Two Nature Papers: AI Makes Validated Medical Discoveries Autonomously
Two peer-reviewed Nature papers published this week document autonomous AI systems making laboratory-validated medical discoveries. Google's co-scientist, a multi-agent virtual lab using an ELO tournament of debated hypotheses, identified and validated novel drug repurposing candidates for AML leukemia (including Cur-61, 18× more effective at killing dormant leukemia stem cells), antimicrobial resistance, and macular degeneration — rated above human experts in blind evaluation. Separately, the Robin system ran a complete drug discovery loop — synthesizing 551 papers, executing code on raw lab data, and producing better drug candidates — in under 2 hours at a cost of $10.76, versus approximately 400 human hours for equivalent work.
Why It Matters
Two independent Nature-quality validations in the same week establish a new benchmark for AI in scientific discovery: autonomous hypothesis generation and laboratory-validated drug finding is no longer a demonstration — it is a reproducible, peer-reviewed result.