Google tests AI that revises molecular designs after experiments reveal a wrong turn
Google had AI design short DNA strands meant to grab one particular protein, with researchers running the experiments and feeding results back. One round looked better but stuck to everything; control tests caught it, the AI changed course and the next round genuinely improved. People still did all the lab work.
Sources
- Google Research · Research announcement, 16 October 2026
- Cutts et al. · Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery
- Delley et al. · Semantic Representation for Agentic Self-Play in Automated Laboratories
- Dai et al. · Autonomous mobile robots for exploratory synthetic chemistry (Nature, 2024)