Google unveils an AI co-scientist whose drug ideas for leukemia held up in lab tests

Google DeepMind/AI co-scientist

Google introduced a multi-agent system on Gemini 2.0 that generates, debates and ranks research hypotheses. In early collaborations, its suggestions were tested in real labs: repurposed drugs slowed acute myeloid leukemia cells, and proposed targets showed anti-fibrotic activity in liver organoids.

• Six agents (generation, reflection, ranking, evolution, proximity, meta-review) under a supervisor.
• Ranking uses an Elo tournament; higher-rated ideas correlated with correct answers on hard science questions.
• Independently proposed a bacterial gene-transfer mechanism Imperial College had found but not yet published.
• Access opened to research groups through a Trusted Tester program.