Google DeepMind Turns EVE Online Into an AI Research Lab
EVE Online's persistent, player-driven universe gives DeepMind a testbed for AI that must plan over weeks or years, a departure from match-based games with clear goals.
Reporting from 1 source: GIGAZINE.
Google DeepMind announced a research partnership with Fenris Creations, the developer of EVE Online, to study AI in the long-running space MMORPG. The persistent universe, active for over 20 years, offers an environment without resets or clear win conditions. Research targets continual learning, memory, long-term planning, and handling cooperation and competition.
The partnership moves DeepMind past the reset-based games that defined its early work, from the Atari titles learned by DQN to AlphaStar in StarCraft II. In EVE Online, no match ends and no score resets: a single universe shared by many players has run since 2003, with economies, wars, and diplomacy shifting constantly.
Fenris Creations said research will initially run in a controlled environment using an offline version of the game. The work builds on SIMA 2, the Gemini-based agent that operates a keyboard and mouse by reading the screen, and that has already been tested in No Man's Sky, Valheim, and Hydroneer.
DeepMind lists four abilities it wants to study: continual learning without losing prior knowledge, memory that accumulates and retrieves information over long spans, planning measured in weeks to years, and handling cooperation, competition, and negotiation with many other actors.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.