Management Science · 2026 · Alex Chohlas-Wood、Madison Coots、Henry Zhu、Emma Brunskill、Sharad Goel
中文摘要
为使算法公平，机器学习文献大多聚焦于在不同种族或性别群体之间拉平决策、结果或错误率。举例来说，设想一个假想的政府网约车项目，为即将面临出庭日期的低收入人群提供交通协助。按照这一文献的做法，人们可能会把乘车机会分配给每美元估计处理效应最高的人，同时约束支出在各种族群体之间相等。然而，这种做法忽视了此类约束的下游后果，因而可能造成意想不到的伤害。例如，若某个人口群体居住地距离法院更远，强制支出相等必然意味着提供的乘车总次数更少，并可能使更多人因未出庭而受罚。在此，我们提出一种以决策后果为核心的替代性公平算法设计框架。在我们的方法中，首先需要引出利益相关者在可能的决策空间及相应结果上的偏好——例如在支出均等与出庭率之间进行权衡的偏好。随后，我们在决策策略空间上进行优化，以最大化所引出效用的方式做出权衡。为此，我们开发了一种算法，能够从数据中高效学习这些最优策略，并适用于一大类表达力强的效用函数。具体而言，我们使用情境多臂老虎机(contextual bandit)算法探索策略空间，同时在每一步求解一个凸优化问题，以基于现有信息估计最优策略。这一后果主义范式有助于以更全面的方式实现公平决策。本文经 Catherine Tucker 接收，收录于“人—算法关联”专刊。资助：本研究得到斯坦福影响实验室 [Start-Up Impact Labs/GEJBU]、斯坦福以人为本人工智能研究院以及哈佛数据科学倡议的支持。补充材料：在线附录与数据文件可在 https://doi.org/10.1287/mnsc.2022.00345 获取。
Abstract
In an attempt to make algorithms fair, the machine learning literature has largely focused on equalizing decisions, outcomes, or error rates across race or gender groups. To illustrate, consider a hypothetical government rideshare program that provides transportation assistance to low-income people with upcoming court dates. Following this literature, one might allocate rides to those with the highest estimated treatment effect per dollar while constraining spending to be equal across race groups. That approach, however, ignores the downstream consequences of such constraints and, as a result, can induce unexpected harm. For instance, if one demographic group lives farther from court, enforcing equal spending would necessarily mean fewer total rides provided and potentially more people penalized for missing court. Here we present an alternative framework for designing equitable algorithms that foregrounds the consequences of decisions. In our approach, one first elicits stakeholder preferences over the space of possible decisions and the resulting outcomes—such as preferences for balancing spending parity against court appearance rates. We then optimize over the space of decision policies, making trade-offs in a way that maximizes the elicited utility. To do so, we develop an algorithm for efficiently learning these optimal policies from data for a large family of expressive utility functions. In particular, we use a contextual bandit algorithm to explore the space of policies while solving a convex optimization problem at each step to estimate the best policy based on the available information. This consequentialist paradigm facilitates a more holistic approach to equitable decision making. This paper was accepted by Catherine Tucker, Special Issue on the Human-Algorithm Connection. Funding: This work was supported by Stanford Impact Labs [Start-Up Impact Labs/GEJBU], the Stanford Institute for Human-Centered Artificial Intelligence, and Harvard Data Science Initiative. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.00345 .