Over the years my friends and I have come up with some variants, with varying degrees of fun/sanity.
1) Start with all pieces off the board. To make a move, you can either put a piece onto your half of the board, or move an existing pieces. Captured pieces cannot be reused (or maybe they can, if you want some sort of bughouse variant).
2) Put a sticker under one pawn on each side. These are the 'assassins'. Be sure to randomise your pawns without looking at their bases. During the game, as a move, you may turn over one of your opponent's pawns - if it is the assassin, you may choose to reveal it. If you do, you now control that piece as if it's your colour and may move it immediately. This of course means you have a 1/4 chance of the opposing king being in check, in which case you can capture it and end the game right there.
3) Ghost pieces. Whenever a piece is captured, remember the square on which it died - it now haunts this square. Any time during the game, a ghost piece may make any capturing move from its square, but instead of capturing the opposing piece, it merely frightens it. Frightened pieces can't move during your opponent's next turn.
4) Tunnels! Cut out some cardboard or paper in a randomish shape to overlay any number of squares in the middle of the board. This represents rock, into which pieces cannot move. The first move a piece makes into the rock weakens it. The second clears the rock out of the way (keep a pair of scissors handy).
Regarding number 4) this seems very similar to the board game Stratego [1] where part of the squares can not be stepped on. Also you only see your opponents pieces when you attack!
I like your (1) and (2), although (3) and (4) seem like the sort of bookkeeping that could only be fun if you had a computer to keep track of them.
> 1) Start with all pieces off the board. To make a move, you can either put a piece onto your half of the board, or move an existing pieces. Captured pieces cannot be reused (or maybe they can, if you want some sort of bughouse variant).
Can a piece be moved into any unoccupied position on your half of the board, or only its usual starting position? Can this be done when your king is in check (in hopes of blocking the check)?
> This of course means you have a 1/4 chance of the opposing king being in check, in which case you can capture it and end the game right there.
I think I don't understand this. The Wikipedia page doesn't seem to describe the starting set-up, but I guess it's the usual one for chess. Are you referring only to the first move? (Otherwise it seems to me that the chance of check depends in an extremely complicated way on the game so far.) In that case, I agree that there's a 1/4 chance of one of the pawns that could check the king being the assassin, but you still have only a 1/2 chance of picking that pawn correctly (even if it's there), so it seems more like a 1/8 chance to me.
You can move to any unoccupied position in your half in (1). We didn't analyse it deeply tbh, mostly we'd just build an ultra-castle of some sort. And yes, you can put in a new piece to block check. We didn't initially realise this, but the king should be the first piece played, or obviously some smart alec will just leave it off the board.
With the assassins, yes it's just a normal setup. But if your first move as white is to turn over the black d7 or f7 pawn, and one of those is the assassin, then you get to take the king. If they're normal pawns, you've spent the move, setting you back in development. Probability-wise I just mean there's a 2/8 chance that it's one of these pawns.
That's not the setup. You don't have two assassins, you just have one assassin, so it's not 8 pick 2. But there are two squares the assassin can be on that give a move 1 mate. That gives the 2/8.
> Probability-wise I just mean there's a 2/8 chance that it's one of these pawns.
Right—there's a 2 in 8 chance that the assassin is one of those pawns, and a 1 in 2 conditional probability, given that one of them is the assassin, that the one you pick is the assassin; so, by 'deconditioning', a 1 in 8 chance overall.
(I think it might make this more plausible—just in terms of probability, not a better rule—to imagine that the rule is changed so that the assassin has to be one of the pawns who can check the king. Then the probability of one of those pawns being the assassin is 100%, by fiat; but you've still only got a 50% chance of picking that one.)
Your variant 1) reminds me of the board game Hive. It feels like a hex-based, lightweight mix between Chess and Go (you win by surrounding your opponent's Queen Bee) where the pieces also create the shape of the board.
I used to play a variant where captures were not announced so the only hint that you lost a piece was when you tried to move it and were informed that that wasn't a legal move
I developed an online version of Hand and Brain Chess to play against a computer. In each move the BRAIN (stockfish engine) will highlight the piece or pieces that are the best move and are the only pieces that you can move. It is your job as the HAND to play the correct move.
At first I thought this was stupid, but after giving it a try it's addicting. It's great to get instant feedback on a move you aren't sure about.
Edit: Is there a variant of this without limit the piece you can move? Basically I just want the instant feedback without getting told which piece to move.
