Chess is often seen as a game of cold logic, but for Vladimir Kramnik, it’s all about beauty. “It’s a kind of creation,” he says. Kramnik’s appreciation for the artistry in the clash of minds on the chessboard helped him dethrone Garry Kasparov in 2000 and hold the title of world champion for several years.
However, Kramnik, who retired from competitive chess recently, believes that the game has lost some of its creativity. He partly attributes this to computers, whose precise calculations have led to a vast repository of openings and defenses that elite players memorize. “At the highest level, half of the game—sometimes an entire game—is played from memory,” he notes. “You’re not even playing your own preparation; you’re playing your computer’s preparation.”
In 2020, Kramnik proposed ideas to bring more human creativity back to chess, with the help of an unexpected ally—the world’s most powerful chess computer. He partnered with DeepMind, an AI research lab, to explore new chess variants using their advanced software, AlphaZero.
In 2017, AlphaZero demonstrated its prowess by teaching itself to defeat the best computer players in chess, Go, and shogi. Kramnik believes its insights reveal exciting new possibilities for chess, provided players are open to minor rule changes.
The project highlighted a new way for chess players and machines to collaborate. “Chess engines were initially designed to defeat humans,” says DeepMind researcher Nenad Tomašev. “Now we see systems like AlphaZero being used for creative exploration alongside humans.”
For around 1,500 years, chess has evolved with various rule changes. About 500 years ago, the game became more dynamic when European players transformed a slow-moving piece into the powerful queen. In 1996, Bobby Fischer introduced Fischer Random Chess (Chess960), which randomizes the starting positions of the powerful pieces to reduce rote memorization and encourage creativity. This variant now has its own tournaments.
DeepMind and Kramnik used AlphaZero to test nine chess variants, including no-castling chess and torpedo chess. No-castling chess, which had its first tournament in January 2020, eliminates the castling move, leading to new strategic patterns. Torpedo chess allows pawns to move up to two squares at any time, altering the value of pieces and the game’s dynamics.
AlphaZero’s experiments produced intriguing results. For instance, no-castling chess resulted in fewer draws compared to traditional chess. The project’s findings weren’t just about numbers but also about the aesthetic enjoyment of the game. Kramnik found beauty in AlphaZero’s adaptation to new rules, particularly in self-capture chess, where players can capture their own pieces, offering new tactical opportunities.
Kramnik hopes that AlphaZero’s exploration will inspire players to try these new variants. “It is our gift to the world of chess,” he says. Chess’s popularity surged during the pandemic, with many seeking intellectual challenges. Interest in Chess960 has also grown, indicating a readiness for new types of play. A Chess960 tournament featuring world No.1 Magnus Carlsen and former champion Kasparov shows this trend.
DeepMind’s work might also encourage more creative use of computers in chess, moving beyond simply making machines unbeatable. “We can focus on chess as an art,” says Eli David, a researcher who develops AI-powered chess engines. One project in his lab aims to create software that mimics the style of specific players, allowing insights into how past and present grandmasters might handle particular situations.
Kramnik’s experience demonstrates that collaborating with machines can enhance both the technical and emotional aspects of chess. “After three moves, you don’t know what to do,” he says. “It’s a nice feeling, like you’re a child again.”