Trustworthy artificial intelligence in insurance: Navigating fairness and performance in predictive modeling
Presented by Milliman
Predictive models in insurance are often built on data influenced by human decisions, structural inequities, and representation bias. Simply removing sensitive attributes does not eliminate bias. Milliman’s paper outlines the legal and business imperatives for fairness in actuarial practices and introduces core concepts of group fairness. It also presents practical bias-mitigation methodologies aimed at reducing legal exposure while maintaining predictive performance.