At the request of the Ministry of the Interior and Kingdom Relations, TNO has compiled a report analysing the use of open and European language models in the Dutch government.
The Netherlands Digitalisation Strategy (NDS) states that the government aims to take a more proactive role in AI. This includes promoting the use of open-source or European generative AI models by government bodies. This is crucial for improving technical flexibility, transparency, reliability, ethical accountability, and digital independence, while also preventing risky strategic dependencies on AI. The TNO research provides recommendations that will be incorporated into further vision and policy development on using generative AI models in government.
Understanding ‘open’ and ‘European’ language models
To encourage the adoption of open and European language models in government, it is essential to first clearly define core concepts like ‘open’, ‘open source’, and ‘European’ language models. The term ‘open’ language models, in particular, can be confusing because an AI model can be considered open in various respects. Therefore, the report proposes using the European Source AI Index and the Model Openness Framework to establish openness on different levels. The 3 most common levels are: closed source, open weights and fully open.
Recommendations for further policy development
The exploration shows that the government needs support in deciding how to deploy generative AI models. The report clarifies the relationship between the criteria ‘European’ and ‘open’. For example, choosing European language models may affect openness. The report also offers recommendations for further policy development. Read the full TNO report (Dutch).
