Report on the workshop „AI in Insurance“

17./18. November 2022 — King’s College London (KCL),

supported by FU Berlin (CIC) and KCL

 

The workshop was dedicated to legal issues and challenges raised by the use of artificial intelligence (AI) in the insurance sector. The following section will first provide an overview of the presentations given by the participants as an introduction to the discussion (sub A). This is followed by a summary of the discussions and the overarching and further insights gained (sub B). Finally, an outlook is given on the prospects for future cooperation (sub C).

 

A. Overview of the presentations

17. November 2022

In the first lecture of the workshop, Prof. Roger Brownsword (KCL) dealt with the question of whether AI programs should be recognized as legal entities under private law. In his view, there are two different approaches to this question: the regulatory approach and the dogmatic approach. Prof. Brownsword came out in favor of the regulatory approach, as the unique nature of such systems requires a completely new approach.

Julian Westphal (FUB) then turned his attention to the contractual challenges of AI. Using a case study, he showed that most declarations of intent made by an AI can be attributed to its owner in accordance with general contract law principles. Only in exceptional cases attribution fails. In these exceptional cases, no contractual obligation can be constructed via the right of representation or the figure of the "blanket declaration"

Prof. Bertram Lomfeld (FUB) then shed light on the ethical challenges in dealing with AI. He addressed the so-called trolley dilemma in the context of self-driving cars. Prof. Lomfeld identified the lack of predictability of an AI system's decisions as one of the biggest problems. He presented a technical solution from one of his current research projects, which is intended to help to better understand the decisions of an AI. In contrast to other technical solutions, the precision and performance of the AI should not be impaired.

After a short break, Prof. Christian Armbrüster (FUB) focussed on the use of AI in personal insurance. He presented an overview of various areas of application in the insurance sector - in private health insurance, for example, AI systems would help to recognise disease patterns as part of the diagnosis. However, the use of data by AI cannot be unlimited: The German Genetic Diagnostics Act prohibits the use of genetic data when deciding whether to take out life insurance. However, a comparable regulation from other legal systems was unknown to those present.

Victor Claussen (FUB) then gave a presentation on the dangers of discrimination through the use of AI, focussing on the algorithmic calculation of insurance premiums. It is particularly problematic when the processing of risk-related characteristics takes place in the "black box" of AI, for example when using "sub-symbolic" AI. The AI Act proposed by the European Commission does not offer a solution to this either.

The final lecture of the first day was given by Dr. Kyriaki Noussia (Reading University), who addressed the topic of self-driving cars. She focussed on ethical issues in data processing and liability law. She spontaneously weakened one of her core theses, as the first lecture of the day by Prof Brownsword had "opened her eyes".

 

18. November 2022

On the second day of the workshop, Prof Baris Soyer (Swansea University) opened with a presentation on the possibilities of the "internet of things" (IoT), particularly in industrial insurance. He presented a number of examples and spoke about the legal and regulatory difficulties involved in implementing these technologies. Unlike Prof Brownsword, he spoke out against strong regulation of AI.

Markus Hoffmann (FUB) followed up with a presentation on telematics tariffs. Such tariffs are a concrete use case of IoT. He discussed the advantages and disadvantages as well as ethical issues that arise in the continuous monitoring of certain risk-relevant circumstances.

Dr. Lukas Böffel (FUB) then turned his attention to the regulatory challenges of AI. He described current use cases and the latest initiatives of the EU Commission and the supervisory authority EIOPA. He presented EIOPA's position on the AI Act, according to which AI systems in the insurance sector should be exempt from regulation. As a result, he spoke out against a regulatory split in favour of industry-specific players.

Prof. Satoshi Nakaide (Waseda University) took an excursion into Japanese earthquake insurance. He explained that tariffs have recently been offered in Japan that link the insurer's cash benefit to a specific index - such as the strength of an earthquake on the Richter scale ("index insurance") - regardless of the actual occurrence of damage. Using an app, the insurer's benefit is paid out without a specific assessment of any damage (e.g. in the form of Amazon vouchers). The German participants in the workshop noticed interesting differences to the basic principles of German insurance law.

Back to the topic of the seminar led Prof. Dan Hunter's (Dean, KCL) lecture on machine learning and volitional behaviour. Legal consequences are often linked to whether behaviour was intentional or unintentional. Prof. Hunter argued that decision-making in AI cannot be categorised as knowing/unknowing. He argued in favour of a regulatory solution.

In the final presentation of the workshop, Prof. Özlem Gürses (KCL) dealt with fraud in the context of machine learning. An AI can draw conclusions about a possible fraud attempt from various pieces of information in a property insurance claim over the phone (e.g. tone of voice, speaking speed, wording). It was remarkable for the participants how (supposedly) little information an AI needs to make a decision; Prof. Gürses cited the example of a private health insurance company that allows a single selfie to suffice as pre-contractual health information for the policyholder.

B. Summary of the discussions

The wide range of topics discussed during the workshop made it possible to draw parallels between different issues. One recurring point of discussion, for example, was whether and to what extent an (additional) regulatory framework is required for AI. Opinions in the discussions were sometimes divided on this point. As a result, the majority of participants were in favour of a regulatory framework in view of the new challenges associated with AI. However, the general consensus was that over-regulation should be avoided in order not to stand in the way of innovation and the obvious benefits of AI.

The second focus of the seminar was on ethical issues, which were addressed in various presentations and numerous discussions. It became clear that the problems in this area are complex and multi-dimensional. They range from data protection and privacy to discrimination and a lack of transparency in decision-making.

In the discussion on telematics tariffs in particular, it became clear that the loss of privacy associated with monitoring was not categorised as problematic by most of those present. One reason for this was that customers are free to choose such a tariff and therefore disclose the information themselves. In addition, the risk (and therefore also the premium) can be better determined, which represents a significant advantage over conventional tariffs. According to the workshop participants, the advantages also outweigh the disadvantages when using AI to detect insurance fraud.

Those present took a more critical view of the processing of very private data (such as findings from genetic analyses) and discriminatory decisions. The use of AI must not lead to policyholders being discriminated against or (indirect) pressure being exerted on them to disclose highly personal data.

C. Prospects for future collaboration

The workshop was very well received by all participants. The conclusion was that the presentations and discussions on an internationally relevant and topical subject area provided new insights and insights into different points of view. The face-to-face meeting was therefore also used to think together about future prospects for cooperation. It emerged that the challenges of climate change for the insurance sector are seen as a current and exciting topic area that can also be linked to the findings from the workshop as far as the use of AI is concerned. It is therefore intended to hold a follow-up workshop at the FUB in 2023.