Accountability in Government Use of AI : Citizen Concerns and Preferences
通过混合方法研究中国公民对政府使用AI的问责关切与偏好，发现公民担忧AI能力不足、缺乏人类监督、算法不透明及问责执行不力，并偏好有问责条款、生态系统透明度、内部审计和行政申诉的系统。
ABSTRACT Government adoption of AI opens the “pandora's box” of accountability , a core aspect of public value. Despite the centrality of governmental AI accountability, citizens' concerns and preferences in holding governments accountable for AI usage remain poorly understood. To address these puzzles and advance knowledge on AI accountability in non‐Western democracies, we conducted a mixed‐method study using data collected from Chinese citizens. A story‐completion study ( N = 50) suggested concerns over AI's incompetence, inadequate human oversight, algorithmic opacity, and weak accountability enforcement leading to harmful outcomes. A discrete‐choice experiment ( N = 2080) that measured preferences toward four algorithmic accountability— procurement clauses , transparency , auditing , and appeal s—supported all hypotheses. Results showed a preference toward systems with accountability clauses over those without, ecosystem‐level over technical transparency, and internal over external auditing. Reactive redress using administrative appeals for AI errors was valued. This study advances knowledge on algorithmic accountability and insights to enhance public trust in government AI use.