On the Fairness of Causal Algorithmic Recourse

Algorithmic fairness is typically studied from the perspective of predictions. Instead, here we investigate recourse actions suggested to individuals remedy an unfavourable classification. We propose two new fair-ness criteria at group and individual level, which—unlike prior work on equalising average group-wise distance decision boundary—explicitly account for causal relationships between features, thereby capturing downstream effects performed in physical world. explore how our relate others, such as counterfactual fairness, show that complementary prediction. study theoretically empirically enforce fair by altering classifier perform a case Adult dataset. Finally, discuss whether violations data generating process revealed may be better addressed societal interventions opposed constraints classifier.