DeepMind partners with EVE Online's studio to test AI in a living world

Google DeepMind announced a new research partnership with Fenris Creations, the studio behind EVE Online, to prototype AI research inside one of gaming's oldest continuously running virtual worlds. The lab laid out the collaboration in a blog post titled "From Atari to EVE Online: Building on 15 Years of AI Research in Games," published August 21.

The partnership extends a research line that stretches back to 2015, when DeepMind's Deep Q-Network learned to play 49 different Atari 2600 games directly from raw pixels, with no game-specific engineering. That paper helped catalyze the modern era of deep reinforcement learning. AlphaGo defeated world champion Go player Lee Sae Dol in 2016, a feat many experts thought was still a decade away, and its Move 37 stunned commentators who initially read it as a mistake before it inspired new lines of strategy. AlphaGo Zero learned entirely from self-play with no human data, AlphaZero generalized the approach to chess and shogi, MuZero learned to play without knowing the rules at all, and AlphaStar reached Grandmaster level in StarCraft II in 2019, handling real-time complexity and imperfect information. DeepMind credits this entire lineage with laying the groundwork for AlphaFold's breakthrough on the 50-year protein structure prediction challenge, work that won the 2024 Nobel Prize in Chemistry.

More recent work shifted from mastering games toward understanding them. SIMA, DeepMind's Scalable Instructable Multiworld Agent, sees what a player sees on screen, understands natural language instructions, and acts through ordinary keyboard and mouse controls, with no API or source code access required. SIMA 2, powered by DeepMind's Gemini models, acts as an interactive companion capable of real-time reasoning and conversation, playing at a human-like level across environments including No Man's Sky, Valheim, and Hydroneer.

• DQN (2015): learned 49 Atari games from raw pixels
• AlphaGo (2016): defeated world champion Lee Sae Dol
• AlphaStar (2019): reached Grandmaster level in StarCraft II
• AlphaFold (2024 Nobel Prize): built on the same research lineage
• EVE Online: continuously running since 2003, over 20 years

EVE Online launched in 2003 as a massively multiplayer space simulation where thousands of players share a single persistent universe, with a player-driven economy, real supply-and-demand dynamics, and trade networks spanning thousands of star systems. DeepMind frames it as an ideal environment for four capabilities central to its frontier research: continual learning without forgetting past skills, memory that spans timescales longer than today's model context windows, long-horizon planning across weeks or months, and complex multi-agent dynamics involving cooperation, competition, negotiation, and emergent social behavior at scale.

The collaboration spans three environments inside Fenris Creations' universe. EVE Online itself offers a large-scale, single-shard persistent world. EVE Vanguard shifts to a first-person perspective with ground-level tactical decisions. EVE Frontier introduces programmable Smart Assemblies inside an open, extensible architecture where the rules of the world can change, forcing agents to adapt to entirely new mechanics rather than a fixed rule set.

Together with Google DeepMind, we're pushing into uncharted territory where AI must learn, adapt and remember on timescales that no other game environment demands, while helping us understand how humans and AI can coexist in a virtual environment before we have to contend with the same questions in real life.

One result has already shipped to players. The Aura Guidance system uses Gemini to surface player-generated knowledge, drawn from real Rookie Help questions and answers, to help new pilots get oriented. The broader research roadmap moves in stages: it starts with an offline instance of EVE Online kept separate from live players, progresses through EVE Frontier to study how humans and agents can coexist in a persistent, open-ended world, and only brings mature capabilities into the live EVE Online and EVE Vanguard games once they clear that bar, with the stated goal of enriching human play rather than replacing it.

DeepMind co-founder Demis Hassabis is a former game developer himself, as are many members of the games research team, and the lab has run similar collaborations with studios including Hello Games, Coffee Stain Studios, and Foulball Hangover.

Nothing here should be taken as financial advice — just information to consider.