Since 2023, there has been remarkable growth in AI capabilities, which became widely available to the public in the progress made by tools like OpenAI’s ChatGPT, Microsoft’s Copilot or Google’s Gemini. As well as generating text, LLMs are also capable of using other generative AI models, such as for image or video generation. Moreover, tools such as Perplexity incorporate internet connectivity for real-time web search and provide accurate sources with their responses. Besides the deployment of more powerful LLMs, AI systems are sometimes allowed to make decisions on their own, granting them a degree of autonomy or agency. Such AI systems are aptly named ‘agentic AI systems’.

The combination of autonomous decision-making and the accelerating pace of innovation creates opportunities for transformative applications, including AI-driven energy-grid management to balance renewable supply and demand, and scientific discovery engines capable of autonomously generating and testing hypotheses. But they also introduce new challenges, for instance in assigning liability in court or challenging unfair or biased decisions. The rapid AI developments affect private companies and public functions in critical domains such as cybersecurity, defence and public infrastructure.

Crucially, all recent AI breakthroughs have been in the domain of so-called ‘narrow AI.’ That is, they are essentially sophisticated statistical models. They are increasingly good at predicting outcomes in a certain narrowly specified domain, for which they have been ‘trained’ by inferring logic and rules from large datasets. There are heavy debates on the future of AI and whether ‘Artificial General Intelligence’ – loosely defined as an AI that is at least as intelligent and adaptable as humans across most tasks – will emerge soon. However, the potential realisation of such technologies is explicitly outside the scope of this paper, which discusses plausible tomorrows extrapolating from current trends in narrow AI developments.

To ensure that the Netherlands remains secure and sovereign in this evolving context, it is vital, first, to take stock of where AI developments stand – from political, technical and societal points of view – to establish a baseline.[10]