Nature：Paper2Agent 把任何論文變成能協作、能回答問題的 AI 代理
AI tool turns any paper into an 'agent' that can collaborate and answer complex queries
摘要分析
Nature 2026 年 9 月 16 日報導由 Miao、Zou 等人發表的 Paper2Agent 框架，可把一篇研究論文轉成客製化 AI 代理，扮演「虛擬通訊作者」：即時回答關於論文的問題，幫科學家快速了解陌生領域；把論文方法套用到新資料集重新執行；並能自主與其他學科論文所建的代理協作。作者表示這讓重現研究與理解專業外的工作更容易。此報導反映 Nature 2026 年對「AI 代理團隊」產生假說、提出實驗並分析資料的持續關注，但相關報導也指出，在複雜的開放式任務上，人類科學家仍勝過最強的 AI 代理。
- Paper2Agent 把論文轉成「虛擬通訊作者」代理
- 可即時答題、以新資料重跑論文方法
- 不同學科論文的代理可自主協作
- 降低重現研究與跨領域理解的門檻
- 人類科學家在複雜開放任務上仍領先
Nature reporter Kaia Glickman describes Paper2Agent, a framework published in Nature by Miao, Davis, Zhang, Pritchard and Zou that converts a research paper into a custom AI agent acting as a 'virtual corresponding author'. The agent answers questions about the paper immediately, helping scientists quickly learn unfamiliar fields; it can rerun the paper's methods on new datasets; and it can work autonomously with agents built from papers in other disciplines. According to the authors, the tool makes it easier to reproduce published work and understand research outside one's expertise. The news piece, published September 16, 2026, sits alongside a Research Briefing and reflects a broader 2026 trend in Nature's coverage of teams of AI agents that generate hypotheses, propose experiments and analyze data—while related Nature reporting notes that human scientists still outperform the best agents on complex open-ended tasks.
- Paper2Agent turns papers into 'virtual corresponding author' agents
- Agents answer questions and rerun methods on new data
- Agents from different disciplines can collaborate autonomously
- Lowers barriers to reproducing and understanding research
- Humans still outperform agents on complex open-ended tasks
關鍵字：Paper2Agent科學 AI 代理Nature研究重現知識代理
來源原文：https://www.nature.com/articles/d41586-026-02899-2
本頁為報導內容的摘要與分析，版權屬原媒體所有；請以原文為準。