Towards Accountability: The Articulation and Formalization of Fairness in Machine Learning

Abstract

The working paper has a twofold aim. First, it will consider how theories of distributive justice have informed, and may inform future, developments and research within the field of machine learning. Under the first section, the paper will analyze current fair machine learning approaches. It will be considered whether these approaches are sufficiently adequate in dealing with the ethical concerns that are generally associated with machine learning, and the contextual challenges brought forth through the deployment of these techniques. After this primary assessment, the paper will discuss how theories of distributive justice might enable a better understanding of fairness and how notions of fairness could help contribute towards future fair machine learning efforts. Second, the paper will establish how the articulation and formalization of fairness, based upon notions found within theories of justice, could (indirectly) increase the accountability of actors who make use of machine learning systems.