Application of AI

Application of AI in social insurance

Overall, the rapid development of AI is impacting social insurance in two ways. On one side of the coin are AI-based applications used by the social insurance institutions themselves. In this context, applications such as those to facilitate internal processes, communication with insured persons or fraud detection are still rather simple. On the one hand, these open up new opportunities for insured persons; on the other, AI systems can also reduce stress of employees in the workplace, thus contributing to better health in the workplace. This use of AI systems results in direct obligations under the AI Act: as operators of AI systems under their own responsibility, social insurance institutions must classify the AI applications they use and implement the provisions of the AI Act that apply to them.

On the other side of the coin is the use of AI in the workplace or in the healthcare sector. In this case, the social insurance institutions are not operators of the AI systems, which is why the AI Act does not specify any direct instruction. Nevertheless, AI systems used in these areas have implications for social insurance, as they affect, for example, occupational safety and health or the quality of medical care for insured persons, which in turn is reflected in any claims for benefits. Hence, AI applications in these areas relevant to social security must be carefully examined and should only be supported if they can lead to better occupational safety and health and better services for the insured persons.

AI in statutory health insurance

So far, only a few AI projects have been implemented in statutory health insurance. Of course, there is potential, for example in the area of utilising health data that is available to health insurance funds through the provision or financing of services. Provided that data protection and data security are guaranteed, this data could be used to determine future care needs, individualise healthcare services and improve care structures, among other things.

The “KI-THRUST” research project sponsored by the Innovation Fund (Innovationsfonds), for example, is moving in this direction. Based on routine data from statutory health insurance, it aims to individually predict the course of illnesses and therapies, thus recognising the care requirements of patients earlier and better.3 While health and healthcare research has so far mostly relied on traditional analysis techniques, KI-THRUST is investigating the potential of AI-enabled prediction methods and comparing the results with those of conventional methods. In the course of the project, for example, the extent to which AI processes can predict certain requirements and difficulties following hospital discharges will be tested. This is based on data records from more than two million hospital discharges.

In contrast to the still rather few AI applications by social insurance institutions, the use of AI in the healthcare sector, which has an indirect impact on health insurance funds, is diverse. Earlier detection of illnesses, better care and a reduction in the workload of medical professionals can significantly reduce healthcare expenditure and thus the burden on health insurance funds. The use of AI ranges from medical diagnostics and drug development to data and process management in hospitals and medical practices, to name just a few fields of application. Medical professionals and patients are increasingly being supported by AI systems, the latter through individualised therapy and follow-up care at home. In contrast, optimised healthcare services are based on the digital networking of patient data, public health data and data from health apps and smart wearables (for example fitness watches). AI is particularly well developed in the area of analysing medical images in combination with data analyses of medical histories, for example to individually predict the course of illnesses and therapies. Other than that, in the field of rehabilitation, AI systems can support people with limited mobility in movement therapy according to their needs, meaning tailored to their individual abilities and requirements.

AI in statutory accident insurance

Accident insurance institutions have also been using AI, for example for the automated, needs and risk-based selection of companies that should be visited and assisted in view of their accident history. The aim of this is to develop preventive measures to avoid accidents and occupational illnesses. This method was used, for example, by the German employers’ liability insurance association for the construction industry (Berufs­genossen­schaft der Bauwirtschaft – BG BAU) as part of a lighthouse project on AI-based support for targeted accident prevention, which ran from February 2023 to May 2024. The employers’ liability insurance association for energy, textiles, electrical and media products (Berufs­genossen­schaft Energie Textil Elektro Medien­erzeugnisse – BG ETEM) also uses AI to select the companies to which inspectors are sent. Being one of the ten prevention services provided by BG ETEM, company inspections should be carried out primarily when the risk of a company is considered to be high – also in order to conserve human resources. An algorithm for predicting loss events helps with risk assessment and thus the selection of companies by correlating individual key figures based on accident and occupational disease incidents and other prevention data. However, the algorithm does not derive any specific company-related preventive measures; these must still be recognised or developed by the inspectors.

However, it is not only AI systems used by the accident insurance institutions themselves that are relevant for the area of statutory accident insurance. The use of AI in the workplace also plays a major role from an occupational safety and health perspective. AI expands the possibilities of technical accident prevention in the form of innovative assistance and protection functions. At the same time, AI systems can influence the physical and mental stress of employees. One prominent example is algorithmic management in the workplace, which was regulated at EU level for the first time and in a pioneering way by the Platform Work Directive. On the one hand, algorithmic management allows tasks and processes to be better controlled, which can help to improve working conditions by preventing accidents and reducing stress, among other things. On the other hand, constant monitoring and assessment by algorithmic management can lead to considerable psychological stress. Against this backdrop, the “Competence Centre for Artificial Intelligence and Big Data” (Kompetenzzentrum Künstliche Intelligenz und Big Data – KKI) at the Institute for Occupational Safety and Health (Institut für Arbeitsschutz – IFA) of the German Social Accident Insurance is involved in research, con­sulting and standardisation of trust­worthy AI and provides support in the development, testing and certification of AI systems.

AI in the statutory pension insurance 

Algorithmic management methods and bots are also being used in the area of pension insurance to speed up the processing of procedures, make decisions more reliable and make it easier for insured persons to access pension insurance. One example is the first AI project of the German Federal Pension Insurance (DRV Bund) concerning risk-oriented employer audits (Künstliche Intelligenz für risiko­orientierte Arbeitgeberprüfungen – KIRA). KIRA supports company audits, during which employees from the DRV Bund’s company audit service check the correct payment of social security contributions every four years. In view of the approximately 400,000 audits that need to be carried out each year, employees hardly have time for a full inspection; instead, they set priorities and limit themselves to spot checks. In future, KIRA will provide support by reading all the companies’ digital data, searching for patterns and flagging anomalies such as unusually high or low contributions or missing evidence in the documents. Based on this information, audit staff decide which cases deserve more detailed consideration and which can be closed quickly. It is important to note that the experience of the employees and the human decision are also essential in this example of use of AI. The AI system simply provides support, thereby contributing to greater efficiency and mitigation of the shortage of skilled labour.

While pension insurance institutions are increasingly using AI-based applications, the indirect impact of the use of AI in the world of work on pension insurance is rather limited. However, it can be argued that appropriately regulated, supportive AI can generally be expected to simplify work processes and simplify everyday working life, thus reducing stress among employees. This can have a positive effect on their health, reducing the number of absences due to illness or even retire­ment. This would also reduce the bur­den on pension insurance by reduc­ing the need for medical reha­bilitation or reduced earning capa­city pensions, for example.