Assessing Game Balance with AlphaZero: Exploring Alternative Rule Sets in Chess

Summary: much ado about nothing: it does not change anything substantial.

Those variants interest me much more than convential variants. I would like to try a few of them. Self-capture and torpedo seem interesting - and if, as mentioned in a couple of articles & videos about them, they allow for uniquely beautiful compositions and gameplay, then I think they have a place in the world of chess. However, I don't see any of them replacing the rules of standard chess regardless of whether or not they result in more decisive games.

If playing a more decisive version of standard chess is the goal - I would probably lean most towards Kramik's idea of "no castling" chess - certainly more than variants that allow sideways pawn moves. And I don't like the idea stalemate=win. That seems to rob the game of a certain artistry regardless of decisiveness. If decisiveness is the goal, I think scoring stalemates differently is a better option.

The paper says differently: no castling, stalemate=win... do not make it substantially more decisive. Torpedo is a bit more decisive, but not that much either.
I thought the proposals of Lasker and Capablanca to award 0.75 point for stalemate were interesting, but this paper learns us that even stalemate = 1 does not change decisiveness in a substantial way.

Yes, surprising indeed. It means after all our rules are not that bad as they are.

@tpr Well, I must admit that my interest in these variants is mostly novel indulgement. New forms of artistry seem possible while maintaining essential elements of the game. That's my own personal interest. I'm a bit of a traditionalist when it comes to arguments about actual rule changes.

That said, if "no castling" doesn't significantly affect the decisiveness of the game, would you go so far as to argue that castling is overrated or unnecessary to play winning chess? Castling certainly seems like an important aspect of successful play.

Castling is very powerful, it is like 3 moves in one. It brings the king to safety and connects the rooks to activate them.
However, if both sides are deprived of castling, it is about the same.
In most chess games both sides castle king's side O-O.
Castling was originally introduced to mitigate the increased power of the queen as compared to the previous rules.
This paper shows however, that introduction of castling would not have been necessary: the game is balanced (draw) even without castling.
My guess is, that if white were deprived of castling rights and black remains allowed to castle, then the game might be in black's favor, but that is just a guess.
This could be tested by engine versus engine games from the starting position, but with both castling rights O-O and O-O-O for white disabled.

#9 Good point. I must admit though, castling as a rule, has always seemed like a bit of a 'patch' job. It has always seemed like an after thought by the creators of the game.. to make up for an inefficiency in play. I suspect that prior to 20th century advances in theory and defense, castling was more necessary for both sides. In other words, today there is little difference in decisiveness if both sides are unable to castle but prior to the 20th century, a lack of castling would likely have resulted in more decisive games.