A game with more than two decades of virtual space combat, piracy and commerce is apparently fertile ground for training future AI models. Google DeepMind this week announced plans to use the popular roleplaying game Eve Online on Wednesday for training, Bloomberg reports.
Eve Online is a massively multiplayer online roleplaying game, or MMORPG, developed by Iceland’s CCP Games. Players explore and interact with each other in a vast universe, forming corporations and taking part in a wide range of activities like mining and space combat. DeepMind is also taking a minority stake in Fenris Creations, the new name for CCP Games that was also announced on Wednesday.
Although a minority stake, DeepMind’s investment is in the millions of dollars, Fenris Creations CEO Hilmar Veigar Pétursson told Bloomberg.
According to a company blog post, the developer will be independent, with its own board of directors, “giving us a more direct structure for the kind of far-reaching decisions that Eve requires.” It had been owned by South Korea’s Pearl Abyss, publisher of recent fantasy hit Crimson Desert, but was sold back to its management last week.
Eve Online launched 23 years ago and has gained a massive following, so it makes sense that the AI lab has taken an interest in training its models on the game.
While it’s currently unclear what type of player data Google is parsing through with its AI models, Eve Online is particularly notable for its emergent player-driven narratives and large-scale battles — many of which last multiple hours and involve virtual assets valued at hundreds of thousands of dollars. The game’s perceived real-world stakes inform player decision-making, which is perhaps what makes Google so interested in this particular MMO.
“Games have always been a huge part of my life — I’ve been a gamer since I was a kid, and I started my career designing and programming complex AI simulation games like Theme Park,” Google DeepMind CEO Demis Hassabis said in a statement. “They’ve also been at the heart of many of Google DeepMind’s breakthroughs — like Atari DQN, AlphaGo, AlphaStar and SIMA — because they’re the perfect training ground for developing and testing AI algorithms.”
Hassabis said the goal is to explore different gaming experiences and “advance AI research safely inside a player-driven universe as amazingly complex as Eve Online.”
According to the blog post, DeepMind will research player behavior on isolated servers of the game so as to not affect the live game itself. As time goes on, the research will likely expand, and Eve Online will also use the research to improve the game in the future.