Can someone explain to me why this was called an "excellent move"? https://imgur.com/a/Fyg4PuW I eliminated a night but I feel like the pawn could eliminate my bishop easily. See also the second picture for another (to me) inexplicable situation.
In the first position that pawn can´t take the bishop because then you can move your rook attacking their queen and she is pinned (can´t move because is aligned with the king). The second position you can move the rook, attack the queen, and he must take it. The common value of the pieces are pawn:1, bishop,knight3, rook:5, queen:9 so winning a quen for a rook is an excellent trade.
You can use lichess.com to analyse the moves. Here is the position if you want to take a look https://lichess.org/analysis/standard/r3k2r/pb1pqppp/1ppb4/4...
Also You can copy & paste the generated moves of the game (like 1. e4 d5 2. exd5 c6 )to the pgn text area.
Really cool! But it doesn't seem to be deterministic. I played a "perfect move", took it back, played it again, and it was downgraded to something less than perfect. Also, I took a different move back and was presented different pieces to move. Maybe the engine, given a chance to run longer, updates the list of best moves?
That's interesting, probably a bug in the take back function. There are multiple engine evaluations running in the background at the same time, current evaluation, opponent evaluation and the brain evaluation, so it could be possible that when you take back there is some problem with synchronization
You can play kriegspiel here at a site that I created and keep alive: http://krgspl.com
I built it in 2013 and am in the slow process of improving it for a relaunch at some point.
I’m of the opinion that this game is a step up from chess, go, and poker, in that current deep learning tech can’t be used to build a competitive engine, due to the lack of information available.
Hill climbing or simulated annealing would work fine, wouldn’t it? You are constantly looking for a goal that maximizes the number of times you can check.
Check != Checkmate. Also you need to find the king before you can do that. And all the other pieces that protect the king. The problem is far more complicated than "throw this algorithm at it".
I recommend playing several games with someone else, and then rethinking your assessment.
> current deep learning tech can’t be used to build a competitive engine.
I’m sorry but I disagree with this point since the process of asking about pawns, knowing about the state of the game (checked, unchecked or checkmate) and knowing if a move is legal can be used with reinforcement learning to accurately model the unknown environment and build a play strategy.
I'm familiar with reinforcement learning and having done some private research in the subject, I disagree with your overall approach. While you might be able to make a basic model you won't be able to reach a global optimum with the vast variation in play you'll have after a dozen moves or so.
Deep learning can't really guarantee to find a global optimum in most nonconvex problems. However, this chess variant is definitely not something that is likely to pose serious issues to modern systems. DeepMind's StarCraft 2 AI does just fine handling vast variation and imperfect information.
> "this chess variant is definitely not something that is likely to pose serious issues to modern systems"
How do you know that? It must be experimented and proven. It's not something that can be handwaved. With machine learning, I would say "not possible until you prove beyond a doubt".
It's a different problem. SC2 is a realtime strategy game, and Kriegspiel is turn-based. The information asymmetry is different in both games, and the goal is different.
In SC2, you can perform realtime monitoring of changes and track them with multiple maneuvers all at once. In kriegspiel you cannot.
Also in SC2, when you see something, you know what it is. You don't know what kind of pieces are where in kriegspiel. Mistaking a queen for a pawn in kriegspiel can be a game-loser.
From a machine learning model’s perspective they are both turn based. In SC2/Dota case one turn is one frame. There is nothing fundamentally different between these and blind chess apart from the possible moves space being vastly smaller
As noted above, the other differences are that you don't know what type of piece you've located (if you've even managed to locate one).
So taking your analogy, and saying an AlphaStar game lasts 5 minutes, at 60 fps, thats 18000 frames. Kriegspiel games average about 50 moves total (or 100 half moves).
So lets tally these up: SC2 has 18000 frames where on visible turf, full knowledge of occupant enemy pieces are available. Kriegspiel has 100 frames where there is no such concept as visible turf of enemy pieces.
Deduction of if and which piece is occupying a square is the only way to have knowledge of the enemy, and the probability for the deduction to be correct drops significantly after each move.
With such a small amount of information available, the problem becomes significantly different.
The key to successful machine learning is available data. There is so little data that the model you are training will collapse or overfit quite quickly and become useless.
The first versions used what you might call 'camerahack', it just sees the entire map at once but as the player himself would see (i.e. fog of war included). The latest version has camera control, only seeing one screen at a time like humans.
