Non-empirical problems in fair machine learning

Abstract The problem of fair machine learning has drawn much attention over the last few years and bulk offered solutions are, in principle, empirical. However, algorithmic fairness also raises important conceptual issues that would fail to be addressed if one relies entirely on empirical considerations. Herein, I will argue current debate developed an framework brought contributions development decision-making, such as new techniques discover prevent discrimination, additional assessment criteria, analyses interaction between predictive accuracy. same suggested higher-order regarding translation into metrics quantifiable trade-offs. Although (empirical) tools which have been so far are essential address discrimination encoded data algorithms, their integration society elicits key (conceptual) questions as: What kind assumptions decisions underlies framework? How do results approach penetrate public debate? reflection deliberation should stakeholders available metrics? outline learning, i.e. how is framed addressed, suggest there non-empirical tackled. While this work focus fairness, lesson can extend other problems analysis decision-making privacy explainability.