监控 AI:面向有效人类监督的框架
计算机与社会
2026-05-19 v1 人工智能
人机交互
摘要
人工智能在高风险决策场景中的应用带来技术、安全与规范性挑战;这些问题只有通过人类监督才能得到缓解。然而,人类监督的概念缺乏统一的基础性理解:监督架构未得到清晰界定,相关角色不明确,实施步骤不透明。因此,研究者与实践者困难确定如何设计、实施和评估能够实现有效人类监督的系统。本文基于计算机科学、人机交互、心理学、哲学与法学的跨学科视角,提出实用的人类监督 AI 系统有效框架。核心贡献包括:(1) foundational 框架,包含工作定义、架构与有效人类监督 AI 系统的流程;(2)用于记录监督架构与流程的初始模板,适用于多个领域;(3)对 emerging 领域有效人类监督 AI 系统开放性研究挑战的综合分析。
引用
@article{arxiv.2605.16278,
title = {Keeping an Eye on AI: A Framework for Effective Human Oversight of AI Systems},
author = {Susanne Gaube and Markus Langer and Tim Miller and Kevin Baum and Raimund Dachselt and Anna Maria Feit and Ujwal Gadiraju and Harmanpreet Kaur and Mark T. Keane and Richard Landers and Johann Laux and Q. Vera Liao and Brian Lim and Linda Onnasch and Tim Schrills and Liz Sonenberg and Chenhao Tan and Nava Tintarev and Ziang Xiao and Hanwei Zhang},
journal= {arXiv preprint arXiv:2605.16278},
year = {2026}
}
备注
The conceptual analysis for this work was undertaken by the authors at Dagstuhl seminar 25272 'Challenges of Human Oversight: Achieving Human Control of AI-Based Systems' (https://www.dagstuhl.de/25272), held at Schloss Dagstuhl (June 29th-July 4th, 2025)