A Programming and Problem-Solving Seminar led by Chris Van Wyk and Donald Knuth in 1979 was, among others, dedicated to programs playing the KRK endgame in Kriegspiel [1].
In my previous life as a teaching assistant, I co-led a seminar that borrowed this problem from Knuth. At the end, our students' White and Black bots played a tournament. The feedback was: "since Black loses this endgame sooner or later, those who developed bots playing White had more fun". Next year, we came up with a symmetric game.
Blindfold chess? Just search for it on youtube. Most GMs can do this easily. Some simultaneously play 10+ sighted players whilst blindfold. Even when not blindfolded they often look away from the board whilst calculating so that the current state of the board doesn't distract them.
It's not exactly blindfold. With blindfold, the blind player still knows the opponent's moves...They just don't have the luxury of being able to _look_ at the state of the board (although top-tier chess players don't need that luxury).
This is a little different, but I think/speculate they would still have enough heuristic knowledge to crush most people.
Suckerpinch (Tom) recently had a video[1] about predicting the position of the oponents pieces from knowning just white. I don't think he's aware of this. Great video anyway.
He trained an NN to predict the pieces and then used Stockfish on the resulting position.
Yeah, the "pawn tries" rule in particular reminded me a lot of his "blind spycheck" algorithm in that video. If you are allowed to rank every move and perform the first legal one, you can get away with some interesting stuff.
An intricate and lengthy account of several different computer chess topics from my SIGBOVIK 2019 papers. We conduct a tournament of fools with a pile of different weird chess algorithms, ostensibly to quantify how well my other weird program to play color- and piece-blind chess performs. On the way we "learn" about mirrors, arithmetic encoding, perversions of game tree search, spicy oils, and hats.
Infochess "is a chess variant designed to simulate the relationship between what is known and what remains unknown in conflict, and to stimulate a deeper appreciation of the interaction between the informational domain and more traditional military affairs." It includes abstracted psyops and electronic warfare.
Having played a number of those, IMO they are nothing like Kriegspiel, which is in another category. I grew up playing chess and at our club and we also played many variants. Kriegspiel was a whole other monster that we would play in teams, one group sitting in on room, and the other next door so you had to whisper and show experiment moves quietly in the group. The arbiter would register on a central board and announce. The experience is difficult to recreate on the web, and I know Lichess hasn’t added Kriegspiel to their variants list even after several requests to do so.
As I was reading the description, I immediately felt that a online version of it would be much simpler and quicker to play than real world version. No need for umpire and multiple boards (handled by game server), all the announcements could be automatic and visual, immediate feedback on blocked moves etc.
Yes, it is! You cannot allow games-in-progress to be viewed (to prevent cheating). So how can you allow a team to be all viewing the same side of the board and prevent spies? I haven't figured out that solution yet.
Yes, thanks for making that note. I've heard it's how/why KriegSpiel was invented in the first place - as chess was a poor substitute for practicing war strategy.
What about scouting? You could have a rule, that you are informed about the pieces in the vicinity. Perhaps even have different view distance for different pieces. Or only reveal the existence but not the type of a piece for longer distances and then type when closer.
In krgspl.com I implemented this as a 'reveal square' rule for any square adjacent to one of your pieces. After playtesting we settled on allowing it once per move, and it would only reveal whether the square was occupied or not, without revealing the occupant piece type.
There's a similar game concept (not based on pieces but rather on decisions to reconnoiter) called reconnaissance blind chess[1]. My understanding is that this game is largely used to explore and experiment with AI decision processes in situations with imperfect/decaying information.
I invented a similar chess variant that doesn't require a referee.
All the pieces are one color, but on the bottom of the pieces you mark the true color (black or white).
When it's your turn, you can either move, or challenge your opponent that their last move was false. If you challenge, they turn over the piece they just moved. If it was their color, you lose your turn. If it was your color, they move the piece back and you get to move.
Don't even get me started on Go Chess, a hybrid of Go and Chess played on a chess board with both Go and Chess pieces, where in the Go world even Chess pieces are treated as Go pieces.
So with a small bit of short term memory this is equivalent to chess? That's not really a similar variant to the hidden-information game that is Kriegspiel.
Seriously awesome information based system design way before computers, it seems like such an advanced game/simulator. One of the things I'm most interested in is how sophisticated past humans could get with their tech & circunstances they had at their time, this is one of those to me.
