To Kramnik, the problem with high-level chess is that it results in draws more often than wins. The logic behind his proposal is pretty straightforward:
- There are too many draws because defense is stronger than offense.
- Castling is a very strong defensive move.
- Banning castling keeps the king in the center, making it more vulnerable and empowering offense.
- It also makes it harder to "link" the rooks.
Most other solutions for high-level chess have focused on introducing randomness and time pressure. This is an interesting proposal because there is no randomness at all. In fact, it's a simpler variant of chess that everyone learns before you learn about about castling. But because castling is so strong, pretty much all established theory assumes that one or both sides castle early in the game.
Kramnik is apparently working with Alpha-Zero, and he says that Alpha-Zero comes up with some very interesting games in this new scenario.
Of course, this isn't a proposal which makes it likely that humans will start beating computers once again. It may turn out that White's moving first is an insurmountable advantage with a stronger offense. And after a few years, new opening theory will be established. But for those few years, high-level chess will be very interesting as the game is figured out once again.