Two amateur chess players guessed the best sequences of WBBWW and then plugged the result of their guesses into an engine to statically analyse the position. There is absolutely no chance that they found the best options for either player.
The sequencing argument that they make has no real foundation. Introducing double moves could have all kinds of side-effects which they later mention when showing their guesswork-variations but skip over when trying to make the previous "logical" argument.
The only semi-useful sentence in this paper is the following:
> More light would be shed on this question if the two leading machine-learning chess programs, AlphaZero and Leela Chess Zero, were taught to play with our proposed change in the order of the 3rd and 4th moves from White-Black to Black-White.
The rest can be summarised as "We thought of this potential rule change and it seemed to us to be more even for black."
If they had wanted to then there are only around 10,000,000 possible game-states after the first WBBWW and potentially looking at computer evaluations of the whole tree of possibilities could be semi-interesting and definitely very achievable. Probably better and easier to guess instead though to be fair.
Lichess has an API endpoint for their cached analysis, so they could have easily automated at least some of the new tree.
Or they could have inserted a couple of `if` statements into Stockfish, recompiled, and simply analyzed the root position.
Even calling this a "paper" is a stretch. I'm usually against twitter post threads as content but that seems about the level of analysis that's actually present here.
> If they had wanted to then there are only around 10,000,000 possible game-states after the first WBBWW and potentially looking at computer evaluations of the whole tree of possibilities could be semi-interesting and definitely very achievable. Probably better and easier to guess instead though to be fair.
This is what I was expecting them to do from seeing the abstract. Instead they just examined a handful of novel lines and evaluated them with Stockfish. An interesting proposed rules change, but a rather low quality analysis of it.
> More light would be shed on this question if the two leading machine-learning chess programs, AlphaZero and Leela Chess Zero, were taught to play with our proposed change in the order of the 3rd and 4th moves from White-Black to Black-White
Also, there has been plenty of exploration around this in the recent past. "No-castling chess" was briefly popular in 2020 following some investigations by DeepMind and Vladimir Kramnik that found that training AlphaZero from scratch to play this led to more exciting games. [0]
They didn't try this specific rule, or didn't publicize their results if they did. It's too different from chess to be of intrinsic interest to serious chess players. Maybe it's interesting as a variant in its own right! But this is the worst imaginable investigation of whether that's true.
To me it seems as if this paper's conclusion is hurt by its data selection (of openings). It uses Stockfish and examines openings like the Ruy Lopez and Queen's Gambit to evaluate the game. But using this new gameplay style, presumably new openings would emerge as predominant openings with the current very played (at a high level) openings potentially falling out of favor.
Many of the famous openings that are still used by top players were elaborated and built over centuries. Stockfish is pretty incredible but it can't solve chess (yet). Meaning, that if you only examine the current most popular openings you can only draw a conclusion based on those start points but you have potentially omitted what could be new ones. Potentially, with this move ordering there may be an opening for white or for black that is extremely imbalanced, but the move ordering to get there is currently beyond stockfish's depth from the start position (or any starting positions they tried). Unless I misunderstood their methods.
I think maybe you misunderstood their methods? They explicitly discuss the most obvious "novel" openings that wouldn't transpose to traditional openings, like 1. e4 d5/dxe4 2. Nc3/Nxe4, or 1. d4 c5/cxd4 2. c3/cxd4.
