From Atari to EVE Online: DeepMind reflects on 15 years of AI games research
dl_weekly · x · 2026-09-02
Google DeepMind has published a blog post tracing 15 years of games-driven AI research: from DQN learning 49 Atari games from raw pixels (the 2015 Nature paper that helped catalyze modern deep RL), through AlphaGo beating Lee Sedol in 2016, to StarCraft — each milestone producing more general, capable systems.
The post stresses that AI games research requires deep partnership with game developers, highlighting the major new research partnership with Fenris Creations and the EVE Universe unveiled earlier this year, which frames EVE Online as a testbed for continual learning, long-horizon memory and multi-agent dynamics. DeepMind is also working with studios like Hello Games, Coffee Stain and Foulball Hangover to prototype gameplay experiences that push both gaming and AI forward.
More from Companies & People
- Prediction Market: 51% Chance of New Claude Opus Release This Month — Polymarket · 2026-09-02
- Model choice is becoming an implementation detail, products should hide complexity — tylerbruno05 · 2026-09-02
- Decoding the Product and Market Signals Behind AI Lab Releases — alliekmiller · 2026-09-02
- DoorDash's moat: Proprietary data and models built on operational edge cases — ypatil125 · 2026-09-02
- MIT Review: Hugging Face hack hints at cultural issues at OpenAI — DavidSKrueger · 2026-09-02
- I Fine-tuned a 511M Parameter Model for Rust Code — StormworksVirtualAir · 2026-09-02