Algorithmic discrimination: Direct or indirect? A study on the inadequacy of this two-dimensional understanding

Abstract

European anti-discrimination law is structured based on the assessment of discrimination according to whether it is direct or indirect. There is a doctrinal debate about what typology can address better the discrimination caused by algorithmic systems that, considering this kind of systems’ particularities, such as unintelligibility, consideration of multiple protected attributes, or absorption of implicit biases, leans for indirect discrimination. However, we consider whether it would not be more interesting to change perspective given the difficulties in elucidating what type of discrimination we are facing and focus the appreciation of discrimination on assessing if the system achieves or not certain precision standards according to particular groups of population (including groups created by combined protected attributes) that may be affected by the system, and using precision results as a starting point to appreciate whether discrimination has occurred.