Note: after 1. e4 d5/dxe4 2. Nc3/Nxe4 Qd5, White's only option to maintain an advantage is to play 3. Nc3, transposing into a mainline Scandinavian. On the other hand, 2. ... BF5 3. Ng3 Bg6 effectively gives black a Caro Kann two knights variation(1. e4 c6 2. Nc3 d5 3. Nf3 dxe4 4. Nxe4 Bf5 5. Ng3 Bg6), but with the moves c6 and Nf3 omitted. This helps black more than white, since c6 hampers the development of Black's knight. As white, I would probably attempt something like 1. e4 d5/dxe4 2. c4/f3, sacrificing a pawn for development and hoping to play d4 to get a big centre(e.g 1. e4 d5/dxe4 2. c4/f3 exf3 3. Nxf3 Bg4 4. d4 Nc6 5. d5 Ne5 6. Be2 Nxf3+ 7. Bxf3 Bxf3 8. Qxf3 Nf6 9. 0-0 when white has a lead in development, central control and pressure along the f file for the small price of one pawn(though I'm sure black has a better line than this)). 2. d4/d5 might also be interesting, depending on whether en passant is legal afterwards :P
Because I find the mainline Scandi to be dreadfully boring, I've been experimenting with playing 2. Nc3 with white. The tempting 2. ... d4 equalises but leads to a more complex game, with white's night rerouting to g3. Black's will often seize the centre with e5 and c5, but white gets attacking chances with Nf5 ideas as well as playing for the f4 break. The frustrating downside is that black can just play 2. ... dxe4 and force me to pick between the scandi I'm trying to avoid and a Caro Kann where black has already equalised...
Yeah, against e4 why would black play anything but d5 followed immediately by dxe4? That would basically turn 1. e4 into the Englund gambit (an awful opening starting with 1. d4 e5 2. dxe5) with reversed colors, giving a clear advantage for black. The next two moves for white would let him equalize, but still the Ruy Lopez is unlikely to occur.
At high levels of play and in computer-vs-computer games, the problem is too many draws. This paper's idea would exacerbate this problem. If you want a fairer outcome, just do what's done already and count one match to be two games, one with white, one with black, and five possible outcomes.
One interesting variant that I would love to see tried is no-black castling (or no-black-short-castling) chess. Draws would count as wins for black (called Armageddon scoring). You could still play two games (one per side) to get the most fair outcome. This starting position should still allow a large variety of openings and it would always be balanced on the edge without the large dead zone of draws. All chess programs can already play this variant as well, though maybe not most optimally.
Chess960 is also a pretty good solution that reduces the value of memorizing openings.
> Other expert programs, including Leela Chess Zero and
Stockfish, when pitted against each other in the superfinal of the unofficial world computer chess championship (TCEC), give White even greater odds of winning, but the outcome is still a draw in the large majority of games (see https://tcec-chess.com).
Despite the fact that computer programs start play from 50 preselected opening positions in the TCEC superfinal (once as White and once as Black), it is remarkable that Black has not won a single game in the last two TCEC superfinals
This line is very dubious. TCEC SuFi openings highly promotes one side over other as otherwise it was seen that almost all matches ends in a draw. It could very well select 1. g4 and black will win all matches. Also there had been an incident where Leela won an opening both with black and white pieces and it happened only once in TCEC history.
As a casual chess player I'm not sure more drawish chess is something we want? I know there has been some exploration into a chess variation without castling which apparently can lead to more vibrant aggressive chess due to the king being stuck in the center of the board.
> Some players, including world champions such as José Raúl Capablanca, Emanuel Lasker, and Bobby Fischer, have expressed fears of a "draw death" as chess becomes more deeply analyzed.
> To alleviate this danger, Capablanca and Fischer both proposed chess variants to revitalize the game, while Lasker suggested changing how draws and stalemate are scored.
The 1984 championship, which was played with a first to 6 wins format, had to be abandoned due to draws. It started on the 10th of September 1984, and was abandoned on the 15th of February 1985 after 40 draws and only 8 decisive games.
The result was controversial (the championship match would be restarted), and the other alternative formats that have been used are also controversial. Such as the reigning champion retaining the championship if a decisive result isn’t produced, and the current format where they play classical games, then rapid games if it’s tied, then blitz games if it’s still tied, then an “armageddon” game if it’s still tied. The most recent championship match was decided by rapid games, which a lot of people basically considered to be a non-result.
