Google DeepMind Partners with EVE Online Developer for AI Research

Google’s AI-focused division, DeepMind, has taken a minority stake in Fenris Creations, the newly independent company behind the popular sci-fi simulation game EVE Online. This strategic partnership aims to leverage EVE Online as a rich environment for studying and developing advanced AI models that can handle complex, dynamic, player-driven systems.

The announcement comes on the heels of CCP Games (now Fenris Creations) securing a $120 million buyout from their former owners at South Korean publisher Pearl Abyss. The rebranded company will continue to operate as usual without any restructuring or layoffs, ensuring that EVE Online players can enjoy the game as they always have.

A Unique Testbed for AI

DeepMind and Fenris Creations believe that EVE Online offers a uniquely rich environment for AI research. The game's complex, player-driven economy and social dynamics provide an excellent testbed for developing AI systems capable of long-horizon planning, memory, and continual learning. These are critical skills for AI models to operate effectively in real-world scenarios.

• Long-Horizon Planning: EVE Online players often engage in strategic planning that spans weeks or even months. This requires AI models to think ahead and make decisions based on long-term goals.
• Memory: The game's persistent world means that actions have lasting consequences. AI systems need to remember past events and use this information to inform future decisions.
• Continual Learning: EVE Online is constantly evolving, with new content and player interactions shaping the game world. AI models must adapt and learn from these changes in real-time.

DeepMind will conduct controlled experiments using a specially designed offline version of the game running on a local server. This setup ensures that the research does not interfere with the online experience for EVE Online players. The two companies also plan to explore new gameplay experiences enabled by these advanced AI technologies, potentially enhancing the game's depth and complexity.

Under the Hood

Google DeepMind has a long history of using games as testbeds for machine learning models. Some notable examples include:

• Go: In 2016, DeepMind’s AlphaGo defeated world champion Lee Sedol, marking a significant breakthrough in AI's ability to handle complex board games.
• Atari VCS Games: DeepMind's agents have outperformed human players in a variety of Atari games, showcasing the versatility and adaptability of their reinforcement learning algorithms.
• StarCraft II: In 2019, DeepMind’s AlphaStar achieved grandmaster status in the real-time strategy game StarCraft II, demonstrating AI's capability to handle fast-paced, strategic gameplay.

More recently, DeepMind has been exploring "virtual world" models like Genie-3, which create real-time interactive simulations. These models help AI systems learn how to operate in physical reality by simulating complex environments and scenarios.

Fenris CEO Hilmar Veigar Pétursson emphasized the unique value of EVE Online as a research platform in an open letter to players:

"EVE is one of the few environments where questions about intelligence can be explored inside something that already behaves like a living world."

This partnership between DeepMind and Fenris Creations marks a significant step forward in AI research, combining the rich, dynamic environment of EVE Online with DeepMind's cutting-edge machine learning capabilities. The results could have far-reaching implications for both the gaming industry and the broader field of AI.

Key Takeaways

• DeepMind Investment: Google’s DeepMind has taken a minority stake in Fenris Creations, the company behind EVE Online.
• Research Goals: The partnership aims to study and develop AI models that can handle long-horizon planning, memory, and continual learning in complex, player-driven systems.
• Controlled Experiments: DeepMind will use a specially designed offline version of EVE Online for controlled experiments without impacting online players.
• New Gameplay: The companies plan to explore new gameplay experiences enabled by these advanced AI technologies.
• Historical Context: DeepMind has a history of using games like Go, Atari, and StarCraft II to advance machine learning research.