Why Europe needs its own AI foundation models

—

a coffee mug with ai written on it on a desk with note db6d713c 6dd0 4147 b8c8 2f7ca767d06f 2

The European Union is currently pushing ahead with the development of its own AI foundation models. The European Commission recently presented the Resilience 2.0 strategy paper, which sets out, among other things, how investment and an adapted regulatory framework could help build European AI infrastructure and data resources. The objective is clear: Europe should become technologically sovereign in a world where powerful AI systems increasingly underpin economic competitiveness, digital infrastructure and societal decision-making.

The strategy paper includes the following measures relating to AI:

  • Developing and supporting major European AI projects (“AI Gigafactories”) and a General-Purpose AI Code of Practice
  • Simplifying financing and regulatory processes
  • Promoting a values-based innovation culture (“EU values-based innovation model”) as an alternative to market-driven models in the United States and state-led models in China

What does this mean for the use of AI in professional research?

At present, companies from the United States and China dominate the market for so-called foundation models – large, powerful AI models that can be used for a wide range of applications. This dependence creates risks, including political influence, limited transparency, security concerns and a lack of control over training data and model behaviour. Chinese AI models, for example, are known in some cases to systematically distort or omit certain political or historical facts without clearly indicating that this is happening. US systems could also become increasingly susceptible to economic or geopolitical interests as ties between major technology companies and the US administration grow closer, and these interests may not necessarily reflect European values such as data protection, freedom of expression or ethical standards.

More and more processes in public administration, the media, academia and business are now supported by AI, from text analysis and research to decision support. If these systems rely on unclear sources, political bias or approaches that raise data protection concerns, this can undermine not only the quality of the results, but also trust and legal certainty. Choosing European AI could therefore be a first step towards greater quality, security and control in the digital workplace.

Switzerland recently made an attempt to provide a trustworthy, secure and transparent alternative with its AI model Apertus. It will be interesting to see what experience is gained from its use.

By Markus Hoffmann