source: Google DeepMind: From Atari to EVE Online: Building on 15 Years of AI Research in Games
level: research
google deepmind announced a research partnership with fenris creations, the studio behind eve online. the collaboration aims to develop ai agents that can handle continual learning, long-term memory, and complex multi-agent dynamics in a persistent game world. the work builds on deepmind's history of using games for ai breakthroughs, from atari to starcraft ii. the partnership includes eve online, eve vanguard, and eve frontier as distinct research environments.
the agents will be tested first in an offline instance of eve online, separate from live players. eve online has run since 2003 with a player-driven economy and thousands of star systems. deepmind's sima 2 agent already plays games like no man's sky using screen input and keyboard controls. the company says the eve universe demands capabilities beyond current models, such as reasoning over weeks or months. no timeline was given for moving agents into live player environments.
deepmind's earlier game research led to alphago, alphazero, and alphafold, which won a nobel prize in chemistry. the new partnership focuses on prototyping gameplay experiences rather than just mastering scores. fenris creations' ceo said the work explores how humans and ai can coexist in virtual worlds before facing similar questions in real life. deepmind also highlighted an existing aura guidance system in eve online that uses gemini to help new players.
why it matters: this research could produce ai agents that learn continuously and remember over long periods, which is essential for real-world applications like robotics and personal assistants.
source: Google DeepMind: From Atari to EVE Online: Building on 15 Years of AI Research in Games