DISA SEMINAR

Addressing Fairness in Machine Learning Predictions: Strategic Best-Response Fair Discriminant Removed Algorithm

Speaker
Dr. Warut Khern-am-nuai, Assistant Professor, Desautels Faculty of Management, McGill University

18 Mar 2022 Friday, 10:30 AM to 12:00 PM

Via Zoom

Abstract:
Discrimination in machine learning (ML) has become prominent as it is increasingly used for decision-making. Although many "fair-ML" algorithms are designed to address such discrimination issues, virtually all of them focus on alleviating disparity in the prediction results by imposing additional constraints. Naturally, in response, prediction subjects alter their behaviors. However, the algorithms never consider those behavioral responses. So, even if the disparity in prediction results may be removed, the disparity in behaviors may persist across different subpopulations of prediction subjects. When these biased behavioral outcomes are used for training ML, the ML can perpetuate the discrimination in the long run. In order to study this issue, we define a new notion called "strategic best-response fair" (SBR-fair). It is defined in a context involving subpopulations that are ex-ante identical and also have identical conditional payoffs. Even if an algorithm is trained on biased data, will it lead to identical equilibrium behaviors of subpopulations? If yes, we define the ML as SBR-fair. We demonstrate that many fair-ML algorithms in the literature are not SBR-fair. Even the one existing SBR-fair algorithm has practical limitations. Based on our findings, we propose an alternative class of ML algorithms that is SBR-fair and free from the aforementioned limitations. We analytically show the theoretical properties of the proposed algorithm and demonstrate its performance in practice vis-a-vis other ordinary off-the-shelf algorithms and existing fair algorithms using both a synthetic dataset and several real-world datasets.

Bio:
Warut Khern-am-nuai is an assistant professor in Information Systems at Desautels Faculty of Management, McGill University. His research interests include platform for online marketplaces, predictive analytics, and management information security. He received his Ph.D. in Management Information Systems from Krannert School of Management, Purdue University in 2016. His research has appeared in Management Science, Information Systems Research, and MIS Quarterly, among other.