AI May Have Cracked 1,700-Year-Old Board Game No One Knew How To Play

For about 40 years, a roughly brick-sized chunk of carved limestone sat in a Dutch museum shrouded in mystery. The pattern of lines scratched into its surface looked like a game board, but nobody could figure out what game. No known ancient game matched the design. Now, a team of archaeologists and computer scientists has used artificial intelligence to take a serious crack at the puzzle, and in doing so, suggests this type of game may date back further than previously documented.

Housed at Het Romeins Museum in Heerlen, the Netherlands, the stone traces back to Coriovallum, a Roman settlement founded under Emperor Augustus that thrived at the crossroads of two major Roman roads. At its peak, Coriovallum covered roughly 119 acres, and its residents clearly had leisure time to fill. Glass game pieces from the site are also in the museum’s collection. Weighing about 7.5 pounds and measuring roughly 8 by 6 inches, the stone was apparently carved during the late Roman period from a piece of ornamental building stone and etched with a series of geometric lines. What those lines meant had stumped researchers since at least 1984.

Rather than shelve the object as unexplained, a team led by Walter Crist of Leiden University combined physical analysis of the stone’s surface with AI-driven game simulations. Their findings, published in the journal Antiquity, suggest the stone was most likely a game board, and that AI could identify a probable ruleset from a wide range of plausible configurations.

Wear Marks Hold the Clues to an Ancient Roman Board Game

Before running any simulations, researchers needed physical evidence that the stone had actually been used as a game board. That required use-wear analysis, a technique that reads the microscopic damage left on a surface through repeated contact. Under microscopy, the surface along the stone’s incised lines showed smooth, leveled grain structure consistent with abrasion from a harder object, the kind of wear produced when small glass or stone game pieces are repeatedly slid across softer limestone.

Roman-era game pieces found at Coriovallum are typically flat, dome-shaped counters. Given their shape, the most natural way to move them is to push them along the lines of a board rather than lift and place them, and that repeated pushing leaves a distinctive mark. As the researchers note, the wear pattern appears only alongside the incised lines, suggesting intentional, line-following activity. Grinding grain or pigments wouldn’t leave that pattern, since those tasks don’t require following a geometric path.

A 3D model of the stone, built using a combination of scanning and light-mapping techniques, confirmed the surface is measurably lower along the lines than in surrounding areas. One diagonal in particular showed heavier wear than the rest, and that asymmetry became the target for the simulations.

AI Simulated Thousands of Matches to Reconstruct This Ancient Roman Board Game

Researchers used a software platform called Ludii, a universal board game engine, to run AI-driven simulations. They drew on the Ludii games database, described in the paper as “the largest comprehensive database of traditional board game rules currently available,” to identify historically documented European games that could plausibly fit this board.

The stone’s lines could be read as six different possible board layouts, and any game involves variables like starting positions, legal moves, and win conditions, so the team tested 130 possible configurations in total. For each one, the AI ran 1,000 simulated games and tracked which lines received the most use. The goal was to find which rulesets would naturally produce the same wear pattern seen on the stone, particularly along that heavily worn diagonal.

Nine configurations matched the wear evidence well enough to qualify. Every single one was a blocking game, a type in which one player tries to trap the opponent’s pieces until they can no longer move. None of the alignment games, where players race to place three pieces in a row, came close.

A Forgotten Game Type May Have Much Deeper Roots Than History Suggests

Seven of the nine matching configurations involved four pieces against two, with the side holding more pieces attempting to corner the opponent’s two. It’s structurally similar to haretavl, a Scandinavian “hare and hounds” game documented primarily in the 19th and 20th centuries, and to related games recorded across Italy, Spain, Greece, and elsewhere in Europe.

That similarity matters historically. Blocking games are not described in texts from the Roman era, which has led some researchers to assume they weren’t part of Roman gaming culture. Yet boards matching the geometry of some European blocking games have turned up in Roman-era graffiti in Rome and at ancient sites in Turkey. As the researchers note, “since the rules for traditional games are rarely written down and simple games are often drawn on the ground and played with stones or seeds, rendering them archaeologically undetectable, these blocking games may have existed for centuries without generating much material or literary evidence.”

Uncertainty remains. Other uses for the stone can’t be fully ruled out, including a mason’s guideline or a tool for shaping objects. AI simulations also model optimal play, not the socially messy, emotionally driven way real humans actually behave at a game board. In simple games like these, though, people tend to play closer to optimal strategy, which narrows that gap considerably.

At its core, the study points to a new approach for archaeological objects that don’t fit neatly into known categories. By pairing physical evidence with AI simulation, researchers moved from “this might be a game board” to “here’s probably how it was played,” and could open a path to do the same with other unexplained objects sitting in museum collections around the world. As the paper concludes, “the act of playing a board game is fundamentally the same today as it was in past millennia,” and the worn lines on a piece of Roman limestone are, across 1,700 years, proof enough of that.

---