Temporal-Logic-Based Causal Fairness Analysis

Recently, there has been growing reliance on AIbased decision-making algorithms and frameworks in different areas such as health-related policies and credit scoring. The AI-based algorithms and frameworks may have been trained using data that is biased against a specific subpopulation or group, e.g., a certain race or gender. Hence, it is crucial to incorporate fairness measures into the decision-making process. However, most of the existing fairness analysis methods do not consider the inherent temporal dependencies in real-world situations. In this paper, we propose Temporal-Logic-Based Causal Fairness Analysis (TL-CFA) as a novel framework that integrates causal reasoning and temporal logic to address fairness concerns in algorithmic decision-making processes. This framework contributes to fairness-aware machine learning by offering a comprehensive approach that considers the temporal dimension, thus promoting fair and unbiased decision-making. We demonstrate the usefulness of the proposed method in health-inequity and gender pay gap case studies. We then compare the results obtained by TL-CFA by Path Analysis and Mediation Analysis.