For the first time, an international research team has harnessed artificial intelligence (AI) to decode the rules of an ancient board game, pioneering a new way to reveal long-lost historical secrets.
By analysing an engraved limestone object from the Roman Netherlands, the team was able to determine likely game rules, based on its distinctive markings
The new research published in Antiquity journal was led by Maastricht University (The Netherlands) and Leiden University (The Netherlands) with input from Flinders University (South Australia), the Université Catholique de Louvain (Belgium) and The Roman Museum and restoration studio Restaura in Heerlen.
The object, found in what is now Heerlen in the Netherlands, features a pattern of unusual intersecting lines that had puzzled archaeologists for decades.
Because most everyday Roman games were drawn in dust or carved into wood (materials unlikely to survive), this carefully shaped limestone piece offered a rare opportunity to investigate ancient gameplay.
"The stone shows a geometric pattern and visible wear that are consistent with sliding game pieces across the surface, which point strongly to repeated play rather than another purpose," says lead author, Dr Walter Crist, an archaeologist at Leiden University who specialises in ancient games.
To determine whether the stone was a game board and how it worked, the research team used AI to simulate hundreds of possible rule sets, to see which produced the same patterns of wear found on the object.
"The uneven wear along the carved lines raises a key question about whether AI‑driven simulated play could reproduce that same pattern," says Dr Crist.
Using the AI-driven play system Ludii, the researchers made two AI agents play against each other using the object as a board, utilising rule sets from many ancient board games documented in Europe, such as haretavl from Scandinavia and gioco dell'orso from Italy.
Flinders University computer scientist, Dr Matthew Stephenson , says that using modern AI techniques can bridge the gap between historical and computational studies of games.
"We ran the simulations repeatedly, adjusting rules each time to see which movements would cause the same concentrated friction seen on the original stone," says Dr Stephenson, from Flinders' College of Science and Engineering.
"The simulations pointed strongly to a type of strategy game known as a blocking game. In blocking games, players try to trap their opponent's pieces by preventing movement rather than capturing them."
Because blocking games are scarcely documented before the Middle Ages, the findings suggest such games may have a deeper history than previously documented, whilst the study also demonstrates the transformative potential of AI for archaeology.
"This is the first time that AI-driven simulated play has been used together with archaeological methods to identify a board game," says Dr Crist.
"It offers archaeologists a promising new tool for understanding ancient games that don't resemble those known from surviving texts or artworks."
This work took place at Maastricht University and as part of the Digital Ludeme Project in Europe, which used artificial intelligence to produce more reliable reconstructions of ancient games that are plausible both historically and mathematically.
By blending archaeology, digital modelling and cultural history, the team provided a clearer understanding of an object that once seemed unexplainable.
"The success of this approach suggests that many other mysterious artefacts may hold hidden stories waiting to be uncovered with the help of modern technology," says Dr Stephenson.
"It shows how AI can contribute to our understanding of materials that would otherwise be difficult to interpret."
The paper, Ludus Coriovalli: using artificial intelligence-driven simulations to identify rules for an ancient board game , by Walter Crist (Leiden University), Éric Piette (Université Catholique de Louvain), Karen Jeneson (Het Romeins Museum), Dennis J.N.J. Soemers (Maastricht University), Matthew Stephenson (Flinders University), Luk van Goor (Restauratieatelier Restaura) and Cameron Browne (Maastricht University), was published in Antiquity. DOI: 10.15184/aqy.2025.10264
Acknowledgements:This research was funded by the European Research Council as part of Consolidator Grant #771292 'Digital Ludeme Project'. Computing resources were provided by the Dutch national e-infrastructure with the support of the SURF co-operative (EINF-3845 'Analysing Traditional Game Properties and Concepts'; EINF-4028 'Evaluation of Trained AIs for General Game Playing'), of the research programme Computing Time on National Computer Facilities (partly financed by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek). Further discussion of results and applications were made possible through European Cooperation in Science and Technology (COST) Action #CA22145 'Computational Techniques for Tabletop Games Heritage (GameTable)'. Open access funding provided by Leiden University.
