Firscher was right about FischerRandom

[lkaufman]

Based on the 12 Classical time limit games played so far in the oddly-named "Freestyle Goat Challenge" event of Chess960 aka FischerRandom, it does appear that for human play 960 solves both problems of playing by memorized computer analysis and excessive draws at Classical time limits. Officially, 9 of the 12 games were decisive, but one of the 3 draws was agreed in a winning position just to clinch the match, so really 10 out of 12 decisive, with all the participants at the very top of World chess! Many games reach a winning position for one side by move ten or so, not due mainly to White's initial advantage but just to the difficulty of figuring out the way to play in the opening without assistance of engines or memorized theory. Basically I think we are seeing chess as it was in the 1800s again, when little was known about how to play the openings properly. I would expect that this event will lead to increased popularity of 960 (the new name freestyle is already used in chess for engine-assisted play, so it is confusing).

Unfortunately it is still too drawish for top engines with many cores at non-blitz time limits. We do have the chess 324 variant that is substantially less drawish. Perhaps we can find other ways to make 960 work just as well for engines as for humans. The problem is that White's advantage is not even halfway to the win/draw dividing line in most or all of the positions. Perhaps some limitation on Black's castling rights or on allowed opening moves (for example no copycat first move or no double pawn first move or first two moves) would be enough to get closer to the win/draw line which would make 960 fully viable (without need for opening books) for engine tournaments and rating lists. Many ideas to consider.

[towforce]

...Many ideas to consider.

Bigger chess board: make the board 10x10 or 12*12 (which would have have over twice as many squares as the standard 8x8 board).

* more move choices would throttle off the power of game tree searching (the branching factor would be too large)

* the positions would be less familiar, and a greater number of familiar features would appear in many positions, confusing both computers and humans

* the games would be longer: let's say white is half a percent better than black. This is obviously a flawed model for many reasons - but it does make the point that a longer game will magnify differences in strength. If a game contains 40 turns for each player, and at each turn white is half a percent better than black, then after 40 moves, white will be...

1.005^40 = 1.22 times better than black - so not likely to win.

If, however, there are 150 moves in the game, then white becomes...

1.005^150 = 2.11 times better than black - so quite likely to win.

Maybe it's better to look at it this way (still a flawed model, but some of you might prefer it to the above model): in an unknown position, white has a 0.1% chance of making a fatal blunder, whereas black has a 0.2% chance. In a 40 move game, the chances of avoiding a fatal error would be as follows:

White: 0.999^40 = 96% chance of avoiding a blunder
Black: 0.998^40 = 92% chance of avoiding a blunder

So probably a draw. However - lengthen this to a 150 move game...

White: 0.999^150 = 86% chance of avoiding a blunder
Black: 0.998^150 = 74% chance of avoiding a blunder

Overall probability of avoiding a blunder: 0.86 * 0.74 = 64% chance of avoiding a blunder.

The positions being less familiar, more complicated, and only searchable to a low depth, would obviously also greatly reduce either side's probability of avoiding a blunder as well.

Conclusion: increase the board size and the number of pieces!

[lkaufman]

Presumably if you expand the board to 10 x 10, you also would want to add two more pieces, which is what Capablanca chess, Grand chess, and Gothic chess are. There are already Fischer random versions, Capablanca Random Chess and Modern Capablanca Random chess. I don't doubt that they would be far less drawish with top engines playing (aside from the fact that top engines don't specialize in these games), and I personally do like those variants. But they are just different games than chess, and have obviously not caught on to any meaningful degree. Even without the extra pieces, they would feel like different games than chess; the pawns wouldn't butt heads when each advances two squares, the bishops would be clearly superior to knights, etc. The attraction of FischerRandom is that it becomes normal chess once the pieces reach normal squares. It is not a fundamentally new game. So I'm looking for a rule change that won't affect play after the first few moves, or else at least a minimal change. Simply forbidding perpetual check (as in Shogi and Shang-Chi) in FRC might be enough to make it not too drawish between engines, that is at least a testable hypothesis that would require only a minimal change to engines. But even that might be too radical for some.

[Modern Times]

I'm a huge fan of FRC and it works fantastically in human chess as above. Unfortunately as you also rightly point out, chess engines play at such a high level that top engines on good hardware simply draw too often, not much different from normal chess.

[CornfedForever]

Too much of an edge for White and the patterns that help chess players understand the game better...are missing. Just a sideshow. Of course, chess players are 'whores' when it comes to some rich dude waving $$$$ under their noses.

[lkaufman]

Too much of an edge for White and the patterns that help chess players understand the game better...are missing. Just a sideshow. Of course, chess players are 'whores' when it comes to some rich dude waving $$$$ under their noses.

The players all really seem to like being able to play without computer preparation. Certainly Magnus is not promoting this for the money (tiny to him), he just thinks that chess should not be a memory contest especially of computer lines. I think most of the other top players agree, at least they say things that imply so. White's edge is not larger on average than in standard chess, maybe a bit smaller overall, though of course a few positions are clearly more favorable for White (but none are anywhere near winning). You are right that players play the openings much worse without being able to rely on patterns that are specific to the normal start position, but this is exactly why most classical games are drawn at top 10 level; the interesting part of the game may all be played from memory, then it's just a matter of being careful. An alternative solution to FRC is to randomly choose openings from some list (maybe the 960 most popular short sequences or some other number); I don't know if that would be more or less popular. Of course there is far more theory on this, but in principle it should be equally hard to memorize theory of 960 random openings or 960 FRC positions. Carlsen gave up the World Championship to push for changes that would devalue opening prep; I don't think he is wedded to FRC but he has certainly made a case that something should change.

