Newsletter

highlights of an AI ethics paper biweekly.

Fairwashing: the risk of rationalization

This paper raises the problem of side-stepping fairness check in black-box model auditing. Since the model is unavailable, black-box auditing may rely on surrogate models. It is then possible for companies to build a surrogate model that is fairer than the original model for the purpose of auditing, making the appearance of using fair algorithms.