跳转至

观点：科学团队中的AI智能体应作为人机系统进行研究

文章背景与核心概要

随着基于大语言模型（LLM）的智能体越来越多地作为积极的协作者参与科学发现，当前的研究大多聚焦于独立“AI科学家”的自主能力。然而，这篇观点文章指出，这种方法忽视了科学团队协作中至关重要的社会动态。作者提出，应当将AI科学家视为人机系统（Human-Agent Systems, HAS）来进行研究，其中分析的核心单位应该是“人-智能体”对。

通过文献综述与实证分析，本文强调了在不考虑人机动态的情况下部署智能体所带来的近期风险——例如科学探究多样性的降低。通过对真实世界案例研究的剖析，作者展示了人类与智能体如何能够实现能力的有效互补，并呼吁开展新的研究以建立数学框架，从而在科学发现中促进人机协同（Human-AI synergy）。

摘要 (Summary)

As large language model (LLM)-based agents are increasingly deployed as active collaborators in scientific discovery, current research largely focuses on the autonomous capabilities of standalone "AI Scientists." This position paper argues that such an approach overlooks the crucial social dynamics of scientific teamwork. The authors posit that AI Scientists should instead be studied as Human-Agent Systems (HAS), where the primary unit of analysis is the human-agent pair.

Through literature review and empirical analysis, the paper highlights near-term risks—such as a reduced diversity of scientific inquiry—when agents are deployed without accounting for human-agent dynamics. By analyzing real-world case studies, the authors demonstrate how humans and agents can effectively augment each other and call for new research to establish mathematical frameworks that foster human-AI synergy in scientific discovery.

随着基于大语言模型（LLM）的智能体日益被部署为科学发现中的活跃协作者，当前的大多数研究主要聚焦于独立的“AI科学家”的自主能力。这篇观点文章认为，这种方法忽视了科学团队协作中的社会学层面，并将AI科学家作为人机系统（HAS，其分析单位为人-智能体对）进行研究既未得到充分开发也未受到足够重视。我们通过文献和实证分析确立了这些观点，并强调了最近的事件和研究：如果在部署科学智能体时不考虑人机动态，将引入近期风险，包括降低科学探究的多样性。通过对真实世界案例研究的分析，我们表明科学家与智能体能够相互增强彼此的能力。我们呼吁开展采用HAS视角的新研究，以开发用于理解和促进科学发现中人机协同的数学框架。

元数据 (Metadata)

  • arXiv ID: arXiv:2608.14667 [cs.AI]
  • 学科分类 (Subjects): 人工智能 (cs.AI); 人机交互 (cs.HC)
  • 提交日期 (Submission Date): 2026年8月2日
  • 会议背景 (Conference Context): 已被 COLM 第二届科学发现语言模型研讨会（COLM 2nd Workshop on Language Models for Scientific Discovery）接受
  • 开源许可 (License): 知识共享署名 4.0 国际许可协议 license icon查看许可

作者 (Authors)

  • Patrick Emami
  • Sameera Horawalavithana
  • Truc Nguyen
  • Gihan Panapitiya
  • Bruno Jacob
  • Siddhisanket Raskar
  • Saumya Sinha
  • Jared D. Willard
  • Andrew Glaws
  • Nithin Somasekharan
  • Ling Yue
  • Brian Lu
  • Shaowu Pan
  • Jason Eisner
  • Patrick Emami
  • Sameera Horawalavithana
  • Truc Nguyen
  • Gihan Panapitiya
  • Bruno Jacob
  • Siddhisanket Raskar
  • Saumya Sinha
  • Jared D. Willard
  • Andrew Glaws
  • Nithin Somasekharan
  • Ling Yue
  • Brian Lu
  • Shaowu Pan
  • Jason Eisner

摘要正文 (Abstract)

Large language model-based agents are increasingly deployed as collaborators in scientific discovery yet most current work focuses on the autonomous capabilities of "AI Scientists". We argue that this overlooks the social aspects of scientific teamwork, and that studying AI Scientists as human-agent systems (HAS)--where the unit of analysis is the human-agent pair--is both underexplored and undervalued. We establish these points through literature and empirical analysis, and highlight recent incidences and studies which show that deploying agents in science without accounting for human-agent dynamics introduces near-term risks, including reduced diversity of scientific inquiry. Through analysis of real-world case studies, we show that scientists and agents can augment each other's capabilities. We call for new research that adopts the HAS lens to develop mathematical frameworks for understanding and fostering human-AI synergy in scientific discovery.

基于大语言模型的智能体正日益被部署为科学发现中的协作者，然而当前绝大多数工作都集中在“AI科学家”的自主能力上。我们认为，这忽视了科学团队协作的社会属性，将AI科学家作为人机系统（HAS——其中分析单位为人-智能体对）进行研究既未得到充分挖掘，也未受到足够重视。我们通过文献和实证分析确立了这些观点，并强调了近期的事件和研究，这些研究表明，在不考虑人机动态的情况下在科学领域部署智能体会带来近期风险，包括减少科学探究的多样性。通过对现实世界案例研究的分析，我们展示了科学家与智能体能够相互增强各自的能力。我们呼吁开展采用HAS视角的新研究，以开发数学框架，用于理解和促进科学发现中的人机协同。

访问与资源 (Access & Resources)