This has always bothered me about soccer. 90+ minutes of somewhat complex team play, positioning, footwork, and endurance… and at the end of it the winner is decided by penalty kicks, which feel like an entirely different game.
I'd like to see such tiebreaks, in both soccer and chess, moved to before the game (soccer) or match (chess) instead of after.
The effect is to give one side draw odds in the regular game or match. The other side, knowing that a tie means a loss for them, will have to play more aggressively for the win. This should lead to more decisive games.
Balancing the one side playing aggressively for the win is the fact that the other side can "aggressively" seek a draw. The "if-we-draw-I-win" side is now way more motivated towards seeking out drawish positions.
the problem with soccer is that good ideas to improve the game (e.g. make the time stop like in Basketball: Always when the ball is in play the clock ticks, when the ball is outside stop the clock. This would mean less players on the ground running out the clock with fake injuries) are not implemented due to how the leagues are organized (UEFA/FIFA etc.). Whereas in the NBA every year they change the rules, e.g. the best rule change came a couple years ago where they put the shot clock reset of an offensive rebound to 14 seconds instead of 24 seconds - faster gameplay. Everyone wins.
Would also love different overtime ideas played out in soccer, e.g. put players out of the game. When it's 8 vs 8 maybe that leads to more goals in overtime or something... also, there was a time/place where penalties were shot differently (with running with the ball like in hockey), so there exist opportunities to change but as far as I see it there's an unwillingness.
Golden goals and silver goals have all been tried, to disappointing effect. The Champions League still uses two-legged ties. Where's the unwillingness?
'the problem with soccer' implies that there's a fundamental problem preventing it from working rather than the reality of it being the most successful sport in the world, regardless of what fans of American sports enjoy and in spite of corrupt leadership and perverse financial incentives.
The issue that I see with all the tie break resolutions is that you end up deciding the outcome of the game, based on something other than playing the game. Rapid chess isn’t classical chess. It’s a different game, with different strategies, and players are rated differently between the two formats. 8v8 soccer would be similar in this respect.
This is simply the nature of games where you need a decisive advantage to turn a draw into a victory. In high level chess you need a decisive material advantage to win, and in soccer, a game where both teams scoring 0 points isn’t uncommon, you also need a somewhat decisive advantage to win.
But these are also both games that are very simple and well balanced. I think they’re better without additional complications, and the best solution is simply to extend play. In the 1984 chess championship I mentioned, the match was abandoned by the organisers. Karpov and Kasparov were both happy to continue playing for another 5 months if that’s what was required.
> The issue that I see with all the tie break resolutions is that you end up deciding the outcome of the game, based on something other than playing the game.
I think that’s the point. You had your chance to win the game and it took too long, so now you play a shorter, modified version of the game to determine a winner.
The same criticism could be leveled at 8v8 soccer that “it’s not soccer”. Your team depth matters less and your stars matter more.
It would create injuries to “extend play”, not to mention creating boring 2 or 3 hour long 1-0 matches.
Used to be a coin toss but penalty kicks are nicer to watch. But yes penalties are a completely different game. That's completely assumed however. You can't really expect player to keep going after 120 minutes and you need a way to decide which team goes through.
actually, a coin toss is a better idea as I guess it's something none of the teams wants, encouraging them to press harder and take more risks. Actually they tried all kinds of things: golden goal, silver goal, replays (FA cup, fe)
That's very much the minority of soccer games. If those are all you've seen, catering to you is less important than catering to the billions of active fans.
Here is a link https://arxiv.org/abs/2009.04374 to the paper you are probably talking about. Coming from DeepMind (and with Former world champion Vladimir Kramnik as a coauthor), they studied a bunch of chess variants by training AlphaZero to play on them. In addition to no-castling, they also studied variants where pawns can always move two steps forward, and where it's possible to capture your own pieces.
To a casual bystander (who at least knows the rules), castling seems a bit arbitrary and so do the other "extras" - en passant et al.
