With 10 more games Mission impossible for Komodo Dragon

[Chessqueen]

10 more games Mission impossible for Komodo Dragon to break even, Komodo Dragon might win 3 more games out of the 10 left

Note: If TCEC decides to experiment where the side that has White can only to castle long 0-0-0, then we will have a more even match regardless of what Opening they use https://tcec-chess.com/

[Graham Banks]

Sad when you have to do such things though.

[lkaufman]

10 more games Mission impossible for Komodo Dragon to break even, Komodo Dragon might win 3 more games out of the 10 left

Note: If TCEC decides to experiment where the side that has White can only to castle long 0-0-0, then we will have a more even match regardless of what Opening they use https://tcec-chess.com/

If you change anything to try to equalize the chances of White and Black, the result (with top 3 engines at Rapid TC with 8 or more threads) is still going to to almost all draws. Giving White a near-decisive opening advantage is the only way to avoid excessive draws without changing the rules of chess. Changing the repetition rule would be the least-disruptive rule change that would have a huge impact on draws.

[bnst]

I think what we should get from this match is that we still have a top engine which is original and not just one of the million clones of Stockfish or Leela. It is very valuable for people analysing chess to have more than one viewpoint.

Keep up the good work !
anst

[mvanthoor]

If you change anything to try to equalize the chances of White and Black, the result (with top 3 engines at Rapid TC with 8 or more threads) is still going to to almost all draws. Giving White a near-decisive opening advantage is the only way to avoid excessive draws without changing the rules of chess. Changing the repetition rule would be the least-disruptive rule change that would have a huge impact on draws.

Could we do something comparable to Komi in Go?

Komi is 6.5 or 7.5 points for white.

Black goes first, which has been determined to have a value of 6 points. Thus white gets a Komi of 6. But, because black goes first, he has Sente and can determine the direction of the game. For this, white gets a compensation of 0.5. Thus if the game ends with exactly equal territory, white wins by 0.5. (The version with Komi at 7.5 is when you count territory including the stones on the board. Because Black goes first and thus plays the uneven points, he has one more stone to place than White, as there are 361 points on the board.)

We could do something similar in chess, maybe.

If I recall correctly, between players of the same strength at levels of over 2600, White will score 54% of the points. Maybe, when a game is drawn, we could count material on the board. The side with a material advantage wins. If material is exactly equal, Black wins because White went first.

We could test this by running thousands and thousands of games between engines to see if this, or a variation of this equalizes the game.

We could also do something with counting. Now we count:
1-0 (white wins)
0-1 (black wins)
½-½ (draw)

We could also do something like this, which is akin to Komi:

1-0 (white wins)
0-1 (black wins)
0.5 - 0.5 (draw with white having a material advantage)
0.4 - 0.6 (draw with exactly equal material; black gets more points because white moves first)
0.3 - 0.7 (draw with black having a material advantage)

And then play with the draw values until the number of points is equal if you have the same engine play games for thousands or millions of games. So if one plays a match of 10 games this could happen:

Player A: Wins 4 games
Player B: Wins 4 games

Game 9: Player A vs. Player B: draw with white a pawn up: 0.5 - 0.5
Game 10: Player B vs. Player A: draw with exactly equal material: 0.4 - 0.6

Player A would then win the match by 5.1 vs. 4.9.

The only reason that there are no draws in Go is because of the 0.5 extra Komi for white; there are plenty of games where one side or the other wins by only 0.5. When White takes a handicap, the first step is to eliminate Komi, but often the 0.5 extra is retained to avoid draws.

[lkaufman]

If you change anything to try to equalize the chances of White and Black, the result (with top 3 engines at Rapid TC with 8 or more threads) is still going to to almost all draws. Giving White a near-decisive opening advantage is the only way to avoid excessive draws without changing the rules of chess. Changing the repetition rule would be the least-disruptive rule change that would have a huge impact on draws.

Could we do something comparable to Komi in Go?

Komi is 6.5 or 7.5 points for white.

Black goes first, which has been determined to have a value of 6 points. Thus white gets a Komi of 6. But, because black goes first, he has Sente and can determine the direction of the game. For this, white gets a compensation of 0.5. Thus if the game ends with exactly equal territory, white wins by 0.5. (The version with Komi at 7.5 is when you count territory including the stones on the board. Because Black goes first and thus plays the uneven points, he has one more stone to place than White, as there are 361 points on the board.)

We could do something similar in chess, maybe.

If I recall correctly, between players of the same strength at levels of over 2600, White will score 54% of the points. Maybe, when a game is drawn, we could count material on the board. The side with a material advantage wins. If material is exactly equal, Black wins because White went first.

We could test this by running thousands and thousands of games between engines to see if this, or a variation of this equalizes the game.

We could also do something with counting. Now we count:
1-0 (white wins)
0-1 (black wins)
½-½ (draw)

We could also do something like this, which is akin to Komi:

1-0 (white wins)
0-1 (black wins)
0.5 - 0.5 (draw with white having a material advantage)
0.4 - 0.6 (draw with exactly equal material; black gets more points because white moves first)
0.3 - 0.7 (draw with black having a material advantage)

And then play with the draw values until the number of points is equal if you have the same engine play games for thousands or millions of games. So if one plays a match of 10 games this could happen:

Player A: Wins 4 games
Player B: Wins 4 games

Game 9: Player A vs. Player B: draw with white a pawn up: 0.5 - 0.5
Game 10: Player B vs. Player A: draw with exactly equal material: 0.4 - 0.6

Player A would then win the match by 5.1 vs. 4.9.

The only reason that there are no draws in Go is because of the 0.5 extra Komi for white; there are plenty of games where one side or the other wins by only 0.5. When White takes a handicap, the first step is to eliminate Komi, but often the 0.5 extra is retained to avoid draws.

As a Go player myself I have of course thought about this idea. There are several points to be made here. First point is that a critical issue which you didn't discuss is the point count to be used for material. If you use the traditional 1-3-3-5-9 count, then the proper result is almost surely a draw with equal material (the Marshall Gambit will be unplayable, but the Berlin or Petroff should probably keep equal material). So Black will normally "win" (or get 0.6 or whatever). Since two minors are normally better than rook and pawn, a better count would be 2-7-7-11-21, which would have fewer tied scores but still probably the normal result will be equal point count draws. But if you make bishop better than knight (as it usually is), say be 2-6-7-10-19 for example, then it is very likely that the proper result will be a draw with White one point ahead, since there are many defenses (especially NimzoIndian) where Black concedes bishop for knight right away to avoid worse problems. With that count it would probably be better to play White even if Black wins on equal point count, though this is far from certain.

It is also far from clear that material point count is the best count for the Komi rule. Another reasonable rule is that whichever king has reached the most advanced rank during the game wins in case of a draw, with Black winning if tied (so 1/2 point komi). Maybe more fair is that if White is exactly one point ahead by this rule the result remains a draw (so a one point Komi, akin to a 6 or 7 point Komi in go without the 1/2).

Finally if you just want to make the game fair with fractional scoring, there is no need to count material or anything else; just count draws as 0.6 for Black and 0.4 for White (idea promoted for decades by Ed Epp). This will cut down tied scores in tournaments and matches quite a bit, but won't really reduce draws.