[chesskobra]

Fischer random with a 2-ply opening book, minus positions that give obvious advantage to one side will already give you of the order of 100k starting positions. If you also want symmetry preserving 2-ply book, that would give you about 10k positions.

[lkaufman]

Fischer random with a 2-ply opening book, minus positions that give obvious advantage to one side will already give you of the order of 100k starting positions. If you also want symmetry preserving 2-ply book, that would give you about 10k positions.

No one has suggested that 960 positions is too few, that isn't the problem. The problem for top engines is that they play so well that nearly equal positions always end in draws. So we either have to change the draw rules themselves, or else unbalance the start position sufficiently to produce winning chances. Since computer events normally allow for the opening to be repeated with colors reversed, there is no problem with unbalanced start positions, as long as they are not actually won for one side. So a 2-ply book that ONLY included positions with "obvious" advantage (but less than 1.0) would be a valid solution.

[pohl4711]

I made huge Chess324 opening sets, very imbalanced, using some single-step pawn moves.

CCRL already does a nice ratinglist using it:
https://www.computerchess.org.uk/ccrl/C ... index.html

"Thanks to Stefan Pohl at SPCC for the excellent unbalanced 3-move Chess324 book used in this tournament Chess324_xxl_big_+090_+119"

Testresults on my website:
https://www.sp-cc.de/anti-draw-openings.htm

Chess324 has the advantage, that all engines can play it, because no new (castling-)rules are needed. Not all engines can do Chess960. But all can do Chess324. And Chess324 games look less "chaotic" for the human eye, than Chess960...


Further information, from my website:
Chess324 64x means: All Chess324 positions combined with 2 single-step pawn-plies (example 1.a3 c6 or 1.e3 h6 and so on): 64 of these 2 single-step pawn-plies are possible, so there are 324 * 64 = 20736 possible different openings for evaluating.
The testing results show, that the classical Chess324 opening set has a pretty high draw-rate. My Chess324 64x sets give white a measuerable advantage, so the draw-rate is much lower. And all of my openings sets have at least 500 lines, which is enough for a 1000 game head-to-head competition (the classical Chess324 set contains 324 lines, only). But, compared to other openings concepts, the Chess324 64x openings sets are pretty small, because I had only 20736 opening lines as raw-data for evaluating, which is a very small number, compared to the other opening concepts... Additionally, I tried to build Chess324 openings with NBC/NBSC (no castlings for black, see below). But unfortunately, this did not work: The draw-rates were quite good, but the Elo-spreading of the results were bad...

Chess324 xxl means: All Chess324 positions combined with 6 single-step pawn-plies. Example: 1. h3 h6 2. g3 d6 3. d3 b6... So more than 600000 different Chess324 openings were possible and evaluated!!! The Chess324 xxl sets are around 30x bigger than the Chess324 64x sets...

[lkaufman]

I made huge Chess324 opening sets, very imbalanced, using some single-step pawn moves.

CCRL already does a nice ratinglist using it:
https://www.computerchess.org.uk/ccrl/C ... index.html

"Thanks to Stefan Pohl at SPCC for the excellent unbalanced 3-move Chess324 book used in this tournament Chess324_xxl_big_+090_+119"

Testresults on my website:
https://www.sp-cc.de/anti-draw-openings.htm

Chess324 has the advantage, that all engines can play it, because no new (castling-)rules are needed. Not all engines can do Chess960. But all can do Chess324. And Chess324 games look less "chaotic" for the human eye, than Chess960...


Further information, from my website:
Chess324 64x means: All Chess324 positions combined with 2 single-step pawn-plies (example 1.a3 c6 or 1.e3 h6 and so on): 64 of these 2 single-step pawn-plies are possible, so there are 324 * 64 = 20736 possible different openings for evaluating.
The testing results show, that the classical Chess324 opening set has a pretty high draw-rate. My Chess324 64x sets give white a measuerable advantage, so the draw-rate is much lower. And all of my openings sets have at least 500 lines, which is enough for a 1000 game head-to-head competition (the classical Chess324 set contains 324 lines, only). But, compared to other openings concepts, the Chess324 64x openings sets are pretty small, because I had only 20736 opening lines as raw-data for evaluating, which is a very small number, compared to the other opening concepts... Additionally, I tried to build Chess324 openings with NBC/NBSC (no castlings for black, see below). But unfortunately, this did not work: The draw-rates were quite good, but the Elo-spreading of the results were bad...

Chess324 xxl means: All Chess324 positions combined with 6 single-step pawn-plies. Example: 1. h3 h6 2. g3 d6 3. d3 b6... So more than 600000 different Chess324 openings were possible and evaluated!!! The Chess324 xxl sets are around 30x bigger than the Chess324 64x sets...

I was the first to jump on the Chess324 bandwagon, and it does indeed have many advantages over chess960. But humans are only playing 960, not 324, and I don't see much likelihood of that changing, especially now that a full tour of 960 has been announced. It is not desirable for engines and humans to be playing two different games.