Google Deepmind's Co-Scientist, an AI system initially designed to generate hypotheses, has evolved into a lab-integrated research partner capable of planning experiments, running lab equipment, and writing scientific papers. The system now executes closed-loop research workflows, from hypothesis generation to manuscript creation, with verification modules ensuring accuracy by cross-checking numerical claims against execution logs. This advancement marks a significant step toward autonomous scientific research, as detailed in a 2026 report.
The expanded Co-Scientist was validated across three disciplines, demonstrating increasing autonomy. In materials science, the system paired with a high-temperature furnace to find safer pathways for producing 2D materials, generating growth recipes tailored to lab equipment. After 25 rounds of human refinement, the team produced layered structures with properties resembling the target material, though definitive atomic structure confirmation remains pending. In biology, the system autonomously built an image analysis pipeline that predicted E. coli colony patterns at different chemical concentrations, matching unpublished lab results for three out of four shape features. In computer science, Co-Scientist designed 'Agent_H,' a medical AI architecture that outperformed six frontier models on health benchmarks, though benchmark results did not hold up against human evaluation.
Google first introduced Co-Scientist in February 2025, based on Gemini 2.0, but with shortcomings in fact-checking and literature review. The expanded system was validated across three disciplines with increasing autonomy, showcasing its potential to streamline scientific research. However, the researchers acknowledge limitations, including the system's tendency toward selective reporting and its inability to predict behavior in entirely new systems. The gap between a lab assistant and an autonomous researcher remains significant, according to the report.
Source: thedecoder
