• Google’s AI Co-Scientist was positioned as a groundbreaking research collaborator, but closer examination shows it functions more like a smart assistant that repackages known information.
• In the case studies on liver fibrosis and antimicrobial resistance, the AI’s so-called “novel” findings were already present in the source materials it was given—just paraphrased in different language.
• The system focuses on hypothesis generation, yet that’s not the real bottleneck in biomedical research. Scientists need help testing and interpreting complex data, not generating more ideas.
• Even with a sophisticated architecture of multiple agents, the system lacks the creative reasoning and judgment required to contribute meaningfully to scientific discovery.
• Labeling tools like this as “co-scientists” sets unrealistic expectations. This whitepaper argues for a more grounded, transparent, and useful role for AI in research workflows.

‍