AI-Generated Chess Puzzles

While strong chess players intuitively recognize the beauty of a position, articulating the precise elements that constitute creativity remains elusive. To address this, the team pre-trained generative models on the lichess puzzles and then applied reinforcement learning, using novel rewards designed for uniqueness, counter-intuitiveness, realism, and novelty. This approach doubled the number of novel chess puzzles compared to the original training data, while successfully maintaining aesthetic diversity.

To get a sense of the positions that this reward prefers, we compiled a study featuring positions from the lichess data with high reward and aesthetic themes like sacrifice, underpromotion, attacking withdrawal, novotny, interference, unprotected position and knight on the rim is dim.

You can try to solve them here: https://lichess.org/study/UNerbDSl

Three distinguished experts—International Master of chess compositions Amatzia Avni (author of "Creative Chess"), Grandmaster Jonathan Levitt (author of "Secrets of Spectacular Chess"), and Grandmaster Matthew Sadler (author of "Game Changer")—evaluated and selected the puzzles they found most compelling. Their preference was for puzzles exhibiting original, paradoxical, surprising, and naturally occurring positions, with particular emphasis on those that integrated aesthetic themes in innovative ways and demonstrated exceptional over-the-board vision.

IM for Chess Compositions Amatzia Avni wrote: "This booklet adds novel, AI-generated puzzles
to the existing chess literature, serving as a resource for both training and enjoyment. A
valuable chess puzzle should be original and creative, with a surprising, counter-intuitive
key move and a smart follow-up. The ideal puzzle is also aesthetically pleasing and offers
a satisfying, flowing solution. It must strike a good balance in difficulty – challenging
enough to avoid being obvious, but not so hard as to cause frustration"

GM Jonathan Levitt wrote: "For years, chess composers have worked with computers to verify
soundness of their work and thus assist in the process of creation too. This nature of
collaboration is evolving, with AI now capable of generating interesting chess positions,
beyond just “mining” databases. The positions in this booklet represent a pioneering step
in this human-AI partnership. While these initial AI-generated endgame compositions
are not yet at a prize-winning level, they clearly demonstrate the potential to be."

GM Matthew Sadler wrote: "It was an intriguing experience to assess the chess booklet. I have
strong preferences about what makes a good puzzle position. In particular, I favor natural
positions resulting from reasonable play by both sides. Puzzles lose my interest if one
side’s pieces are clearly misplaced or if a complex solution yields a minimal advantage,
like being up a single pawn after sacrificing multiple pieces. Even with those stringent
conditions, I enjoyed many of the positions in this booklet".

Solve the puzzles here https://lichess.org/study/dLYe8R4f

• Read the full review and analysis here
• Read the technical paper here
• Watch the analysis by GothamChess here https://www.youtube.com/watch?v=e47VomRFhAU&t=15s