OECD publishes report on use cases and further potential.

In June, the Organisation for Economic Co-operation and Development (OECD) released a report entitled “AI and the future of social protection in OECD countries”. It presents use cases of artificial intelligence (AI) in the area of social protection and identifies untapped potential – for example in forecasting needs and social shocks with consequences on both individual and societal levels, or in improving take-up of social benefits. However, OECD member states are still very cautious in this field due to potential data protection risks and the possibility of errors in automated decision-making processes.

Widespread use cases in OECD countries

Current applications mainly focus on supporting benefit recipients, automating internal administrative processes and tackling fraud. In the area of user support, chatbots and digital assistant systems to improve online interaction are particularly popular. In most cases, they are used to provide citizens with information and respond to enquiries. Some countries use them proactively, for example to ask about users’ well-being.

In addition to these support tools, AI applications offer the possibility to automate internal processes and reduce the time required for administrative tasks. Examples include processing large volumes of data from traditional databases and analysing unstructured content such as texts and images from scanned paper documents.

Finally, AI is frequently used for error and fraud detection, with the OECD expecting a steep increase in this area in the coming years. AI can identify potential fraud cases at an early stage and thus support prevention efforts. In Portugal, for instance, AI-based facial and voice recognition is used to verify proofs of life from civil servants living abroad who are claiming pension benefits.

Untapped potential of predictive analytics

According to the OECD, predictive analytics is a particularly promising area of AI application in social protection. These analyses offer opportunities in three areas:

First, they can help to prevent social shocks and crises or at least enable an early response. In Germany, for example, AI is already used to predict the likelihood of workplace accidents. With the help of an AI application, the statutory accident insurance for the construction sector (BG BAU) identifies companies with the greatest need for support and accordingly directs inspection and supervision personnel. If shocks cannot be prevented, AI systems can at least help to mitigate their consequences. In Togo and Ireland, for instance, AI is used to forecast floods.

Second, predictive analytics can be used to identify vulnerable individuals at an early stage and initiate targeted preventive measures. Likewise, potential beneficiary groups – for example by region or sociodemographic characteristics – can be identified in order to develop tailored support offers. Given the significant returns of some preventive measures, the social and economic benefits of such AI applications can be considerable. There are already examples of AI being used to prevent homelessness, repeated domestic violence, and to support care and income security for older people. Despite the potential AI offers in this context, the OECD points out that substantial progress can also be achieved using existing tools – without AI – and high-quality, linked personal data.

Third, predictive analytics can help to better reach eligible individuals and reduce the non-take-up of social benefits. Some population groups most in need of services are often difficult to reach. For example, young people not in education, employment or training benefit particularly from early interventions to prevent long-term unemployment. However, they are difficult to engage in social programmes, as they are often unaware of relevant services and rarely contact the authorities themselves. Against this background, AI is already being used in a few cases to identify people in need of support and offer them help proactively.

Outlook

According to the OECD, governments should continue to pilot AI applications in social protection and determine where the use of AI is the appropriate tool. In many cases, significant progress can still be made using other approaches – such as simplifying registration procedures through linked administrative data. For each specific use case, a clear rationale should be formulated, aligned with policy objectives. At the same time, early and transparent public engagement is necessary, as public trust remains fragile – only 40 percent of respondents in 27 OECD countries currently see a benefit for users in AI-based application processing. These findings are also relevant from the perspective of the German social insurance. The increasing use of AI in the responsible organisations must always be linked to a benefit for the insured and the greatest possible transparency.