Two of the many problems with chess is that the white pieces have the advantage and that most games end up as draws.
Wikipedia says:
In the last five matches in Adorjan’s survey, all between Anatoly Karpov and Garry Kasparov, White won 31 (25.8%), drew 80 (66.7%), and lost 9 (7.5%), for a total white winning percentage of 59.2%.
Now, I don’t think it’s fair to refer to 31 vs 9 as “59.2%”. That looks more like 77.5% to me.
What they do is that a win is 1pt for the winner, and a draw is a half point for both players, and so it looks like 31+40 vs 9+40, which is the 59.2 they were talking about. To me, this is misleading when comparing white’s chance to win vs black’s chance to win. We don’t want to dilute the statistics with a bunch of drawn games. It’s not a good that two thirds of the games are drawn.
In 2017 AlphaZero, playing 100 games against Stockfish, won 25 and drew 25 as White, but won 3 and drew 47 as Black.
So with my way of looking at it, that’s a 89.3% advantage for white. Even with their draw-diluted perspective it’s 58% advantage for white.
In chess, there are seven ways to draw.
So we have two problems: too many draws, and, not enough black wins. One of those problems can be the other problem’s solution and vice versa. Give some or all of those seven ways to draw as alt wincons to black. RIP Gordian knot. Call it “black holds the fort” instead of a “draw”.
Since there’s a point system it doesn’t even have to be a full 1 point win.
If a full win is 1 pt to either player but holding the fort is a half point that only goes to black, suddenly black has a huge advantage. 49 vs 31 (61.25%) in Adorjan’s study, 39 vs 25 (61%) in the AlphaZero vs Stockfish series. That’s already more fair than the 77.5% or 89.3% advantage white had before, but, we can do better.
FIDE has access to a lot of data of human vs human games across all levels of play, or you could look at the latter AI vs self games, to determine a constant instead of .5 that makes black and white equal. Just as a quick example, with the AZ vs Stockfish series, giving black 0.305 points (and white 0) on holding the fort would make black and white equal, while with Adorjan’s study it’s 0.275 points instead.
Basically for any dataset it’s just this formula: b+xd / w+b+xd = 0.5, where b is number of black wins outright, d is number of drawn games, and w is number of white wins outright, and just solve for x.
When the goal of the game changes, that of course reverberates throughout play, so of course both black and white would play differently, very differently, when a rule such as this is in place. That is the point!
Of course you’d select a nice round number. For example instead of 0.305, a win for either could be 10 pts to that player, while holding the fort would be 3pts for black.
In baduk (a.k.a. go) something similar was introduced in the late 1930s, which in Japanese is called 込み (“komi”, “compensation points”). In baduk the black pieces move first and have the advantage, so white was given 4.5 points as compensation (after playing the game for thousands of years without such a rule). This was increased to 5.5 in 1973 and to 6.5 in 2002.
It’s rare, but white might get in a position where they can’t win outright but can choose between black getting a full victory or just black holding the fort. In that situation, white can’t affect her own tournament standings but can help or harm her opponent. That’s not great, it’s going to lead to kingmaking, collusion, or just bad feel situations.
I asked for patches, and Richard sent:
Your issue of white throwing games is likely helped by making the score zero sum. So +10/-10 or +3/-3. There will almost always be opportunity to abuse tournaments. So in standard chess if I can draw or lose and I don’t care about my rating I might choose to lose since I am out of the prize pool anyway and I can get my opponent into the top 16. Very common issue. But - if you care about your score the difference between 0 and 1/2 will generally have you play honestly, and presumably that would be true if you made your system zero sum.