grandmaster-level-in-starcraft-ii-using-multi-agent-reinforcement-learning/
Not sure what point you’re responding to.
I wrote:
“I think you could build a dominant Kriegspiel chess engine, but not to the level of a 700 point ELO advantage.”
Yes, the computers will probably end up beating the best humans at Kriegspiel (if they are not already). But the *difference* in performance will likely be narrower. This narrower differential is because of the uncertainty.
I just feel pretty confident that you would be able to beat humans. AI doesn’t actually know the future. “The best it can do (at this point) is see patterns and make probabilistic guesses” humans can’t predict the future either, so we have to use the information available to us and make educated guesses
I’m glad we agree!
I had never heard of Kriegspiel chess! I find it quite terrifying. But I don't buy your analysis of AI capabilities.
There are AIs already that can be trained to play a game (Go being one of the first done this way) by making random moves, at first. Over thousands or millions of iterations, it gets very very good at the game. This is how I would train a system to play Kreigspiel chess. Pit 2 AIs against each other and just let them go. Eventually, they will reach the level of beating a human every time.
The good news (if you can call it that) in war is that you can only train the machine on simulations. The real thing doesn't happen often enough for the machine to build a workable model.
Hi David, thanks for commenting.
Just to clarify, your position is that a Kriegspiel engine will eventually be able to obtain the equivalent of a greater than 700 ELO rating over the best human players?
The possibility can't be ruled out.
Sure anything is possible. I’m just saying, based on what I know about how chess engines work and the work that was done on building Kriegspiel engines about a decade ago, it seems unlikely that Kriegspiel engines will achieve the same level of dominance that traditional chess engines have achieved. A Kriegspiel AI might eventually beat the best humans, but the *level* difference between the AI and the best humans will simply be narrower.
What you write is interesting, but this "But how would AI fair on a Kriegspiel board where it does not have perfect information? I would submit to you that you could never build a chess engine that is as dominant over humans at Kriegspiel as it is at regular chess." is likely wrong.
NEVER is a mighty powerful claim. I am confident even under imperfect information chess engines would consistentl beat even grandmasters.
The reasons machines can consistently outperform humans are: PERFECT memory, FAST access. Imperfect information changes neither. The computer knows exactly where all your pieces are at start, & minimax STILL works for that reason. Moving pawns to take the center board is correct whether I know or do not know your positions. That is, the correct strategy wins EVEN FASTER if you deviate from it.
Examples may help. I am white boldly move to take center but clever you opens with ... a knights paw. Cool! I have center board!
There are basically only six "good" opening moves, in chess. They are still good even if opponent makes bad moves, even if information is imperfect.
Fischer thought you could get out of this sort of problem by varying the start positions of non-pawns randomly.
If you wish to play chess with less than perfect information simply build a set like stratego. This is reasonably accurate since its semi perfect. You know Where the enemy is but not What he has. Then you neeWhat you write is interesting, but this "But how would AI fair on a Kriegspiel board where it does not have perfect information? I would submit to you that you could never build a chess engine that is as dominant over humans at Kriegspiel as it is at regular chess." is likely wrong.
NEVER is a mighty powerful claim. I am confident even under imperfect information chess engines would consistentl beat even grandmasters.
The reasons machines can consistently outperform humans are: PERFECT memory, FAST access. Imperfect information changes neither. The computer knows exactly where all your pieces are at start, & minimax STILL works for that reason. Moving pawns to take the center board is correct whether I know or do not know your positions. That is, the correct strategy wins EVEN FASTER if you deviate from it.
Examples may help. I am white boldly move to take center but clever you opens with ... a knights paw. Cool! I have center board!
There are basically only six "good" opening moves, in chess. They are still good even if opponent makes bad moves, even if information is imperfect.
Fischer thought you could get out of this sort of problem by varying the start positions of non-pawns randomly.
If you wish to play chess with less than perfect information simply build a set like stratego. This is reasonably accurate since its semi perfect. You know Where the enemy is but not What he has. No referee then.
Your account of war is very likely inaccurate, but there is too little there to accurately account for. First, the problems (domain) is vast, second you haven't defined (your quotes mean what?) "solved".
Your account of war is very likely inaccurate, but there is too little there to accurately account for. First, the problems (domain) is vast, second you haven't defined (your quotes mean what?) "solved".
Thank you for taking the time to leave such a lengthy comment, Eric! I think once you’ve played a few games of Kriegspiel you might see things from a different perspective. You did not refute the central point of my argument: the computer relies on decision trees to make the strongest moves, and it cannot generate those decision trees if it does not know where the pieces are. The lack of perfect information changes the game substantially.
yes it can.
see also:expectimax.
instead of using a known payoff for each node in the decision tree you just use a probability range. even if we decide that i don't get to know which piece killed me i can still infer which ones likely do. i can also infer positions and pieces from trying possible prohibited moves. imperfect information does not mean no information.
Correct, you repeated the same point I made when I said, “For example, if you attempt to make a move and the referee announces that the move is illegal for 3-reasons, the computer may be able to, over time, build an idea of where an opponent’s pieces are. If you discover an enemy piece in an unexpected location, the computer algorithm could make a very good guess about which piece it is, how it got there, and where other pieces are as a result.”