What are the potentials and limits of using artificial intelligence (AI) in nuclear waste management? Selected experts met on 1 December to get to the bottom of this question, and to develop initial answers. The Federal Office for the Safety of Nuclear Waste Management (BASE) had invited the participants.

AI is currently being discussed in many areas of society, such as customer service, journalism, and science. Intelligent programmes can take over tasks from humans everywhere - sometimes with astonishing results. AI methods are also gaining importance in the repository site search for high-level radioactive waste.

This applies, in particular, to the natural sciences and technology, where AI applications are intended to reduce uncertainties, for example in the computer-based modelling of phenomena and processes. In the social and political sciences, AI may become relevant in the future for the systematisation of data material or in processes of public participation.

Can AI make a repository safer?

But what does that mean in concrete terms? Can these technologies make repository site search decisions 'better'? Can we actually achieve more safety by using them, and can risks be reduced?
Such questions have hardly been asked in the context of the repository site search far, and formed the impetus for BASE's discussion series. There were diverse contributions at the kick-off event. They addressed topics such as:

• Prerequisites for the use of AI methods in nuclear waste management,
• the potential of AI in dealing with geodata,
• AI experiences from other environmental policy application fields,
• Findings from the AI-supported evaluation of participation processes that could also become relevant in the repository site search, as well as
• ethical and legal framework conditions that already determine the use of AI today and are currently being negotiated at EU level with a view to the 'Artificial Intelligence Act'.

Quality of the data material is crucial

The discussions showed that AI methods certainly hold potential for the disposal of radioactive waste. However, their limits also became apparent, in particular with regard to the available data material and the availability of data. If AI is to be used for image recognition, for example, sufficient image material must first be available to train the AI method applied. In addition, corresponding IT infrastructures as well as human resources are needed where AI is used. Furthermore, it is unclear how robust the results of AI methods would be, and whether AI applications should not rather act as 'control instances' that reduce errors and uncertainties.

Actors involved in the site search for a repository are also faced with specific questions: What would happen, for example, if the Bundesgesellschaft für Endlagerung (BGE) actually used AI in the search for the most suitable site for a repository? As a supervisory and licensing authority in the repository site search process, BASE would have to clarify potential regulatory needs.

The panel kicked off a series of discussions on AI in the context of nuclear waste management that BASE will be hosting.