Original
A Public Values Framework for AI Governance: Evidence From US Federal AI Policymaking
Abstract
ABSTRACT Governments at all levels are grappling with AI's implications for public administration. Because federal actions often establish the principles guiding wider governance efforts, this study asks how US federal AI policies reflect—or fail to reflect—core public values. Drawing on a dataset of US federal AI policies from 2020 to 2023, we systematically analyze the risks, harms, and governance strategies employed across government‐issued and enacted policy documents. Specifically, we analyze 120 discrete policy sections or subsections of federal laws and regulations, utilizing the parsing approach of the AI Governance and Regulatory Archive (AGORA), all of which regulate the operation of the government itself rather than private sector behavior. We develop a theoretical and empirical crosswalk linking AI‐related risks and harms to prominent public values. Results indicate that while values such as individual rights, the common good, and accountability are regularly invoked, additional guidance around public values like citizen engagement, workplace standards, and responsiveness may be necessary to prevent public value failure. The study further categorizes governance strategies in AI policy documents into regulatory, service‐based, and informational instruments, and finds that regulatory policies are most likely to engage with public values.
中文
人工智能治理的公共价值框架：来自美国联邦人工智能政策制定的证据
摘要
摘要：各级政府都在应对人工智能对公共行政的影响。由于联邦行动往往确立了指导更广泛治理努力的原则，本研究追问美国联邦人工智能政策如何反映——或未能反映——核心公共价值。基于2020年至2023年美国联邦人工智能政策数据集，我们系统分析了政府发布和颁行的政策文件中所采用的风险、危害与治理策略。具体而言，我们分析了联邦法律和法规中120个相互独立的政策条款或子条款，采用人工智能治理与监管档案（AGORA）的解析方法，这些条款均规范政府自身的运作，而非私营部门行为。我们构建了一个将人工智能相关风险与危害同重要公共价值相连接的理论与实证对照表。结果表明，尽管个人权利、共同善和问责等价值经常被援引，但为防止公共价值失败，可能还需要围绕公民参与、工作场所标准和回应性等公共价值提供更多指引。本研究进一步将人工智能政策文件中的治理策略划分为监管型、服务型和信息型工具，并发现监管型政策最有可能涉及公共价值。
关键词
人工智能治理、公共价值、美国联邦政策、政策工具、风险与危害、公共行政、问责、公民参与