TLDR: Google DeepMind has developed an AI system capable of generating original and creative chess puzzles, a feat that has garnered praise from chess Grandmasters and experts. The AI, trained on millions of puzzles and refined with reinforcement learning, aims to demonstrate genuine creativity in chess composition, producing aesthetically pleasing and counterintuitive challenges.
In a groundbreaking study titled “Generative Chess Puzzles,” researchers at Google DeepMind, in collaboration with the University of Oxford and Mila (University of Montreal), have unveiled an artificial intelligence system that can create original chess puzzles. Published on October 29, 2025, the 75-page study explores whether AI can exhibit true creativity in chess, moving beyond mere mechanical combinations to produce puzzles that are aesthetically pleasing, surprising, and counterintuitive.
The AI system underwent extensive training, initially learning from four million puzzles sourced from Lichess. This foundational knowledge was then enhanced through large neural networks. The models were further refined using reinforcement learning, employing a reward function designed to prioritize puzzles that were both unique (requiring only one winning move) and counterintuitive (solvable by powerful engines but challenging for weaker ones).
To evaluate the AI’s creative output, a panel of distinguished chess experts, including GM Matthew Sadler, GM Jonathan Levitt, and FM Amatzia Avni (an International Master in chess compositions), reviewed a selection of the generated puzzles. Their separate 39-page conclusion provides detailed feedback. The experts were tasked with identifying their favorite puzzles and articulating what made them compelling, considering factors such as creativity, challenge level, and aesthetic design.
While opinions among the experts varied, particularly in defining a ‘compelling puzzle,’ their overall feedback was positive. They did, however, offer some criticism, noting instances where positions were too trivial or unrealistic. GM Jonathan Levitt lauded the project as a “pioneering step.”
DeepMind views this project as a significant milestone in AI creativity. The authors stated, “Our work demonstrates a significant advancement in the AI-driven generation of creative chess puzzles.” They added, “The positions represent a pioneering step toward human–AI partnership in chess composition. The experts found them beautiful, counterintuitive, and original—proof that creativity and beauty in chess remain deeply subjective.”
Also Read:
This research highlights AI’s evolving capacity for creative problem-solving, pushing the boundaries of what is considered achievable for artificial intelligence in complex domains like chess.
