AI & Archaeology · Roman, 27 BC–AD 476 · Heerlen (Roman Coriovallum), Limburg, Netherlands
How AI rebuilt the possible rules of a Roman board game from the Netherlands
Wear marks on a carved limestone object from Roman Heerlen, fed into game simulations, point to a blocking game in which four pieces trap two.
Peer-reviewed paper
Heerlen (Roman Coriovallum), Netherlands Researchers have used artificial intelligence to suggest how people may have played a game on a carved stone from Roman Heerlen, in the south of the Netherlands. Their simulations point to a blocking game, in which one player tries to stop the other from moving. The study is published in the journal Antiquity.
The stone is object 04433 in the collections of Het Romeins Museum in Heerlen. It was excavated from the remains of Coriovallum, the Roman town under modern Heerlen. The town was occupied from the reign of the emperor Augustus (27 BC–AD 14) until the fall of the Western Roman Empire around AD 476. It lay in the province of Germania Inferior.
The team comes from institutions in the Netherlands, Belgium and Australia. They combined use-wear analysis, the study of tiny marks left on an object by handling, with computer simulations of play.
An odd stone with no known parallel
The object measures 212 by 145 by 71 mm (8 by 6 by 3 inches). It is made of white Jurassic limestone, probably quarried at Norroy in north-eastern France. In the Roman northern provinces this stone was usually used for large architectural pieces. A small object made from it is therefore unusual.
The stone is worked on all sides. One face was smoothed and cut with lines that form symmetrical geometric shapes. It had been interpreted as a game board before. Nothing like it is known, though, and its use has been much debated.
Board games have a long history, known from ancient texts and art and from excavated dice and playing pieces. Game boards are rarer finds. They are seldom found with pieces or other objects that show their purpose. Written sources and images that show such boards do not give the rules of play.
Wear marks fed into a game engine
To test whether the stone was a board, the team made a depth map of its surface. They used photogrammetry together with a photometric stereo technique. The smoothed face was generally more worn around the incised lines. Microscopic examination confirmed this. The researchers think hard objects were dragged across the lines.
With these marks hinting at play, the team turned to Ludii. This general game system was developed as part of the Digital Ludeme Project. It holds the largest database of traditional board-game rules currently available. It can also run AI-driven simulations of games, called playouts.
The team entered criteria taken from the wear pattern. One was that the board had fewer than 20 likely playing spots. That narrowed the options to two kinds of game. One was a blocking game. The other was a game where the aim is to place three pieces in a row.
Next the team ran playouts under various criteria to see which matched the wear best. They also added biases to the simulated players. These made the games reflect more closely how humans actually play board games.
The results suggested at least nine ways the game could have been played. The one that best fits all the criteria is a blocking game. One player starts with four pieces on a vertical side of the main rectangle. That player tries to trap the opponent's two pieces on the opposite side.
The reconstruction is the best fit among several possible rule sets. In their paper, the researchers write that combining AI simulation with use-wear analysis can help identify potential game boards. They add that it can rebuild playable rule-sets that may give indications of how people played games in the past.
Sources
Paper: Antiquity, 2025, https://doi.org/10.15184/aqy.2025.10264
Spotted a mistake? Tell the editor about a spelling error or a fact that is wrong. We correct stories openly.
Readers' picks show on the front page.
Follow @AncientHeraldHQ on X for the day's finds as they come in.
Comments
House rules. Stay on the subject, be kind to other readers and say where your facts come from. No advertising and no personal attacks. The editor reads along and hides comments that break these rules.
Loading comments…