Explainable AI (XAI) systems are a necessity of our present era. Stakeholders must be able to trust AI in order to actively use it. In 2021 EIOPA's Consultative Expert Group on Digital Ethics in Insurance published six principles for the ethical use of trustworthy AI in the insurance sector. These principles include fairness & non-discrimination, transparency & explainability, data governance & record keeping and robustness & performance. We will focus on the principle of transparency & explainability. For this purpose, a use case of ratings of German life insurance companies will be considered.

Two different types of acceptance towards XAI have to be taken into account. On the one hand, technical acceptance and, on the other hand, user acceptance in the ratings of life insurance companies. Business figures of life insurance companies are often regarded as intransparent. This holds even true when national GAAP accounting (e. g. HGB) is supplemented by the Solvency II figures in the Solvency and Financial Condition Reports (SFCR).

Transparency is a basic prerequisite for a high level of explainability of the generated results by the XAI. There are, for example, mathematical sensitivity analyses (ratio of input data to result data, i.e. ex post explainability) or algorithms that already provide explainability by design (e. g. decision trees). They are used in our use case of the company rating. Especially, relatively small, structural networks are used which are given in equation form and represent expert knowledge. This guarantees explainability on the basis of a directed graph. Users therefore need less professional and technical expertise to understand. In our use case this means making the strengths and weaknesses of the insurer visible in a graph. Expert knowledge and AI methods are combined and used in such a "hybrid" model for the visualisation of the results, which operationalises the explainability also for the users of the company ratings. In this respect, the discussion of explainable AI methods in this context is a useful contribution to data ethics in insurance.