标题： 公平性在政策制定中使用机器学习方法估计异质处理效应的影响 显示英文标题

标题： Fairness Implications of Heterogeneous Treatment Effect Estimation with Machine Learning Methods in Policy-making

摘要： 因果机器学习方法能够灵活地生成异质性治疗效果估计，对于试图制定和实施政策的政府来说可能是非常有用的工具。然而，正如关键的人工智能文献所示，政府在使用机器学习模型时必须非常小心，以免产生意想不到的后果。 为了尽量避免出现意外的不良结果，可以采用人工智能公平性方法，这些方法旨在创建不会受种族或性别等敏感变量影响的机器学习模型。 在本文中，我们论证了标准的人工智能公平性方法，这些方法是为预测性机器学习开发的，不一定适用于所有的因果机器学习应用，因为因果机器学习通常（至少到目前为止）是通过建模来告知最终决策者的，而人工智能公平性方法假设模型直接做出决策。 我们将这两种情况分别定义为间接决策和直接决策，并建议政策制定最好被视为一种联合决策，在这种决策中，因果机器学习模型通常只有间接影响力。 我们为此情景定义了公平性——即提供决策者所需的信息，使其能够准确判断公正的政策结果——并认为因果机器学习模型的复杂性使得实现这一点变得困难。 这里的解决方案不是传统的调整人工智能公平性，而是仔细建模以及意识到这些方法可能鼓励的一些决策偏见，这正是我们要描述的内容。 显示英文摘要

摘要： Causal machine learning methods which flexibly generate heterogeneous treatment effect estimates could be very useful tools for governments trying to make and implement policy. However, as the critical artificial intelligence literature has shown, governments must be very careful of unintended consequences when using machine learning models. One way to try and protect against unintended bad outcomes is with AI Fairness methods which seek to create machine learning models where sensitive variables like race or gender do not influence outcomes. In this paper we argue that standard AI Fairness approaches developed for predictive machine learning are not suitable for all causal machine learning applications because causal machine learning generally (at least so far) uses modelling to inform a human who is the ultimate decision-maker while AI Fairness approaches assume a model that is making decisions directly. We define these scenarios as indirect and direct decision-making respectively and suggest that policy-making is best seen as a joint decision where the causal machine learning model usually only has indirect power. We lay out a definition of fairness for this scenario - a model that provides the information a decision-maker needs to accurately make a value judgement about just policy outcomes - and argue that the complexity of causal machine learning models can make this difficult to achieve. The solution here is not traditional AI Fairness adjustments, but careful modelling and awareness of some of the decision-making biases that these methods might encourage which we describe.