[论文] Value-Sensitive Delegation in Everyday AI Agent Use: Evidence from Ope...

论文概要

研究领域: ML 作者: Renkai Ma, Ruyuan Wan, Xuan Lu, Fan Yang, Chen Chen, Lingyao Li 发布时间: 2026-09-18 arXiv: 2609.22067

中文摘要

用户越来越多地将工作委托给自主 AI 智能体，然而评估通常衡量任务完成度，而非用户优先看重的价值。我们采用价值敏感设计（Value Sensitive Design），在 LLM 辅助下分析了 73,093 篇关于使用 OpenClaw 的第一人称 Reddit 帖子，每篇都标注其人类价值、智能体方面、价值满足度和用户结果。21 项价值归为六个价值组：自主、可靠、可负担的运行，有界触及范围、可审查性和公平可及性。相对于各维度在语料中的占比，价值并非聚集在智能体的输出上，而是聚集在用户为一次运行所设定的操作条件上。在用户描述智能体交付成果的帖子中，价值通常得到满足（六组中五组如此）；而在用户描述监督智能体的帖子中，价值大多未得到满足（六组全部如此）。我们将这一模式概念化为“价值敏感委托”。要支撑人类价值，不能只关注智能体完成了什么，还要关注用户围绕委托设定的条件，包括成本、可及性和监督。

原文摘要

Users increasingly delegate work to autonomous AI agents, yet evaluations typically measure task completion rather than the values users prioritize. Using Value Sensitive Design, we analyzed, with LLM assistance, 73,093 first-person Reddit posts about using OpenClaw, each for its human value, agent aspect, value fulfillment, and user outcome. The 21 values form six value groups, including Autonomous, Dependable, and Affordable Operation, Bounded Reach, Reviewability, and Equitable Access. Relative to each aspect's corpus share, values clustered not at the agent's outputs but at the operating conditions users set around a run. Values were usually met where users described what the agent delivered, in five of six groups, and mostly unmet where users described supervising it, in all six group...

#论文 #arXiv #ML #小凯