AI 中文总结
本研究通过四个AI部署案例，提炼出意图性、可解释性、道德作者身份和公正性四项治理原则，为高风险组织中的AI委托决策提供问责框架。
AI 中文摘要
人工智能（AI）在高风险决策中的快速整合已超越了人类监督和问责的既有机制，导致组织在负责任地委托决策权方面缺乏指导。本研究考察了四个被广泛记录的AI部署案例：新冠疫情期间实施的英国A-Level评分算法、亚马逊的自动化招聘系统、美国刑事司法系统中使用的COMPAS再犯风险评估工具，以及荷兰的SyRI福利欺诈检测系统。我们利用61个公开来源（包括政府报告、组织文件和媒体报道），基于两阶段扎根理论编码方法进行了比较定性分析。分析生成了一个包含32个条目的编码手册，随后将其应用于110个编码片段，并使用定量分析来评估各案例间编码的一致性。研究结果中浮现出四项反复出现的治理原则：（1）意图性，即领导者有意识地决定何时应使用AI；（2）可解释性，要求决策过程足够透明，以便进行解释和审查；（3）道德作者身份，即可识别的人类行动者对委托决策保留责任；（4）公正性，要求委托安排尽量减少对现有不平等的强化。这些发现为审视高风险组织环境中的领导问责和AI治理提供了一个基于经验的框架。
英文摘要
The rapid integration of artificial intelligence (AI) into high-stakes decision-making has outpaced established mechanisms for human oversight and accountability, leaving organisations with limited guidance on the responsible delegation of decision authority. This study examines four widely documented AI deployments: the UK A-Level grading algorithm implemented during the COVID-19 pandemic, Amazon's automated hiring system, the COMPAS recidivism risk assessment tool used in the U.S. criminal justice system, and the Dutch SyRI welfare-fraud detection system. Using 61 publicly available sources, including government reports, organisational documents, and media accounts, we conducted a comparative qualitative analysis based on a two-phase grounded-theory coding approach. The analysis produced a 32-item codebook, which was subsequently applied across 110 coded segments, with quantitative analyses used to assess coding consistency across cases. Four recurring governance principles emerged from the findings: (1) Intentionality, whereby leaders deliberately determine when AI should be used; (2) Interpretability, requiring decision processes to be sufficiently transparent to enable explanation and scrutiny; (3) Moral Authorship, whereby identifiable human actors retain responsibility for delegated decisions; and (4) Justice, requiring delegation arrangements that minimise the reinforcement of existing inequities. These findings contribute an empirically derived framework for examining leadership accountability and AI governance in high-stakes organisational settings.
Journal refInternational Journal of Organizational Leadership 14(2025) 637-655
