Descriptions:
Google DeepMind’s Co-Scientist is a multi-agent AI system purpose-built for scientific discovery, and this video brings together DeepMind researchers and external scientists to explain how it works and what it has already produced. Unlike a general-purpose language model, Co-Scientist runs specialized agents in parallel: some scour the literature, others generate and evolve hypotheses, a dedicated agent ranks and compares ideas, and the system synthesizes connections across previously separate research fields.
A researcher studying liver fibrosis describes prompting the system to explore epigenomic treatment approaches. Co-Scientist processed tens of thousands of papers and tested thousands of hypotheses over days — work that would otherwise take months — and surfaced a novel, rigorous hypothesis the team could not find a way to disprove. The result accelerated lab experiments and contributed to published findings, with more in progress. The system can effectively deploy what one researcher describes as “50 scientists in one day.”
The video highlights that knowledge at the frontier of science now doubles roughly every two months, making comprehensive literature coverage nearly impossible for individual researchers. Co-Scientist is framed as infrastructure for that gap — not a replacement for scientists but a tool that dramatically compresses the time from hypothesis to testable experiment. Google DeepMind describes the trajectory as moving from moonshot to mission, with the system already in active use across multiple research domains.
📺 Source: Google DeepMind · Published May 19, 2026
🏷️ Format: Showcase
