The widespread integration of Artificial Intelligence (AI) driven algorithms into various aspects of modern society raises fundamental questions about their impact on decision-making processes. This talk addresses a crucial inquiry: does the integration of AI improve human decision-making compared to human or AI alone?

I will introduce a causal inference framework to evaluate three distinct decision-making paradigms: human-alone, human-with-AI, and AI-alone systems. Utilizing standard classification metrics based on potential outcomes, the evaluation will consider both the prediction accuracy and the fairness of these systems. Furthermore, I will introduce strategies to improve decision-making by combining different systems and optimizing AI algorithms.