Insights and Highlights: Unpacking “Fairness” in Insurance

Panelists included Lisa A. Schilling, FSA, EA, FCA, MAAA, Director of Practice Research, Society of Actuaries Research Institute and Peggy Tsai, Chief Data Officer, BigID. The session underscored the challenges posed by AI, emphasizing the importance of strong governance, transparency, and ongoing process enhancements to maintain fairness in data practices and ensure equitable outcomes in insurance.

Fairness in insurance products and processes has been a long-time hallmark of good management for successful insurance companies. Regulations require companies not be unfairly discriminatory to consumers in their processes and practices. This issue has come to the forefront in the industry recently amid advances in artificial intelligence (AI). Panelists underscored that AI and advanced analytics have heightened both the positive potential and negative implications of existing insurance practices. The discussion emphasized the need for a nuanced approach to fairness that addresses the complexities introduced by these technologies.    

A pivotal theme was the significance of data quality and governance in ensuring fairness. Highlighting the inherent biases that can emerge during data collection, panelists stressed the ongoing recalibration and transparency necessary in model outputs to mitigate these biases effectively. Robust stewardship practices should prioritize data integrity before model building and decision-making. Ensuring accurate risk classification aligned with expected claims values can serve as a fundamental aspect of actuarial fairness.

The panel then examined the challenges posed by data proxies and synthetic data in insurance models. Synthetic data is data that is produced by machines, sometimes to represent human behaviors. Data proxies similarly involve analysis informed by machines processes to represent real-world behavior. Concerns were raised about the accuracy and representativeness of these proxies, particularly in reflecting real-world demographics. The difficulty of removing synthetic data once integrated into models underscored the importance of rigorous validation and transparency throughout the modeling process, including at the beginning of a development process. A critical aspect of the discussion addressed the use of proxies for race and ethnicity in insurance, highlighting the ethical and regulatory implications. Panelists stressed the necessity of rigorous data management and model validation processes to ensure compliance and fairness in risk assessment practices.

The discussion concluded with a consensus on the imperative for continuous monitoring, recalibration, and transparent communication in insurance practices. Balancing data-driven decision-making with fairness and objectivity remains a paramount challenge, requiring ongoing efforts to align technological advancements with ethical standards.

To learn more about AI in financial services, you can explore further with research from the Center for Ethics in Financial Services.