Games with simple rules and deep structure.
9 topics · 9 curated works
No prior grounding assumed.
Japanese Chess (Shogi): The Science and Art of War or Struggle, Philosophically Treated
Choyo Suzuki · 1905
Frames shogi as a philosophy of struggle for an early Western audience, arguing that letting a captured piece change sides and return to the board…
+2 more at this level
Assumes you know the vocabulary.
Go and Go-Moku: The Oriental Board Games
Edward Lasker · 1934
Carries a chess master's eye across to Go, arguing the game rewards patient accumulation of influence over the forcing tactical exchanges that chess…
Primary sources and full treatments.
Temporal Difference Learning and TD-Gammon
Gerald Tesauro · 1995
A neural network trained purely by playing itself, with no expert knowledge encoded, reached championship-level backgammon, evidence that self-play…
+4 more at this level
9 works
Choyo Suzuki
Frames shogi as a philosophy of struggle for an early Western audience, arguing that letting a captured piece change sides and return to the board makes it a fundamentally different model of conflict from chess, where a captured piece is simply gone.
242 pageslink checked 17 Sept 2026Arthur Smith
The first full English-language account of Go, arguing its strategic depth comes from simple rules generating positions too numerous for the exhaustive calculation chess allows.
about 3 hours to readlink checked 17 Sept 2026Google DeepMind
Follows the 2016 match against Lee Sedol and shows that the decisive move was one no human would have played, raising the question of whether the system understood Go or merely searched it better.
link checked 17 Sept 2026Edward Lasker
Carries a chess master's eye across to Go, arguing the game rewards patient accumulation of influence over the forcing tactical exchanges that chess trains a strong player to look for first.
link checked 17 Sept 2026Gerald Tesauro
A neural network trained purely by playing itself, with no expert knowledge encoded, reached championship-level backgammon, evidence that self-play reinforcement learning can substitute for human expertise.
link checked 17 Sept 2026Erik D. Demaine & Robert A. Hearn
Surveys how the computational difficulty of playing a combinatorial game perfectly is itself a formal property of its rules, so an abstract game's depth can be measured rather than judged by feel.
link checked 17 Sept 2026Rémi Coulom
Introduces Monte Carlo Tree Search, arguing that random playouts averaged over many simulations evaluate a board position more reliably than a hand-built heuristic function, which is what later let computers surpass humans at Go.
link checked 17 Sept 2026Schaeffer et al.
Nineteen years of computation proves perfect checkers is a forced draw, settling a question about the game left open since the first learning programs of the 1950s.
link checked 17 Sept 2026Li et al.
Xiangqi strategy is non-transitive in a way chess mostly is not, so an engine trained only to beat its most recent self can cycle indefinitely rather than converge on one dominant style.
link checked 17 Sept 2026