AI-rewritten: This is a summary of an article from Singularity Hub, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

James Zou of Stanford University introduced Paper2Agent, a system designed to transform static scientific papers into interactive AI agents. Unlike standard chatbots that summarize text, these agents are trained on a paper’s full content, including figures and data, to reproduce its experimental results in a virtual environment. This process allows the AI to act as a "virtual author" capable of answering complex questions, applying methods from one study to new datasets, and collaborating with agents from other disciplines.

The project aims to address challenges like the overwhelming volume of scientific publishing and low reproducibility rates. While only 39 percent of psychology studies in a 2015 project yielded consistent results, Paper2Agent attempts to verify claims by running experiments. The system scans papers to capture details like experimental setups and stores them in a digital vault called an MCP server. This enables scientists to link their own large language models to the agent for deeper analysis rather than simple summarization.

In testing, the tool successfully converted 76 percent of 110 papers into working agents, with failures often indicating missing code or data. One agent based on AlphaGenome answered genetics questions with near-perfect accuracy and was significantly faster than other AI scientists. Furthermore, multiple agents collaborated to identify new genetic variants linked to conditions like psoriasis and cholesterol issues, demonstrating how cross-disciplinary research could be automated without human matchmaking.