Google introduced an AI co-scientist for scientific discoveries

Google has developed a multi-agent system based on Gemini 2.0 that helps scientists generate new hypotheses and accelerate scientific and biomedical discoveries.

The AI co-author does not just analyze literature but also independently formulates original research hypotheses, evaluates them, refines them, and adapts them to specific goals. This is made possible by a coalition of specialized agents that operate according to the scientific method: generation, reflection, ranking, evolution, and so on.

The system has already demonstrated its effectiveness. In the field of biomedicine, the AI co-scientist predicted new drugs for treating acute myeloid leukemia. Experiments confirmed that the proposed compounds actually suppress the viability of cancer cells. The AI co-scientist also uses an Elo rating system for self-improvement—the longer it analyzes data, the more accurate its conclusions become. By several metrics, the system has already surpassed existing AI models and even some human experts.

Source: research.google