Promotion of trustworthy AI in social insurance
ed* No. 02/2024 – Chapter 6
While transparency, trust and acceptance are the basis for greater use of AI systems overall, the social insurance sector, in particular, faces the difficult task of wanting to benefit from the possibilities of AI systems for the sake of the insured on the one hand. On the other, ethical issues must not be lost sight of and decisions that directly affect people must not be handed over to potentially prejudiced AI systems. This challenge can be met by taking certain precautions.
For example, to prevent people in decision-making positions from developing excessive trust in automated tools, outputs from AI systems could be alternated with random outputs to ensure that decision-makers maintain a critical mindset and do not adopt AI decisions without reflection. Another idea are AI assistants that are developed in such a way that they ask questions based on an extensive database instead of giving answers. In this way, employees could be assisted in making appropriate decisions themselves.
These two examples show that humans can be supported by AI systems on their path to decision-making. Ultimately, they do have to make the decisions themselves in order to ensure human control over the AI, responsibility and accountability. This can also contribute to greater public acceptance of AI applications.
How can the EU support?
Despite a lack of competence in the field of social security, the EU can help with promoting the use of AI in the area of social security. The challenge of using trustworthy and tailored AI systems profitably for both insured persons and social insurance institutions is being faced throughout Europe. Against this background, the EU can, above all, provide a platform to promote cross-border cooperation, collect use cases and strengthen the exchange of good practices. In this way, challenges and solutions can be brought together to pool resources and avoid duplication of work. At the same time, the EU can offer technical support in a coordinating role, without prescribing requirements for the implementation of AI solutions. A harmonised approach to the use of AI systems in social security can also be useful in terms of interoperability, particularly in view of the increasing need for cross-border data exchange and coordination of social security systems.
Another possibility for support can be to provide shared databases for the training of AI models. Small countries in particular may only have a limited volume of non-representative data, which can easily lead to biases. However, linking and sharing large volumes of routine data from different areas of social insurance across national borders can also significantly improve the quality of, for example, predictions about potential future insurance claims.
Furthermore, the EU can define common basic principles and objectives based on cooperation between the responsible stakeholders of different Member States. This may include ensuring human oversight and accountability when using AI, as set out in the Platform Work Directive for the use of algorithmic management. Thus, the EU can play an important role in ensuring that ethical AI systems are used in the Member States.
Another possibility for support can be to provide shared databases for the training of AI models. Small countries in particular may only have a limited volume of non-representative data, which can easily lead to biases. However, linking and sharing large volumes of routine data from different areas of social insurance across national borders can also significantly improve the quality of, for example, predictions about potential future insurance claims.
Furthermore, the EU can define common basic principles and objectives based on cooperation between the responsible stakeholders of different Member States. This may include ensuring human oversight and accountability when using AI, as set out in the Platform Work Directive for the use of algorithmic management. Thus, the EU can play an important role in ensuring that ethical AI systems are used in the Member States.