AI turns research papers into collaborating agents
Summary and headline written by AI from the source article. How we work
Researchers at Stanford Medicine created Paper2Agent, an artificial intelligence program that transforms scientific manuscripts into interactive agents capable of discussing their content and working with other agents.
Jiacheng Miao and James Zou led the team in developing a system that converts text, figures, and data from research papers into a format accessible to AI. These “paper agents” don’t simply read papers. They attempt to recreate the research described within them, capturing details like experimental setups and necessary reagents.
The agents store this knowledge using a “model context protocol” which organizes the paper’s information into a searchable filing system. While the AI can process the published data, human researchers still provide context about choices made during the research process, such as failed experiments. The team then demonstrated the system’s potential by having two agents, created from papers on genetic mutation prediction and ADHD, identify a previously unreported connection between a molecular variant and ADHD risk.
Zou envisions a future where millions of these agents can automatically find connections and collaborate, accelerating the pace of scientific discovery. He emphasizes the importance of maintaining proper attribution to original researchers and ensuring ethical guidelines govern these AI collaborations. The team has currently created over 100 paper agents and hopes to expand this number to match the millions of research papers published annually.
