As artificial intelligence becomes more deeply integrated into decision-making across healthcare, finance, and employment, concerns have grown over biased outcomes stemming from the data on which these systems are trained. Reported cases—such as preferential lending to men and higher parole approval rates for White individuals—highlight how AI can replicate or amplify societal biases. Accordingly, research efforts increasingly focus on developing training methods that mitigate such bias and promote socially acceptable, equitable decision-making.

This presentation surveys prominent examples of AI bias and outlines key methodological approaches for mitigation, including various formal definitions of fairness and techniques for training models that satisfy them. It also introduces a new clustering algorithm specifically designed to improve fairness within clustering analyses.

🔗 2025 SNU AIHC Seminar(Oct.) Summary