AI技术正悄然改变学术研究的协作方式

大语言模型等AI工具虽然提升了个人研究生产力，但可能导致学术研究变得日益孤立。1这一现象被称为"Waymo效应"——当技术消除了与他人互动的摩擦时，我们体验到纯粹的收益，同时忽视了这些摩擦所隐含的价值。1大语言模型恰似"知识生活的Waymo"，随处可得、无需协商议程、不涉及自我管理，使研究人员能够绕过协作的"不便"。1然而，他人思维中的这种"不便"——提出意外挑战和不同视角——正是研究创新的关键来源。1

学术激励体系在推动这一趋势中扮演了重要角色。1资金压力、对速度的崇拜以及无需协商作者顺序等因素共同推动了"去协作化"。1研究表明，小团队更易产生破坏性创新，大团队则倾向于渐进式发展，而跨领域的非典型配对往往能产出最具新颖性的工作。1然而，当研究人员将写作外包给AI时，他们实际上是在跳过思维过程本身，因为写作与思考是不可分离的。1

应对这一挑战需要重新审视研究机构的结构与激励机制。1有研究者提出"船长制科学"概念，即研究者为船长，AI为船员，人类必须保持权威地位。1这一框架呼应了自动化的历史教训——1983年，自动化专家Lisanne Bainbridge就曾指出，系统运行越可靠，人类操作者在紧急情况下所需的技能流失越快。1

A phenomenon termed the "Waymo effect" describes how technology that eliminates friction in human interaction can paradoxically undermine the value it seeks to enhance.1 The concept, drawn from the experience of autonomous vehicles, illustrates how large language models function as an "intellectual life Waymo"—constantly available, free of competing agendas, and requiring no negotiation or compromise.1 Yet this convenience comes at a hidden cost: the very friction that AI removes often carries unrecognized benefits for research and discovery.

Collaboration in academic work derives much of its value from what observers describe as the "inconvenience" of engaging with others' minds—the unexpected challenges, alternative perspectives, and friction that emerge when researchers must negotiate ideas across different viewpoints.1 However, current incentive structures within academia actively discourage such collaboration through three converging forces: funding pressures that reward speed, a cultural emphasis on velocity over depth, and the ability to sidestep negotiations over author attribution.1 Research by Dashun Wang suggests an alternative framework: "pilot-in-command science," in which human researchers maintain authority while AI serves in a supporting role, preserving essential human oversight.1

The structural patterns of collaborative research reveal further complications. Small teams tend to produce disruptive innovations, while large teams drive incremental progress; the most novel work emerges from cross-disciplinary pairings that defy conventional collaboration patterns.1 Writing and thinking are inseparable processes, such that outsourcing the writing stage effectively bypasses the intellectual work itself rather than merely accelerating it.1 This dynamic echoes a 1983 observation by Lisanne Bainbridge on automation's irony: as systems become more reliable, human operators lose the skills needed to respond when emergencies arise.1 Research institutions must therefore treat collaboration as foundational infrastructure and restructure incentives accordingly, rather than allowing AI-driven efficiency to erode the collaborative foundations of knowledge production.