Why Google DeepMind Is Using EVE Online to Train AI, w/ CEO Hilmar Pétursson Artwork

AI-Curious with Jeff Wilser

Every week, Jeff Wilser sits down with the people building, breaking, and reckoning with AI — from the CEO of Upwork to the pioneer who coined "AGI" to an AI social network where bots wrote manifestos and had existential crises. Wilser is the author of eight books, AI keynote speaker, and the kind of interviewer who'd rather find the story no one's telling than rehash the headline everyone's read. Named by Inc. Magazine as one of the best ways to get AI-savvy. Included in UC Berkeley's data science curriculum.

All Episodes

AI-Curious with Jeff Wilser

Why Google DeepMind Is Using EVE Online to Train AI, w/ CEO Hilmar Pétursson

July 09, 2026 • Jeff Wilser • Season 2 • Episode 27

0:00 | 59:23

Can a 23-year-old video game help train the next generation of AI agents?

In this episode of AI-Curious, we talk with Hilmar Pétursson, CEO of Fenris Creations, the studio behind EVE Online, about why Google DeepMind has invested in the company and why EVE may be one of the hardest tests for artificial intelligence.

EVE Online is not just a space game. It is a persistent virtual civilization with player-run economies, alliances, corporations, spies, propaganda, logistics, years-long wars, and decades-old rivalries. That makes it a fascinating training ground for some of the biggest unsolved problems in AI: long-horizon planning, memory, continual learning, imperfect information, coordination, and agents that can operate inside messy real-world systems.

We explore why EVE is so much more complex than games like chess, Go, or StarCraft, and why Hilmar sees it as a possible “final boss” for AI in games. We also discuss how DeepMind could use EVE as a research environment, what an “AlphaEve” might eventually mean, and why virtual worlds may help AI systems learn in ways that resemble how humans and animals learn through play.

Along the way, we get into the business implications of AI agents, the parallels between EVE alliances and real-world companies, the idea of AI chief-of-staff agents, and the possibility of giving frontier AI labs like OpenAI, Anthropic, DeepMind, and xAI their own territories in EVE Frontier to see which AI best helps humans thrive. Hilmar also shares the story behind his conversations with Elon Musk, how EVE players are already using AI, and why he believes AI could help us unlock new versions of ourselves.

Guest:
Hilmar Pétursson — CEO, Fenris Creations

Check out:
Fenris Creations: https://fenris.com/
EVE Online: https://www.eveonline.com/

Follow AI-Curious on your favorite podcast platform:

Apple Podcasts
Spotify
YouTube
All Other Platforms

For anyone interested in Jeff’s AI Workshops for their company:

Reach out directly at jeff@jeffwilser.com

— Intro: Google DeepMind’s unconventional investment in EVE Online

— Why DeepMind sees EVE Online as one of AI’s hardest tests

— Why EVE is less a game than a simulation of civilization

— Alliances, spreadsheets, spies, betrayals, and 40,000-player coordination

— Why EVE’s complexity matters for real-world AI agents

— The AlphaGo comparison and why EVE may be a harder AI test

— The idea of giving frontier AI labs their own territories in EVE

— Hilmar’s correspondence with Elon Musk and xAI

— Hilmar joins AI-Curious

— What EVE Online is and why it is different from most games

— EVE’s single persistent world and 23 years of player history

— The game’s economy, strategy, spies, heists, and “five-dimensional chess”

— Why major EVE wars can take a year to prepare

— Decades-old rivalries and why EVE mirrors real-world conflict

— How Hilmar and Demis Hassabis first discussed AI and EVE

— AlphaGo, AlphaZero, AlphaStar, and DeepMind’s history with games

— Why EVE is much harder than chess, Go, or StarCraft

— EVE as a life you live, not a game you simply win

— False information, propaganda, psychology, and coordination inside EVE

— Why large language models made an “AlphaEve” more plausible

— The game outside the game: Discord, Slack, email, and real-world coordination

— Why the DeepMind partnership is still a research challenge, not a solved problem

— Long-term memory, imperfect information, and EVE as an AGI-style test

— Why EVE matters for business leaders, not just gamers

— EVE alliances as a model for complex organizations and AI agents

— The idea of an AI chief of staff for large-scale coordination

— Games as testing grounds for AI, from chess and Go to EVE

— Early AI guidance tools for new and veteran EVE players

— Long-horizon planning, memory, and continual learning

— Why mammals learn through play, and what that means for AI

— DeepMind’s history of using games as AI playgrounds

— DeepMind’s private EVE server and AI playground

— Why AI will be tested in a fake version of EVE before touching the live game

— EVE Frontier and lower-stakes experiments with AI agents

— Smarter NPCs, AI-run organizations, and helping new players find the magic of EVE

— AI as a sparring partner for human judgment

— The Black Mirror version: what if AI learns to manipulate or dominate a virtual society

— A frontier lab “bake-off” with OpenAI, Anthropic, DeepMind, and xAI

— Betrayal as a strategy and what EVE can teach us about trust

— The Guiding Hand Social Club heist and why EVE’s rules matter

— Hilmar’s conversations with Elon Musk and AI-native games

— How EVE players are already using AI tools

— How AI helps Fenris Creations manage legacy code, data, and complexity

— Hilmar’s prediction for AI, games, science fiction, and human potential