Google DeepMind's "Co-Scientist"

A multi-agent LLM system from Google DeepMind designed to generate, rank, and iteratively refine novel research hypotheses in collaboration with a human scientist, described in Gottweis et al. (2026, Nature).

Matters to this vault as a contemporary, published comparator in the automated-science lineage: a 2026 multi-agent NV-diamond research paper (claim-mansouri-2026-pi-agent-rejected-75-proposals-manager-broke-stall) cites and implicitly contrasts itself against Co-Scientist alongside Sakana's "The AI Scientist" — one a hypothesis-ranking collaborator for human-led research, the other (that paper's own Auctopus system) an end-to-end agent laboratory that runs physical experiments without a human in the loop.

References

• Capture: 10-inbox/raw/2026-11-01-outside-arxiv261040589-collaborative-ai-agents-for-autonomous-experimental.md