Google DeepMind

Generating Novel Scientific Hypotheses with Co-Scientist

The video introduces Co‑Scientist, DeepMind’s multi‑agent AI platform designed to accelerate scientific discovery by automatically scouring the literature and generating testable hypotheses. Researchers face a deluge of data—knowledge doubles every two months and over 17,000 rare diseases exist, with only 5% treated. Co‑Scientist tackles this by deploying specialized AI agents that retrieve papers, synthesize ideas, rank proposals, and iterate on hypotheses at a speed humans cannot match. In a demo on epigenomic drivers of liver fibrosis, the system produced a hypothesis that “took the presenter off his chair,” evaluating thousands of papers in days and suggesting drug candidates previously unseen. Demis Hassabis and other DeepMind scientists praised the rigor and creativity of the output, noting that several ideas have already led to published findings. If the technology scales, it could compress years of research into weeks, dramatically shortening the path from discovery to clinic, especially for underserved rare‑disease patients. The tool promises to give scientists “super‑powers,” reshaping the productivity of labs worldwide.