5 Case Studies: Challenges for insurers While the Discrimination Acts apply to all decisions made by insurers, AI-informed decision making may raise novel challenges and uncertainty around what constitutes unlawful discrimination. This section uses case studies to explore challenges faced by insurers in avoiding discrimination and complying with the Discrimination Acts when using AI-informed decision making. These hypothetical case studies are provided as simple examples, to aid understanding and illustrate some of the issues discussed above. While they are not exhaustive, they aim to illustrate a range of challenges in different areas of insurance, and to outline some general principles that may be applied to more complex situations and other circumstances. The case studies are not based on any actual experiences of insurers or any other individual experiences known to the Commission or the Institute. 5.1 Case Study: Car Insurance This case study explores the challenges that arise when data is correlated with a protected attribute, which may be unseen, and identifying where further action may be required to avoid unlawful discrimination. (a) Part A Car Insurance Pty Ltd (CIPL) is considering how to price its car insurance policies. For simplicity, the following assumptions apply: • There are only two types of cars: expensive cars and cheaper cars. • The cost of repairing an expensive car is significantly more than the cost of repairing a cheaper car. • All cars have the same probability of crashing during a policy period. As such, the expected cost of claims (per car) is significantly more for expensive cars compared to cheaper cars. CIPL is considering charging higher premiums for the drivers of expensive cars in accordance with the difference in the expected cost of claims. CIPL is concerned that there may be an unknown relationship between a protected attribute and the cost of the car people drive. It would be very difficult for CIPL to determine any such relationship on the available data. This unknown relationship might mean that people with certain protected attribute statuses (Group A) are more likely to drive the expensive car, whereas people without that protected attribute (Group B) are more likely to drive the cheaper car. If such a relationship exists, CIPL is concerned that charging higher premiums for the expensive cars may constitute discrimination against Group A. However, it is unlikely that CIPL will have engaged in unlawful discrimination by charging premiums as proposed. Is there direct discrimination? There is no direct discrimination as Group A are not being treated differently from Group B because of, or by reason of, a protected attribute. The premium charged is due to the cost of repairing the car, and the relationship between the cost of the car and the protected attribute is unknown to CIPL. Guidance Resource: Artificial intelligence and discrimination in insurance pricing and underwriting • 2022 • 27