From Atari to EVE Online: Building on 15 Years of AI Research in Games
2026-08-26 · Google DeepMind
From Atari to EVE Online: Building on 15 Years of AI Research in Games
Gaming at the Heart of Google DeepMind
Since DeepMind’s foundation in 2010, the constrained yet rich worlds of games have played a critical role in understanding intelligence. They have driven some of the biggest AI breakthroughs and remain central to the organization’s work. Gaming is in Google DeepMind’s (GDM) DNA. Founder Demis Hassabis is a former game developer, as are many team members who bring decades of hands-on experience and deep respect for game craftsmanship.
DeepMind has consistently emphasized that meaningful AI research in games requires close partnership with game developers. This includes long-term collaborations with acclaimed studios such as Hello Games, Coffee Stain Studios, Foulball Hangover, and the major new research partnership with Fenris Creations and the EVE Universe announced earlier this year.
Games as the Engine of AI Research
The journey began with training a deep neural network to play Atari 2600 games directly from raw pixels. The Deep Q-Network (DQN) learned to play 49 different games — from Pong to Breakout to Space Invaders — without game-specific engineering. The 2015 Nature paper on DQN helped launch the modern era of deep reinforcement learning.
Subsequent work tackled increasingly complex games, producing more capable and general systems:
- AlphaGo (2016): Defeated world champion Lee Sedol in Go — a milestone experts believed was still a decade away.
- AlphaGo Zero: Surpassed previous versions by learning entirely from self-play with no human data.
- AlphaZero: Generalized the self-play approach to master chess, shogi, and Go with a single algorithm.
- MuZero: Learned to play games without even knowing the rules.
- AlphaStar (2019): Reached Grandmaster level in StarCraft II, handling real-time strategy, imperfect information, and high complexity.
In each case, the AI enriched the playing experience. AlphaGo’s famous Move 37 was so unexpected that commentators initially thought it was a mistake, overturning centuries of Go wisdom and inspiring new strategic exploration. AlphaZero similarly generated novel lines of play in chess.
The spirit of exploration developed in games had broader impact. The same foundations powered AlphaFold’s solution to the 50-year grand challenge of protein structure prediction, recognized with the 2024 Nobel Prize in Chemistry.
From Mastering Games to Understanding Them
Earlier systems showed AI could master any game given a clear objective and sufficient training. However, the real world lacks scores and rulebooks. This led to a fundamentally different question: can AI understand and interact with any game world the way a person would?
This question drives the SIMA (Scalable Instructable Multiworld Agent) project. Instead of optimizing scores, SIMA is a generalist agent that sees what a player sees on screen, understands natural language instructions, and acts using ordinary keyboard and mouse controls — without needing APIs or source code access.
Powered by Gemini frontier models, SIMA 2 functions as an interactive companion capable of real-time reasoning and conversation. It achieves human-like performance across complex 3D research environments and commercial games including No Man's Sky, Valheim, Hydroneer, and others.
Benefits for Game Development and New Gameplay
A truly general gaming agent would unlock powerful capabilities for existing games without code modification. Potential applications include:
- AI companions that genuinely understand the game world
- NPCs that adapt and respond dynamically, far beyond scripted behaviors
- Robust QA testing during development, even as the game changes with every commit
- Real-time adaptation to new content and unpredictable player behavior after launch, without constant re-scripting
Responsible Research Through Partnership
To develop SIMA safely and responsibly, DeepMind partners with leading game studios and maintains a growing portfolio of games for AI research. This enables increasingly complex challenges that may one day transfer to real-world problems.
Game studios contribute expert craft, extraordinary worlds, and deep player knowledge. DeepMind contributes frontier AI (including Gemini), research in generative interactive environments and embodied agents, plus the team’s unique game development background. Together, the focus is on discovering breakthrough, never-before-seen gameplay experiences that would be impossible without AI.
The approach is “show, don’t tell” — working hand-in-hand with developers to explore ideas and build playable prototypes to find what is genuinely fun.
A New Chapter with Fenris Creations and EVE Online
The latest research partnership with Fenris Creations, the independent studio behind the EVE Universe, represents a new chapter. Fenris Creations has spent more than two decades building one of the most extraordinary persistent worlds in gaming. EVE Online, launched in 2003, is a massively multiplayer space simulation where thousan