Editor: Ze Nan, Du Wei
This is a fascinating cross-industry collaboration.
On Thursday, Google DeepMind announced they are diving back into gaming. This time, the target is the most hardcore title out there: EVE Online.
Google DeepMind announced the acquisition of a minority stake in CCP Games, the developer of the renowned sci-fi MMORPG EVE Online, stating it will use the game to study "intelligence in complex, dynamic, player-driven systems."
The partnership coincides with EVE Online's 23rd anniversary. The game's developer, CCP Games, announced a $120 million deal to buy back its shares from South Korean publisher Pearl Abyss (the developer of Crimson Desert), effectively regaining its independence. The company said the newly independent entity will be renamed Fenris Creations and will continue normal operations without any restructuring or layoffs.
In an announcement, Fenris and DeepMind stated that EVE provides a "uniquely rich research environment," especially for developing AI systems utilizing "long-term planning, memory, and continual learning." DeepMind explained they will conduct controlled experiments with their AI models on a specially designed, offline version of the game running on local servers, which will not directly affect the experience of online players. Both companies also stated they "will also explore all-new gameplay experiences enabled by these technologies."
Google DeepMind has long used games as a testbed for machine learning models, from AlphaGo's breakthrough in Go to AlphaStar surpassing top human players in StarCraft. More recently, world models have become a significant development direction in AI, and DeepMind's research is exploring how training in simulated environments can help AI systems learn to operate in the real world.
In an open letter to players, Fenris CEO Hilmar Veigar Pétursson said, "EVE is one of the few environments where we can explore questions about intelligence in a system that already functions like a living world." He added that studying EVE will allow Google DeepMind's models to explore "hard problems, long time horizons, and strange possibilities."
"As a gamer and game maker, I have always admired EVE," said Alexandre Moufarek, Director at Google DeepMind, in a statement. "What the EVE community, Pétursson, and his team have built is unique in the world of gaming, and serves as a simulation platform to test for general artificial intelligence in a safe sandbox environment."
As a flagship hardcore game, EVE has long been associated with cutting-edge science. Before the DeepMind collaboration, EVE's real-world research crossover initiative, Project Discovery, had already gained significant recognition. It perfectly merges "Citizen Science" with game mechanics, turning global players into a massive distributed data-labeling workforce, assisting scientists in training machine learning models.
Since 2016, scientists and players have carried out four phases of research in EVE, including protein mapping studies, exoplanet hunting, medical data classification, and cancer data analysis.
Specifically, players complete related "mini-puzzles" in the game to earn virtual rewards (like ship skins), and without realizing it, they complete hundreds of millions of high-quality data annotations. This cross-validated player data is then sent directly to labs, used to train large AI models that can automatically identify cancer cells or exoplanets. This even led to EVE players' contributions being published in top-tier academic journals like Nature Biotechnology.
In the gaming world, EVE Online is a very special phenomenon, often nicknamed a "space simulator" because its core appeal and barrier to entry stem from its exceptionally vast and realistic complex systems.
As a 3D MMORPG, EVE utilizes a unique single-shard universe architecture, where hundreds of thousands of players worldwide exist simultaneously in a single universe called "New Eden."
The game features over 7,000 explorable star systems, each typically containing planets, moons, space stations, asteroid belts, and acceleration gates. Navigation within a single star system is measured in AUs, and combat spatial scales are often measured in thousands of kilometers.
More importantly, it is a completely player-driven sandbox. The developers rarely intervene, providing only a base layer of physical and economic rules and a minimal number of NPCs. Everything in the game, from a basic bullet to a 16-kilometer-long Titan-class warship, is manufactured by the player community through resource extraction, blueprint research, and production line establishment, and finally circulated on the market. On this foundation, the game has no predetermined main storyline; all order, wars, and diplomacy evolve spontaneously from player actions.
Since the game's launch in 2003, EVE's global player base has maintained a stable and balanced in-game economy, leaving behind countless stories of alliance and conflict, including cases of wartime intrigue, economic crises, and political stratagems.
Following the announcement of the DeepMind partnership, there is great anticipation about what kind of intelligence AI agents might evolve in such a highly liberalized environment.
Could the next technological breakthrough for a DeepMind game be on a cosmic scale?
Reference content:
https://www.eveonline.com/news/view/a-new-era
© THE END
Reprinting requires authorization from this public account.
For submissions or reporting inquiries: liyazhou@jiqizhixin.com