Oct 26, 2021 9:00 AM

This AI Resurrects Ancient Board Games—and Lets You Play Them

What tabletop games did our ancestors play in 1000 BC? A new research project wants to find out, and make them playable online too.

The AI Database →

Application

Games

Prediction

End User

Research

Sector

Games

In 1901, on an excavation trip to Crete, British archaeologist Arthur Evans unearthed items he believed belonged to a royal game dating back millennia: a board fashioned out of ivory, gold, silver, and rock crystals, and four conical pieces nearby, assumed to be the tokens. Playing it, however, stumped Evans, and many others after him who took a stab at it. There was no rulebook, no hints, and no other copies have ever been found. Games need instructions for players to follow. Without any, the Greek board’s function remained unresolved—that is, until recently.

Enter artificial intelligence and a group of researchers from Maastricht University in the Netherlands. Thanks to an algorithm the team used to analyze the playability of one suggested ruleset, the century-old guesswork could soon be taken out of the Knossos game. Today, not only can its recognition as a game be further assessed, with hopes of a clearer answer in future, a version of it is also playable online. And for the first time, so are hundreds of other games thought to have been lost to history.

Board games go back a long way. Centuries ago, before the chess we know today, there was Chaturanga in India, Shogi in Japan, and Xiangqi in China. And long before them was Senet, one of the earliest known games, which, along with others played in ancient Egypt, may have ultimately inspired backgammon. “Games are social lubricants,” explains Cameron Browne, a computer scientist at the university who received his PhD in AI and game design. “Even if two cultures don’t speak the same language, they can exchange play. This happened throughout history. Wherever people spread to, wherever soldiers were stationed, wherever merchants were trading. Anyone who had time to kill would often teach those around them the games they knew.”

Whether discovered buried in rubble, stashed away in tombs, or inscribed on tablets, the archaeological evidence left behind reveals that nearly all cultures created and played games. But like many odds and ends excavated, our knowledge of ancient games is fragmented. We know their origins, but the gameplay has long been a stumbling block, since the rules were typically passed on by word of mouth instead of being written down. The little that is known is left open to modern interpretation.

It’s these lapses in board game history that gave legs to the five-year Digital Ludeme Project, which Browne leads. “Games are a great cultural resource that’s been largely underutilized. We don’t even know how so many of them were played, especially when you go farther back in time,” he says. “So the question for me was, can we use modern AI techniques to shed insight into how these ancient games were played and, together with the evidence available, help reconstruct them?”

As it turns out, the answer is a resounding yes. It’s been three years since Browne and his colleagues set to work, and already they have brought nearly a thousand board games online, ranging from across three time periods and nine regions. Thanks to them, games once popular in the second and first millennia BC, like 58 holes, are now just a few clicks away for anyone on the internet.

Interestingly enough, this reconstruction process begins with the opposite. Games are first broken down into fundamental units of information called ludemes, which refers to elements of play such as the number of players, movement of pieces, or criteria to win. Once a game is codified in this manner, the team then fills in the missing pages of its rulebook with the help of relevant historical information, like when it or another game with similar ludemes was played and by whom.

The riddle however is only partly solved at this stage. Others who do similar work–manually–usually hit a dead end here. It’s because what looks good on paper might not translate as well in reality, Browne explains. “The rules might make sense when you read them, but you don’t know how well they actually work unless you play the game. Quite often, rules that make perfect sense play terribly as games.”

And so modern AI can get the project over the hump. Every hypothesis the team has about a game they’re studying is then fed through the Ludii software, where, after thousands of rounds of play-testing, its playability is assessed in just a matter of hours. The algorithm is a constant “work in progress,” though. A next step is tweaking it so that it also determines a game’s quality: whether or not an iteration would be fun to play, interesting enough to pass on, and easy to learn. Multiple yeses would mean it is likely to survive the tides of time, and therefore its most reasonable set of rules can be concluded.

But computers can have blind spots too, in that they only measure what’s measurable. Here’s where Walter Crist comes in. Crist, an anthropologist on the team, brings a human touch to the very computational project. He accounts for the intangibles the algorithm cannot calculate, like the social aspect of games. While rules may be integral, they don’t account for every possible scenario, whereas decorum might. A player for example could make the same move over and over again and prevent the game from ending. But people generally don’t do that because of social pressure and the desire to build relationships, according to Crist. “Not every situation has a rule for it; sometimes it’s one that’s worked out between the two players.”

Reconstructions aside, the researchers are also making exciting rediscoveries. It doesn’t matter if an artifact looks, swims, and quacks like a duck, it could still be anything but. Patterns found on surfaces at ancient sites that appear to be part of a game could also just be decoration. To differentiate, Crist looks at the social and spatial context. Was the graffito located in a public place, away from traffic, where games would traditionally be enjoyed? Was there evidence of socializing, like eating and drinking, which people commonly did while merrymaking?

Technology again is a huge help in piecing together the puzzle. Take one ancient Roman board for example. Although it was unlike anything that Browne and his team had seen before, 3D scans and x-ray fluorescence “revealed grooves consistent with those made when game pieces are repeatedly dragged along a stone board.” Whether it's a game is still not conclusive, but running it through the Ludii software next could make it so, and that would mean the board currently on display at the ThermenMuseum in the Netherlands may soon have a playable version online. It’s a promising preview of what’s to come in the next two years, and beyond.

Ultimately, the Maastricht researchers want to draw a fuller picture of how games evolved over time. The aim is for all this work to culminate in a family tree “going back as far as recorded history itself,” in addition to a growing world map. Board games didn’t just start at one point but rather in many places at different times, Browne says, and both genealogical efforts may help trace the paths they traveled and the ways they developed from one another. At the same time, the project is part of a larger cultural effort. Tracking how they dispersed throughout history can reveal how humans and cultures did the same. For example, seeing similar games played across vastly different geographic areas could point to how cultures interacted. As Crist puts it: “You have to have a movement of people to have a movement of games.”

Meanwhile, the Digital Ludeme Project is laying the foundation for bigger and better things. For one, the Ludii software, free for all to use and modify, has the potential to help game publishers and independent developers play-test and refine their games. Equally important is that it’s spearheading a new field of study: digital archaeoludology.

“We’re still in the experimental phase, but this project has taken on problems that had yet to be solved or even tackled before,” Browne says. “Now we’ve got the tools developed, we’ve got the evidence, and we’ll have even more results in the coming years. This could advance progress toward true artificial intelligence.”