Google DeepMind to train AI in war and diplomacy inside EVE Online

Google DeepMind has entered a research partnership with Fenris Creations, the developer of EVE Online, and will begin training AI models in one of the most complex MMORPGs. This week the studio split off from its owner Pearl Abyss for $120 million and at the same time received an investment from Google.

Why EVE Online is an ideal lab for AI

EVE Online is not just a game about space. It is a full simulation with its own economy, currency, fully fledged player corporations (with hierarchies, diplomacy and wars), a dynamic market and political alliances. Thousands of people live in this world at once: they trade, build coalitions, divide power and plan operations.

For AI this is an ideal sandbox, because the environment is unpredictable. There is no single correct solution, and constant adaptation is required. An agent interacts with other AIs and with players at the same time, a mistake can be costly, and success requires understanding motives and negotiating.

In ordinary simulations, AI learns either to win or to optimize a single metric. In EVE Online it has to do everything at once: compete, cooperate, trade, fight and survive.

What DeepMind's AI agents will learn

The research program will cover several areas at once:

• Cooperation: how to act in concert with other agents to reach a goal
• Conflict and war: strategic planning under opposition
• Economics: managing resources, trading, balancing supply and demand
• Diplomacy: negotiating, building alliances, agreeing on cooperation
• Adaptation: changing tactics quickly when market conditions shift

The experiments will run on a local (offline) server, an isolated copy of EVE Online with its own economy and rules. This lets DeepMind iterate fast: run a model, check the result, roll back failed versions, without risk of disturbing the live economy of the main game, where real people spend their time and money.

How Google positioned itself strategically

Google's investment in Fenris Creations is not just sponsorship of academic research. The company received a minority stake in the developer itself, which guarantees it long-term access to the environment and influence over its development. Fenris Creations, meanwhile, has fully retained its independence: management, teams, and the development plans for EVE Online, EVE Vanguard and the new game EVE Frontier remain in the studio's hands.

Google does not control the studio, but it has secured its place as a strategic partner in AI research. It is a typical Big Tech move: the company invests in the future while also obtaining research rights.

What this means for AI and games

Games are increasingly becoming laboratories for AI research. Complex economic systems, social interactions, multi-agent scenarios and unpredictability are all needed to prepare AI agents for real-world tasks. If an AI can wage war, trade and negotiate in EVE Online, it may be able to handle more abstract human scenarios too.

EVE Online is a fully fledged, hypercomplex system that reflects reality better than any artificial task. For players themselves there is an interesting consequence: in the future they will meet AI agents inside the game, competing on equal terms with them.