Scientists Simulate the Rules of an Ancient Game Using AI

At first glance, the stone doesn’t seem like much. It could be mistaken for a discarded architectural fragment as it sits in Heerlen under museum lighting, slightly uneven and with faint scratches on its surface. People stroll by it. A few hesitate. The majority don’t. However, there is a moment when the grooves start to feel deliberate, almost restless, as if something had once repeatedly moved across them when you are standing very close.

Walter Crist stopped there. Initially, not physically, but mentally. When he examined the item in a nearby museum in 2020, he noticed something subtle: the wear wasn’t random. It traveled along routes. ones that are repeated. It’s difficult to ignore how some lines seem deeper, almost polished by friction, implying that long-gone hands had repeatedly pushed pieces along the same paths.

That tiny, nearly coincidental observation completely changed the narrative. Suddenly, what had been loosely referred to as a “game board” felt more precise and alive. However, there was an issue. Nobody was able to play it.

In that sense, ancient games are peculiar. Boards and occasionally pieces are left behind, but instructions are rarely left behind. It’s possible that, similar to modern games like tic tac toe or checkers, people learned by observing others. Rules didn’t exist in text; they lived in memory. The game ends in silence when that chain breaks.

Crist and his group had to deal with that quiet. They had a Roman-style carved limestone slab with roughly 22 possible piece locations. Their patterns of clothing suggested movement. However, they lacked descriptions, drawings, and a rulebook. Only scratches. Thus, they resorted to an unexpected solution: artificial intelligence.

The team collected more than a hundred potential rule sets that might fit the geometry of the board using the Ludii database, an extensive collection of classic board game rules. They then allowed AI to simulate thousands of matches for each set rather than testing them manually, which would take years. The system played out scenarios silently and repeatedly, monitoring the movement of pieces, their clustering locations, and the paths that would deteriorate over time.

The idea of machines playing imagined games from two millennia ago in an attempt to imitate the behaviors of people who never left instructions behind is strangely poetic. In doing so, the AI might have been simulating human behavior rather than merely solving a puzzle.

The findings made things more specific. Wear patterns that matched the stone were only produced by a small number of rule sets. And they all pointed to a particular type of game: blocking games. In an effort to trap their opponent and put them in situations from which they could not escape, players would move pieces along the lines.

It sounds easy. However, that type of play has a subtle tension as players anticipate moves, close off options, and gradually tighten control. There’s a feeling that this wasn’t merely amusement as you watch the reconstructed logic take shape. It needed to be attended to. Plan of action. Be patient.

The fact that these games are primarily connected to medieval Europe rather than the Roman era is even more fascinating. That begs the question of whether this was an isolated invention or proof of a larger, unrecorded tradition.

Whether this game was popular or limited to a small community is still unknown. The stone doesn’t say anything. It simply indicates that it was used for a long enough period of time by one or more people to leave traces that have endured for centuries.

And that survival seems almost coincidental. The majority of ancient games were either played with perishable materials or sketched in sand. They vanished along with the performers. For decades, this one persisted, carved into limestone, buried, rediscovered, and then misinterpreted.

The past doesn’t seem as far away when you stand in front of it or even just imagine it. It hasn’t really changed in thousands of years to move a piece across a board, block an opponent, or think two steps ahead. Seeing this develop via AI makes that continuity oddly apparent.

There’s also a hint of unease. AI successfully recreated a mode of thought, deriving intentions from scratches, rather than merely analyzing the artifact. That is strong, but it also begs the question. How much of this is interpretation influenced by contemporary presumptions, and how much is discovery?

Crist appears to be conscious of that tension. The approach is effective, the patterns align, but certainty is still elusive. And perhaps that is appropriate. Like ancient lives, ancient games seldom provide comprehensive solutions.

Nevertheless, something has been found. It’s not perfect, but it’s enough to picture two people sitting somewhere in the Roman Coriovallum, leaning over this exact stone, moving pieces back and forth, maybe laughing, maybe quietly arguing, maybe just killing time. The scratches are still there. And at last, they start to make some sense.