"'Hell no' (or "Impossible"), when the attempted move is always illegal regardless of the opponent's position. For example, moving a bishop as if it were a knight."
It would be cool if there were a variant of this variant where one also had to discover 1) what the legal moves were, and 2) whether there is an opponent there at all.
We have research lab at CTU in Prague where our focus is on imperfect-information games. We are working on scaling up algorithms that have worked well in Poker, and extend them to other domains as well. Poker has some nice properties - the hidden information (player's private cards) does not change through the course of the game so the value functions needed for neural networks can be computed "nicely", as they have fixed input/output sizes.
We definitely see Kriegspiel as one of the next challenges.
In my thinking, if there was some RNG added to the turn by turn play of chess, with means of improving the odds through tactical play, then it would be the perfect 2 player board game.
Otherwise, it's almost impossible to fight back and win a chess match through audacious, if risky play, in normal chess. Once a person is in a position of advantage, they almost always win, unlike say in backgammon, where the player at disadvantage can deny victory to the other, for sometime, during which they might figure out some striking counterplay.
There are a number of games which amount to some random element added to standard chess, one example is Knightmare chess: http://www.sjgames.com/knightmare/
I seem to remember a simpler variation involving rolling dice to determine what pieces are allowed to move, but I can't seem to find anything about it.
In the rules I used to play by at chess camp, you had to ask "are there any pawn captures", and if there were, you had to make one--this removes the additional information issue.
My daughter invented one version with regenerating pawns. I play regular chess on my side, and on her side, she has the king, queen, and 8 pawns. On her turn, she can choose to move a piece as normal or put a pawn on any square on her half, but only if she's had a pawn captured (so off the board).
It creates a strange desire for her to get her pawns captured so she can plop it down somewhere else, and a strange desire for me not to capture any pawns.
There is not much that I remember from 5th grade, but what I do remember is that after half a year of teaching us chess during Chess Club, our teacher has beaten all of us simultaniously while facing the wall and announcing his moves after we announced ours.
It's rare to see that much expertise and skill in action.
Is AI as good at Kriegspiel as it is at regular chess?
Is there a Go equivalent? If so, is AI any good at that?
And, if either answer is no: then why did DeepMind go straight to trying to solve a game that was both incomplete-information and realtime (Starcraft) instead of merely incomplete-information (Kriegspiel)?
Kriegspiel is a very niche game. There are no professional players and it's not clear what it would mean to achieve superhuman performance.
At one point Stratego was the standard for incomplete-information (but non-random) games in AI research. That may no longer be considered challenging, but either way, crushing an AI tournament in that wouldn't attract the level of publicity DeepMind get from chess, go or Starcraft.
On an extremely tangential note: If you need some fun, short shows to relax, watch White Gold on Netflix. Stratego plays a small part in the second series.
Nein danke. I cannot envision the "future board" well enough to play NORMAL chess very effectively. I cannot imagine not being able to see the enemy pieces at all.
Thanks for sharing. The whole time reading this, I was thinking how straightforward it would be to re-factor online/electronic chess games to also support a Kriegspiel mode. I'm surprised I have not seen the option.
What I don't get about hackernews is why regularly someone just dumps a link to a random wikipedia article with no context of why it's here and why it should be interesting. Can we just change this and add some context?
1) Start with all pieces off the board. To make a move, you can either put a piece onto your half of the board, or move an existing pieces. Captured pieces cannot be reused (or maybe they can, if you want some sort of bughouse variant).
2) Put a sticker under one pawn on each side. These are the 'assassins'. Be sure to randomise your pawns without looking at their bases. During the game, as a move, you may turn over one of your opponent's pawns - if it is the assassin, you may choose to reveal it. If you do, you now control that piece as if it's your colour and may move it immediately. This of course means you have a 1/4 chance of the opposing king being in check, in which case you can capture it and end the game right there.
3) Ghost pieces. Whenever a piece is captured, remember the square on which it died - it now haunts this square. Any time during the game, a ghost piece may make any capturing move from its square, but instead of capturing the opposing piece, it merely frightens it. Frightened pieces can't move during your opponent's next turn.
4) Tunnels! Cut out some cardboard or paper in a randomish shape to overlay any number of squares in the middle of the board. This represents rock, into which pieces cannot move. The first move a piece makes into the rock weakens it. The second clears the rock out of the way (keep a pair of scissors handy).
Some of this seemed like a good idea at the time.