For a few years now, government and administrative bodies have been devoting more and more time to exploring potential uses of Artificial Intelligence (AI) within their respective work. Administrative processes can be improved by using AI – for citizens, for the business community and for the administrative work itself. However, as clearly demonstrated in recent years by various international usage cases, AI systemscarry certain risks, including discrimination by AI systems against specific sections of the population.

There are three key factors that make it essential to address the issue of how to use AI for public administration:

1. The ongoing shortage of trained specialists is already causing many public sector job vacancies to remain unfilled. As the baby boomer generation reaches retirement age, official bodies will be particularly affected as employers. According to a 2016 study by the WifOR Institute and PwC, by 2030 every ninth job among administrators and office workers – one of the key areas in public administration – could be vacant.
2. Meanwhile, the amount of data flooding into the public sector continues to grow. This is a result of new dynamics in the labour market, such as more frequent job changes, periods of non-working (e.g. sabbaticals) and new forms of employment. Changes in the labour market and in employment relationships are also causing a rise in the workload handled by social security authorities.
3. The requirements of the population as regards the availability and quality of public services are changing. Younger generations in particular have grown used to the services provided by modern companies (e.g. digital access to information or 24-hour services) and expect the same from state-run institutions.

The potential benefits offered by AI in terms of administration are considerable, affecting nearly every function of the public sector. Some sections of public administration, including social security authorities, have recognised this potential and are already applying it in selected areas of activity.

First and foremost, the statutory social insurance authorities, including the Federal Employment Agency, Deutsche Rentenversicherung (the German pension insurance association) and the statutory accident insurance institutions, offer good conditions for the deployment of AI: many tasks in this sector are highly structured and the execution process tends to be subject to clear rules. Moreover, social insurance authorities have access to a large amount of data.

At the same time, this kind of data is subject to a particularly high standard of protection. Add to this the fact that social insurance authorities are also making decisions on applications for state benefits, decisions which in some cases can directly affect individuals’ livelihoods. Given the above, the AI-based automation of decision processes that involve margins for discretionary assessments or judgements is a controversial issue. Even now, automated decisions in public administration are permitted only under certain circumstances, and they are likely to be subject to additional regulation in the future.

Technical feasibility and legal permissibility are not the sole factors that determine whether or not an AI application will be able to fulfill its intended purpose. Scientific studies have shown that AI systems can only be implemented successfully within an organisation if they are also accepted by the employees. This means AI will only be able to develop its full potential if the administrative employees and officials affected are able to use the new applications productively (competency is key) and want to do so. They also need to see there is a concrete benefit for them in terms of getting their day-to-day work done.

As such, agile working methods and co-creative design processes play an important role in the development, introduction and “maintenance” of AI, because they can help to integrate the specialist knowledge and experience of the operators. However, as yet there is a lack of empirical findings on the conditions under which AI is accepted in the context of social insurance authorities, either by the administrators or by citizens.

This research project aims to delineate the prerequisites and success factors for artificial intelligence in the public sector. How should AI systems be designed in order to meet the exacting requirements of employees and find acceptance with the public? What conditions must be met in order for the potential of AI in the public sector to be exploited?

This research project complements the work of the Policy Lab Digital, Work & Society by generating scientifically grounded findings on what form the use of AI in labour administration and social administration might take. Using the research findings as a basis, it also formulates recommendations for policy decisions regarding the use of AI in the welfare state.