Application of Artificial Intelligence in the Insurance Industry

Abstract

This report first outlines the current state of the application of Artificial Intelligence (AI) in the insurance industry and addresses the present role of actuaries in its use. In line with the general public discussion and terminology, we already use the term Artificial Intelligence, although the insurance industry currently employs no applications that meet the strict definition of AI. Rather, Machine Learning (ML) and Statistical Learning (SL) methods are in use. In the following, we summarize AI, ML, and SL systems under the term “AI systems.”

The Ethics Guidelines for Trustworthy AI [1] by the EU’s High-Level Expert Group on Artificial Intelligence (HLEG) define ethical principles and concrete requirements for trustworthy AI systems. For the insurance sector as a whole, as well as for each individual insurance company, the question arises of how the trustworthiness of deployed AI systems can be tested, ensured, and demonstrated—or more generally, whether specific regulatory needs exist in this regard.

Due to the already extensive regulation of the insurance sector—such as Solvency II, the German General Equal Treatment Act (AGG), and the EU General Data Protection Regulation (GDPR) [2]—the ethical principles are largely covered by the existing regulatory framework. Taking into account these regulations, the DAV’s Professional Code of Conduct [3], and the report Handling Data in the Field of Data Science [4], this document analyzes the topic from the actuaries’ perspective.

The third section discusses the current state of algorithmic explainability and interpretability. With the increasing use of machine learning and artificial intelligence algorithms, there is a growing demand to “make algorithms explainable.” This aims to ensure that algorithm-based decisions are comprehensible and not unjust or flawed. For common methods, interpretability approaches are described and illustrated with concise examples that show interpretability is indeed quite feasible.

Overall, the DAV sees no need for additional regulation specifically targeted at the use of AI systems. However, for certain specific aspects of the requirements for trustworthy AI, we do see the need for more detailed analysis and, if necessary, recommendations for action—particularly for actuaries.

This report serves as a basis for discussion and is intended to inform members and committees of the DAV about the findings. It does not represent a professionally endorsed position of the DAV.