Kramnik And AlphaZero: How To Rethink Chess
For chess to flourish, we need a new challenge.
The increasing strength of chess engines, the millions of computer games and the volumes of opening theory available to every player are making top-level chess less imaginative. Decisive games in super-tournaments have declined, while the number of games with what I'd call "creative" content is also on the slide.
The 2018 world championship match between Magnus Carlsen and Fabiano Caruana, for example, ended with zero decisive classical games. (Carlsen defended his world title by winning a rapid-game playoff.)
This is not the players’ fault, but the reality they face. It would be strange to expect them to deliberately decrease their chances of a positive outcome by taking unreasonable risks for the sake of playing more “entertaining” games. From my own experience, I know how difficult it has become to force a complex and interesting fight if your opponent wants to play it safe. As soon as one side chooses a relatively sterile line of play, the opponent is forced to follow suit, leading to an unoriginal game and an inevitably drawish outcome.
Of course, there are still some fascinating top-level games being played, but to keep chess alive, I believe we must reverse this trend before the game’s spirit fades away.
Working with DeepMind:
So I started thinking, if the outcome is always the same, perhaps there’s something we can do to reinvigorate the game. I spoke with Demis Hassabis, the founder and CEO of the artificial intelligence lab DeepMind. Hassabis was once one of the strongest junior chess players in the United Kingdom and is still a devoted chess aficionado. I was granted an opportunity to test my theory with the famous machine-learning chess engine AlphaZero.
Working with DeepMind researchers Ulrich Paquet and Nenad Tomasev, we used AlphaZero as a petri dish to test different variations and see how the game might unfold. Ultimately, our mission was to find an adjustment to the rules to allow more space for human creativity.
Tweaking the rules of chess is not a new idea. One of the most widely played variants is Fischer Random, also known as Chess960. The recent World Fischer Random Chess Championship drew quite a bit of attention and resulted in Wesley So’s triumph over Carlsen in the finals.
Fischer Random is an interesting format, but it has its drawbacks. In particular, the nontraditional starting positions make it difficult for many amateurs to enjoy the game until more familiar positions are achieved. The same is true for world-class players, as many have confessed to me privately. Finally, it also seems to lack an aesthetic quality found in traditional chess, which makes it less appealing for both players and viewers, even if it does occasionally result in an exciting game.
No-castling chess:
My aim was to find a chess variant that would not only have the potential to bring the excitement and decisive victories back to chess, but is also aesthetically pleasing. The goal was to reignite interest and introduce players and audiences to the immense complexity and creativity of the original game of chess.
To begin, we tasked AlphaZero with exploring a variant that prevented either side from castling, trying different opening moves from both sides. The outcome was beyond our expectations!
We let AlphaZero learn how to play "no-castling chess" from scratch, allowing the program to incrementally learn how to master the game through a process of trial and error, similar to how it taught itself to play classical chess. After playing millions of games, AlphaZero became a no-castling expert, allowing us to analyze how it plays and assess the overall game balance.
The win/loss percentages for both White and Black are similar to classical chess, suggesting that the no-castling variant should be quite playable without favoring a particular player. Preventing the king from retreating to a safe distance means that all of the pieces have to engage in the melee, making the play more dynamic and entertaining, with a number of original patterns.
Two AlphaZero no-castling games annotated by Kramnik.
The advantages do not stop there. The no-castling restriction means that players cannot rely on memorized patterns; they are forced to think creatively from the beginning. Even if a player wants to force a draw, it is nearly impossible to control everything. Plus, this variant makes it practically impossible to play it safe, even as White, because it is so much harder to find a completely secure place for the king.
Finally, it also levels the playing field, so that amateur players have a better chance at playing against more seasoned opponents, who often memorise opening theory.
Old boundaries, new creativity:
A tournament using this variant will definitely see a significant increase in the number of decisive games—far beyond 50 percent, I estimate—along with an explosion of creative and unexpected ideas.
There are still many details to investigate. For example, the openings seem to be more complicated in the AlphaZero no-castling games, meaning that new opening theory has to be developed and players will need to explore new approaches to king safety. But none of this should prevent the chess community from trying this game.
I highly recommend that organizers of upcoming tournaments and chess lovers across the world give this variant a try.
No-castling has the potential to reset the clock, making creativity and depth of thought more important than just memorizing patterns and spending preparation time pressing the spacebar to get the next computer-engine move. By adjusting our limits, we can harness a new generation of players, original ideas and a future of exciting, decisive and creative chess.
Would you play no-castling chess? Let us know in the comments.
More on Vladimir Kramnik:
More on AlphaZero:
- AlphaZero's surprise debut
- AlphaZero beats Stockfish again
- Official AlphaZero resources
- Game Changer: AlphaZero's Groundbreaking Chess Strategies and the Promise of AI by Matthew Sadler and Natasha Regan
- 10 more unannotated AlphaZero-vs-AlphaZero no-castling games (below)