Artificial intelligence, bias and insurance – A technical and legal analysis

Abstract

Artificial intelligence (AI) can bring significant efficiencies to the data-driven insurance industry. This is also true for the use of AI in relation to potential customers, policyholders, and beneficiaries. However, AI carries the risk of bias, a systematic forecasting error, and thus of discrimination. This interdisciplinary article examines the risk of bias and its causes from the perspective of computer science. Based on this, it discusses from a legal point of view the extent to which non-discrimination law applies in the case of bias and what the legal consequences are. The subsequent analysis of whether and how data scientists can technically eliminate bias leads to the legal question of how insurance supervisory law ensures that bias is avoided as far as possible. Finally, we examine the implications of the EU draft of an AI regulation for the avoidance of bias. How to cope with the assumed inevitability of bias is one of the important questions that future interdisciplinary research of information technology and insurance law has to solve.