The paper “People are not coins: Morally distinct types of predictions necessitate different fairness constraints” will be presented at the leading conference in fair AI ACM FAccT ’22 taking place in Seoul on June 21-24 and published in its proceedings.
This work is co-authored with Corinna Hertweck, Christoph Heitz, and Michele Loi and already available on arXiv.
In the paper, the authors identify two types of practices in predictive algorithms: human-group-based practices, which are about a person and based on data from similar people, and human-individual-based practices, which are based only on data of the person under consideration. They contend that there is a morally salient distinction between the two practices, which is grounded on the moral principle of treating people as individuals and not as specimens of a group, as in the case of coins. Furthermore, the authors show that the group fairness metrics that are necessary conditions for the fairness of human-group-based practices are not necessary conditions for the fairness of human-individual-based practices.