What would happen if chess was reduced to purely "normal" piece moves? What happens if pawns lose their initial two square move? Obviously it would reduce the space of potential games by quite a lot but would it reduce chess to draughts/checkers?
> To a casual bystander (who at least knows the rules), castling seems a bit arbitrary and so do the other "extras" - en passant et al.
Those rules were added in to make faster games. Taking them out would slow the game down, and we've at least a hundred years of the new rules (faster games) being preferred.
> Would you mind pointing me at a canonical source on that please. Chess is way older than a few centuries, so why do the extra rules sidle in and when?
It's literally on the wikipedia entries, respectively:
https://en.wikipedia.org/wiki/Castling
"Castling was added to European chess in the 14th or 15th century and did not develop into its present form until the 17th century. The Asian versions of chess do not have such a move. "
"Castling was added to allow the king to get to a safer location and to allow rooks to get into the game earlier (Davidson 1981:16). "
https://en.wikipedia.org/wiki/En_passant
"The motivation for en passant was to prevent the newly added two-square first move for pawns from allowing a pawn to evade capture by an enemy pawn.[14]:16 "
The two-square pawn move is commonly regarded as being introduced in order to make the game faster/more dynamic.
> Chess is way older than a few centuries, so why do the extra rules sidle in and when?
Not exactly what you were asking, but you may be shocked to learn that Chess has had two significant patches very recently - namely clarifying the rules to prevent promoting a pawn to an enemy piece (sometimes causing a smothered mate), and to prevent castling with a pawn promoted to a rook on the 8th rank (annotated humourously as 0-0-0-0-0-0!!). These patches were seemingly implemented in the late 19th century and 1970s respectively, although it's surprisingly hard to find out precisely when.
Obscure, but also completely to do with how the pieces on the board can actually move, which I bet most people would have suspected that we worked out a few centuries ago.
The decision of who goes first doesn't really matter, it's a purely aesthetic choice. Similarly there's a lot of regulatory changes to this day around things better described as "playing conditions". But banning 0-0-0-0-0-0 is fundamentally the same category of change as saying "oh and also the queen can move like a knight now", which seems like an absolutely ridiculous thing to have still been refining in the 1970s.
There's no way they would have allowed 0-0-0-0-0-0 or "promote to opposite color piece" in a real game if GMs tried it in 1965. It's only ambiguous wording of the rules that was fixed in 1970, and not actually a capability of pieces.
IRL games don't blindly execute the rules like video game code: in rare cases it's possible for the rules are poorly worded in a way that would allow something stupid, and yet that wouldn't be actually permitted play by any reasonable player or arbiter.
It's happened several times in MtG tournament history for example, where some mechanism was so clearly against the intended rules that the judges just didn't allow it even though it would have worked as a strict reading of the rules (nonsense about actions "between turns" and I recall Yawgs Will exiling as a triggered effect both just getting shut down by judges).
The two things you mention are solely gags for funny chess puzzles about what is an obvious oversight in some specific written rulesets, not actually something that would have actually been legal play.
> in rare cases it's possible for the rules are poorly worded in a way that would allow something stupid
I would not describe either of the chess examples as "stupid". "silly", sure, but neither of them is really any more ridiculous than en passant is.
> It's happened several times in MtG tournament history for example, where some mechanism was so clearly against the intended rules that the judges just didn't allow it
I feel like it's fair to make a distinction between a game like MtG where interactions and synergies are sufficiently complex enough to make those weird edge cases requiring an arbiter an inevitability, and chess, a game that's surely simple enough that it's reasonable to expect the rules to cover all on-board actions.
That said, if you have any specific examples here I'd love to hear them. I know of some absolutely batshit things that are technically possible in MtG, but I'm not sure I've heard of any stories of them happening in ostensibly legal tournament play.
