[论文解读] AI Researchers' Views on Automating AI R&D and Intelligence Explosions

Severin Field, Roy Douglas+1 more|arXiv (Cornell University)|Feb 13, 2026

Ethics and Social Impacts of AI被引用 0

一句话总结

本文对25位前沿AI研究人员进行了半结构化访谈，围绕ASARA（AI Systems for AI R&D Automation）、风险、时间表、治理与部署等话题，强调对递归自我改进可行性与治理挑战的观点不一。

ABSTRACT

Many leading AI researchers expect AI development to exceed the transformative impact of all previous technological revolutions. This belief is based on the idea that AI will be able to automate the process of AI research itself, leading to a positive feedback loop. In August and September of 2025, we interviewed 25 leading researchers from frontier AI labs and academia, including participants from Google DeepMind, OpenAI, Anthropic, Meta, UC Berkeley, Princeton, and Stanford to understand researcher perspectives on these scenarios. Though AI systems have not yet been able to recursively improve, 20 of the 25 researchers interviewed identified automating AI research as one of the most severe and urgent AI risks. Participants converged on predictions that AI agents will become more capable at coding, math and eventually AI development, gradually transitioning from `assistants' or `tools' to `autonomous AI developers,' after which point, predictions diverge. While researchers agreed upon the possibility of recursive improvement, they disagreed on basic questions of timelines or appropriate governance mechanisms. For example, an epistemic divide emerged between frontier lab researchers and academic researchers, the latter of which expressed more skepticism about explosive growth scenarios. Additionally, 17/25 participants expected AI systems with advanced coding or R&D capabilities to be increasingly reserved for internal use at AI companies or governments, unseen by the public. Participants were split as to whether setting regulatory ``red lines" was a good idea, though almost all favored transparency-based mitigations.

研究动机与目标

• 了解研究人员如何设想AI系统自动化AI研发（ASARA）及其中的里程碑。
• 评估对ASARA capable模型的公开部署与内部部署的看法与原因。
• 探讨感知的约束、发展路径以及能改变对自主AI研发信念的证据。
• 考察风险感知，包括权力集中、就业影响与安全考量。
• 评估对治理选项（如红线与基于透明度的缓解措施）的看法。

提出的方法

• 对2025年8–9月的前沿AI实验室研究人员与学者进行了25次半结构化访谈。
• 对访谈原文进行归纳编码以识别参与者之间的主题与模式。
• 采用AI辅助编码（Claude）对分类变量进行分类，并由人工复核。
• 通过参与者驱动的术语表（论文中的表1）界定ASARA、智能爆炸与部署等概念。
• 对参与者进行了匿名处理，且在 publication 之前由访谈对象审阅引用。

Figure 1: Participant predictions on whether AI companies will publicly deploy models capable of meaningfully accelerating AI research, grouped by affiliation type. The figure shows a categorical dot plot with participants grouped into “Expects Internal,” “Nuanced,” and “Expects Public” categories.

实验结果

研究问题

• RQ1 ASARA在AI系统对编程的辅助与最终开展自主研究任务时可能呈现何种形态？
• RQ2 研究人员对ASARA能力的内部部署与公开部署的模式有何预期，原因是什么？
• RQ3 有哪些约束、里程碑与可观察的证据表明自主AI研发在前进或倒退？
• RQ4 研究人员对风险（如权力集中、就业、治理）及可能缓解措施的担忧是什么？
• RQ5 研究人员如何看待红线与基于透明度的治理作为应对ASARA风险的方式？

主要发现

• 大多数研究人员预见从编码助手向自主AI开发者的渐进发展，潜在引发智能爆炸，尽管时间线观点存在差异。
• 约25位研究者中约有20人将自动化AI研究视为最严重的AI风险之一；对递归改进是否会发生以及何时发生存在分歧。
• 大约一半的参与者在讨论部署时预期内部（仅限实验室）部署，因经济与监管压力也存在公开部署的可能；也有一部分人认为结果将因组织与政策环境而异而呈现细微差异。
• 研究人员强调的风险包括权力集中、在封闭环境中加速进展、对人类相关性的潜在威胁；对缓解措施与红线的意见分歧，既有支持设定阈值的观点，也有倾向透明度导向方法的观点。
• 里程碑包括长期任务能力（40小时的研究任务）以及对自主研究的可观察证明，如编码突破、数学/算法发现中的存在性证明，以及可解释性方面的担忧。

Figure 2: Participant views on the clarity of the trajectory toward ASARA, grouped by affiliation. The figure shows participants distributed along a spectrum from “Clear Path” through “Major Obstacles” to “Unknown/Unknowns.” Frontier lab researchers predominantly cluster toward viewing the path as c