Google introduces a new AI assistant for biomedical research to help scientists reach hypotheses and insights faster using advanced models. The software, called "Co-Scientist," can identify gaps in scientific knowledge, develop related questions, and suggest solutions in conjunction with existing databases. Initial tests at Stanford University, Imperial College London, and Houston Methodist Hospital show that the AI can generate novel ideas in just a few days, which would normally take research teams years. The assistant independently arrived at the same theory of a novel gene transfer as a team at the British elite university – even before their peer-review was completed.

For the application, Co-Scientist scours freely accessible papers and databases, combines insights with AlphaFold, the in-house protein folding tool, and provides precise solution approaches along with sources. Particularly convincing was an experiment at Stanford University, where the AI system identified proposals for repurposed drugs for liver fibrosis. "Thus, AI becomes a potential accelerator in medicine," said Google's research director Alan Karthikesalingam. The secret of the tool lies in various divided "AI Agents," which consist of generating, reflecting, and evaluating ideas—a process considered a reflection of human research methodology.

In the competitive environment where OpenAI, Perplexity, and BioNTech's AI offshoot InstaDeep are also experimenting with similar solutions, Google is focusing on particularly close collaboration with leading scientific institutions. According to José Penadés from Imperial College, this could "change the entire approach to scientific problems." The next step is now broader availability: Google wants to further develop the tool to make it usable for more researchers and thus effectively address the increasing information floods in many fields.