(Since we're discussing absurd rules lawyering in TCGs now, you may be interested in the legendary 3000 card Yu-Gi-Oh! deck that was once played in a tournament setting)
The two cases MtG I know of off hand are to do with accidentally broken rules updates. One was when the added a moment "between turns" where you could activate abilities but not cast spells (added to make Time Vault work). The interpretation that you could activate wall of roots infinitely is similar to the chess cases you're pointing at ("once per turn doesn't constrain what I do between turns"). The other was Yawgs Will exiling effect was briefly a triggered effect instead of a replacement effect which let Yawgs Will + Dark Ritual also be a trivial infinite, which wasn't even rules lawyering it just was so clearly broken and unintended that Arbiters didn't allow the official rules and eratta to ride.
But really I don't think any of the rules of chess ever unambiguously allowed either of the things you said, they are both extreme rules lawyering primarily known for their humorous value, and the rules being clarified they aren't allowed doesn't prove they were allowed before. E.g. 0-0-0-0-0 claimed a pawn promoted to a rook is "a piece that hasn't moved", and easily could have been rejected as saying it is the same piece that moved while it was a pawn.
En Passant is widely know and explicitly specified in the rules, taught, and played by even low level players, so there's no way it's similar.
I think it's impossible to know unless you can find a case where someone actually attempted the moves, but I feel pretty confident they just wouldn't have been allowed and if you tried a similar gag at a rated chess event today the arbiter would just tell you that you're crazy.
I don't think we want more drawish chess but ideally we want the outcome of the game to be a function of the players' skill and not of who had the white pieces.
> In mathematics, the Thue–Morse sequence is the binary sequence obtained by starting with 0 and successively appending the Boolean complement of the sequence obtained thus far. The first few steps of this procedure yield the strings 0 then 01, 0110, 01101001, 0110100110010110, and so on, which are prefixes of the Thue–Morse sequence. The full sequence begins 01101001100101101001011001101001....
Something somewhat similar happens in the game of Connect6.
Connect 4 and Connect 5 (aka Gomoku) significantly favor the first player, but Connect6 follows a BWWBBWW... pattern where each player places two stones per move expect for the first move where only one is placed. It ends up being very balanced.
I think a better way to state this would be: player A makes white's first move, then player B decides whether to play white or black. Play proceeds as normal with black to move.
But tbh it seems a bit silly; at least TFA realizes there at least naively seems to be more at stake in moves 2 and ..2.
As a chess variant fan, this is a cool idea that has a lot of potential. I would like to see it extended to include say, the first ten moves. Black may announce a double move any time during the first ten moves after which white gets his double move and then play continues with single alternating moves as usual, with the sole restriction that mate may not be delivered by a double move.
white gets the first tempo on the first move, black gets the first tempo on the second move, and then white gets the first tempo on remaining moves, and black's chances are improved.
I wonder what it would be like to continue the pattern? it would drastically alter the game as there would be many more opportunities to take pieces, but if it could be grokked it might be fun.
all the variants that breath a little new life into the game can be fun for semi serious players, I find.
One of the most amusing Chess variants I learned as a kid is the one where white gets the full complement of pieces, and black gets only a king, with the proviso that black gets two consecutive moves and can move through check on the first move. This means that black can capture two pieces in a single turn. While white still wins with optimal play, less experienced players playing white will get absolutely wrecked.
I’ve been saying for some time the first mover advantage in football penalty shootouts should be removed by doing a Team A/Team B B-A A-B etc. as they do in tennis tiebreaks. I’m not sure if Chess would ever make such a change.
”As part of a trial to reduce a potential first-mover advantage, the IFAB sanctioned in March 2017 to test a different sequence of taking penalties, known as "ABBA", that mirrors the serving sequence in a tennis tiebreak (team A kicks first, team B kicks second)
[…]
During the IFAB's 133rd Annual Business Meeting in Glasgow, Scotland on 22 November 2018, it was agreed that due to the lack of strong support mainly because of its complexity, the ABBA option would no longer be used in future competitions.”
