[3N5-OS-11b] OS-11 (2)

[3N5-OS-11b-04] Fairwashing: the risk of rationalization

〇Hiromi Arai^1,2, Urlich Aïvodji^2,3, Olivier Fortineau⁴, Sébastien Gambs³, Satoshi Hara⁵, Alain Tapp^6,7 (1. RIKEN, 2. JST PRESTO, 3. Université du Québec à Montréal, 4. ENSTA ParisTech, 5. Osaka University, 6. Université de Montréal, 7. MILA)

Keywords:Fairness in machine learning, Interpretability in machine learning

Black-box explanation is the problem of explaining how a machine learning model produces its outcomes. While current model explanation techniques provides interpretability, they can be used in a negative manner to perform fairwashing, which we define as promoting the perception that a machine learning model respects fairness while it might not be the case. We demonstrate systematic rationalizations taken by an unfair black-box model using the model explanation with a given fairness metric. Our solution, LaundryML, is based on a regularized rule list enumeration algorithm whose objective is to search for fair rule lists approximating an unfair black-box model. We empirically evaluate our method on black-box models trained on real-world datasets and show that one can obtain rule lists with high fidelity while being considerably less unfair at the same time.