TL;DR

Researchers at Maastricht University are using AI to reconstruct the rules of ancient board games whose instructions have been lost to time. The Digital Ludeme Project, led by Cameron Browne, has compiled all known board games from before 1875 and uses algorithmic generation and AI play-testing to infer plausible rulesets.

Breaking Games Into Parts

The project’s system, called Ludii, borrows computational techniques from genetics and applies them to game design. Researchers break games down into their constituent parts — called “ludemes” — and codify them in a database alongside cultural and historical information. These ludemes function somewhat like genes: small, modular components that can be recombined to produce complete rulesets.

Using algorithmic procedural generation, the system creates candidate rulesets for games where only fragments of the original rules survive. The approach has already produced results with a Roman board game discovered in a tomb in Slovakia in 2018, where physical game pieces were found but no instructions.

AI Agents as Quality Testers

Generated rulesets need evaluation, and that is where AI agents come in. The project uses Monte Carlo tree search — a simpler relative of the approach behind DeepMind’s AlphaGo — to play thousands of games under each candidate ruleset. The AI evaluates games against several quality criteria: strategic depth, dramatic tension, clear victory conditions, reasonable game length, and a low frequency of draws.

This automated play-testing allows the team to rapidly filter hundreds of rule variations and identify the ones that produce the most engaging gameplay. The result is not necessarily the exact historical ruleset, but a set of plausible reconstructions that are consistent with the available archaeological evidence and produce games worth playing.

Looking Forward

Browne’s team is clear that the system is intended to complement manual historical research, not replace it. The project has compiled all known games from before 1875 and created hundreds of variations, building a resource that historians and archaeologists can use alongside traditional methods. As the database grows and the AI agents improve, the project may help recover not just individual games but broader patterns in how human societies developed rules, strategy, and competitive play across cultures and centuries.