A persistent universe that has been live for more than two decades is becoming an AI proving ground. Google DeepMind published the full account of its games research arc on August 21, ending at its partnership with Fenris Creations, the independent studio behind EVE Online, and the staged program the two will run inside that world.
The retrospective walks the milestones: the Deep Q-Network mastering 49 Atari games from raw pixels, documented in a 2015 Nature paper, AlphaGo’s 2016 defeat of Lee Sedol, then AlphaGo Zero, AlphaZero, and MuZero, each removing more human priors.
The lab is shifting from systems that dominate a single game with a clear score to agents that operate in any world the way a person would. The vehicle is SIMA, which sees what a player sees and acts through ordinary keyboard and mouse input. SIMA 2, released in November with a Gemini core, added goal reasoning, conversation, and self-directed practice.
EVE Online’s single-shard economy has run since 2003, holding thousands of players in one market with real supply and demand and player-built politics. DeepMind says the environment should exercise continual learning, memory beyond context windows, planning over weeks or months, and complex multi-agent dynamics.
Fenris CEO Hilmar Petursson said the partnership pushes into territory where AI must learn and remember on timescales no other game demands. The program spans EVE Online, the tactical EVE Vanguard, and EVE Frontier, where programmable Smart Assemblies let the rules change.
