Fairness as adequacy: a sociotechnical view on model evaluation in machine learning

Abstract

AbstractThis paper develops an account of model evaluation—with an emphasis on fairness concerns—that takes the social situatedness of ML models as its starting point. Such a view entails that ML models are not deemed isolated entities, but rather tools, used for specific purposes and potentially impacting their social environment in manifold ways. This shift of perspective opens up a new problem space and facilitates rethinking criteria for model evaluation. By drawing on the adequacy-for-purpose view in philosophy of science, epistemic norms and desiderata for an adequate deployment of ML models along the dimensions of Social Objectives, Measurement, Social Dynamics, and interaction are then identified. The account thus developed also highlights why any auditing of ML models that ought to assist in consequential decision-making cannot be limited to an assessment of statistical properties, but needs to incorporate a variety of methods from the social sciences instead. Moreover, while the process of model evaluation might be deemed as a mere technical exercise, it is in fact riddled by epistemic and morally normative considerations.