System from Stanford Turns Scientific Papers Into AI Assistants

Researchers at Stanford University have developed a system called Paper2Agent that turns scientific papers into functional toolkits for artificial intelligence agents. According to a report published in the journal Nature, the system packages a paper's code, methods, and data in a format that allows an AI assistant to perform new analyses.

James Zou, a computer scientist at Stanford University, describes the system as a virtual co-author that can be used seamlessly. During the testing phase, the team attempted to convert 100 computational biology papers and succeeded in 74 of them. The failures were noted to stem from missing code, data, or non-functional software. Paper2Agent toolkits demonstrated the ability to run analyses and answer result-oriented questions more accurately and at more than twice the speed compared to general-purpose agents using original papers.

Peter Clark from the Allen Institute for AI noted that the system makes research code easier to use, though it currently focuses on reusing methods rather than answering general open-ended questions. The system aims to move scientific knowledge from passive digital formats to active formats that facilitate collaboration and rapid data processing.