Publication Details

Download

Rifat Amin
Multi-Objective Counterfactuals for Counterfactual Fairness in User Centered AI
UCAI'23: Workshop on User-Centered Artificial Intelligence @ Mensch und Computer 2023 (MuC'23), September 3-6, Rapperswil, Switzerland

This position paper emphasizes the role of user-centered artificial intelligence in critical decision-making domains in machine learning models. In this paper, I introduce MOCCF (Multi-Objective Counterfactuals for Counterfactual Fairness) as an extended method that generates realistic counterfactuals by leveraging multiple objectives. Furthermore, to increase transparency, I propose two fairness metrics, Absolute Mean Prediction Difference (AMPD), and Model Biasness Estimation (MBE). I argue that these metrics enable the detection and quantification of unfairness in binary classification models both at the individual and holistic levels consecutively and contribute to user-centered artificial intelligence.