I wouldn’t know whether that decision was the best possible, but guess that the popularity of football makes them more resisting to change than other sports (“Never change a winning team”)
The football (soccer) offside rule is often held up as an example of complication, so I doubt a reordering of penalty kicks will be too hard to follow.
> The football (soccer) offside rule is often held up as an example of complication
I don't follow the sport much - what's so complex about it? It always seemed fairly simple to me, especially in comparison to other sports' more complex rules (good luck explaining the Duckworth-Lewis-Stern system to someone).
On the average game of chess white scores 0.55 points and black 0.45. In the age of computers it should be pretty trivial to award black 1.2 points for a win (or some such appropriate number) and vastly improve balance.
The first player chooses white's move. The second player chooses whether to play as black or to take over playing as white.
Problem solved. The first player is incentivized to make white's first move as value neutral as possible. It's exactly the same principle as the "sister cut's the cake, brother chooses which piece to take" trick many will have used as kids.
It's also trivial to make chess fair by agreeing that every match ends in a draw. The concern is in how we can make it fair without making it less interesting. The cake-cutting solution would severely reduce opening variety.
I fail to see how this would be the case. I think it may even increase opening variety.
I think you'd see many more opening with oddly position a-c/f-g file pawns, or horses thrown into the rail. I think it could make for a radically different game opening game that would be just as diverse as the current game. Perhaps more so.
I wonder if you misunderstood? Since this is very far from agreeing to a draw at the outset. I meant the cake cutting for only the very first move. Not the subsequent moves. Only the very first move played on the board is incentivized to be value neutral.
> I wonder if you misunderstood? [...] I meant the cake cutting for only the very first move.
No, I got that.
> I fail to see how this would be the case. I think it may even increase opening variety.
At a high level players will be relatively comfortable with playing at least one opening from any (sensible) opening move, so swapping would offer very little disadvantage in terms of not being prepared. So the decision on whether or not to swap is almost entirely based on how good that opening move is.
Now, white has a significant advantage on just about every opening move that's currently played with any regularity. Therefore black will almost always swap, and gain that same advantage. So it's in white's interest to not play those moves, and instead pick whichever moves are as close to 50/50 as possible, so that their opponent cannot get an advantage.
The crucial thing to note is that most possible moves white can make are either good (preserve a meaningful advantage for white) or bad (immediately gives advantage to black). There are not many first moves that result in a 50/50 game. Therefore the number of viable first moves is decreased.
They're certainly different first moves, and we'd see a lot of variety for a year or two, maybe even up to a decade as people play and learn entirely new lines. But in the long run the number of opening moves that don't immediately give up an advantage would be reduced.
It begs another question though, is chess a worse game if white begins with a pawn on a3 instead of a2 (assuming this is the result every time the suggested procedure is used), and black moves first? Maybe. But I think it's an okay solution if you want a neutral platform for a tie-breaking game.
I just checked and "horses thrown into the rail" (by which I assume you mean Na3 or Nh3) would still never get played because they're terrible for white; stockfish puts them at -0.5 and -0.7 respectively, compared to the white advantage is only +0.2.
Two amateur chess players guessed the best sequences of WBBWW and then plugged the result of their guesses into an engine to statically analyse the position. There is absolutely no chance that they found the best options for either player.
The sequencing argument that they make has no real foundation. Introducing double moves could have all kinds of side-effects which they later mention when showing their guesswork-variations but skip over when trying to make the previous "logical" argument.
The only semi-useful sentence in this paper is the following:
> More light would be shed on this question if the two leading machine-learning chess programs, AlphaZero and Leela Chess Zero, were taught to play with our proposed change in the order of the 3rd and 4th moves from White-Black to Black-White.
The rest can be summarised as "We thought of this potential rule change and it seemed to us to be more even for black."
If they had wanted to then there are only around 10,000,000 possible game-states after the first WBBWW and potentially looking at computer evaluations of the whole tree of possibilities could be semi-interesting and definitely very achievable. Probably better and easier to guess instead though to be fair.