From Atari to EVE Online: Building on 15 Years of AI Research in Games
deepmind ·
Google DeepMind reflects on 15 years of AI research in games, from early Atari-playing systems to partnerships with modern game studios. The article highlights milestones like the Deep Q-Network, AlphaGo, and AlphaStar, noting how game-based breakthroughs also contributed to scientific advances such as AlphaFold. It introduces SIMA, a generalist agent that understands natural language and plays games through normal controls, aiming to work in any game without code changes. A new research partnership with Fenris Creations, the studio behind EVE Online, is presented as a major step for studying continual learning, memory, long-horizon planning, and multi-agent dynamics. The collaboration uses offline and separate game environments first, with the eventual goal of enriching live player experiences. DeepMind emphasizes that AI should act as a catalyst for new gameplay and real-world problem-solving, not as a replacement for players.
deepmind