From Atari to EVE Online: Building on 15 Years of AI Research in Games

Games have been central to major AI breakthroughs over the past 15 years, from teaching AI to play Atari games to mastering Go and chess. A research organization is now developing a general gaming agent that can understand and play games the way humans do, using natural language instructions and standard controls, without needing access to game code. This agent could create new gameplay experiences like AI companions and adaptive non-player characters, and could improve game development through robust testing and real-time adaptation. The organization has partnered with game developers and studios to test these capabilities in complex game worlds, including a major partnership focused on a large-scale multiplayer space simulation with a persistent player-driven economy.