If the European Union is to develop competitive artificial intelligence (AI) models, it will need to unlock substantial cross-border investments to catch up with the US and China. Fortunately, there is a precedent: the CERN nuclear research centre.
The idea of creating a “CERN for AI” has been embraced by European Commission President Ursula von der Leyen, who outlined a proposal in July to “to set up a European AI Research Council where we can pool all of our resources, similar to the approach taken with CERN.”
Over the past 30 years, the EU has fallen behind in developing and deploying new technologies, critically undermining its economic performance. In his report on competitiveness published last month, Mario Draghi noted that “only four of the world’s top 50 tech companies are European”, identifying this as the main reason for the productivity gap between the EU and the US.
Draghi identified AI as the driver of a new “digital revolution” and an opportunity that the bloc can’t afford to miss. The technology can make workers more productive, science more successful, and countless economic sectors more profitable. These proven and promising use cases justify a projected global market value for advanced AI in the trillions of euros.
Not all AI models are created equal, however: trustworthy AI is still lacking and is an area where European developers, by working closely with regulators, can steal a march on global competitors.
Our research has found that provably lawful, ethical and robust models do not yet exist. AI models frequently “hallucinate”, a polite euphemism for “making things up”, with results ranging from the amusing – “eat a minimum of one rock per day” – to the potentially dangerous.
This undermines trust in the technology and prevents companies, governments and individuals from using it in sensitive areas where it could add great value, such as critical infrastructure operations.
Pooling AI resources
Huge investments are necessary. Our research suggests that a world-class AI developer would cost between €30bn and €35bn over three years, a sum equal to a third of the EU’s entire seven-year flagship science and innovation fund Horizon Europe.
When it comes to AI investment, bigger is better. Building a powerful and profitable AI model requires large quantities of data and computational resources, but top-shelf AI chips are too expensive for any individual European company or country to afford. With investments fragmented across the EU, initiatives like EuroHPC have been too small to create globally competitive advanced AI.
The CERN model would therefore be an EU-wide initiative to pool resources, talent and ambition into a focused effort to develop trustworthy AI models. This approach would position the EU as a global leader in the research and development of advanced general-purpose AI.
The institute should be built on a public-private model, which would provide the right incentives and capabilities for responsible AI innovation. While private companies are best equipped to rapidly develop models that meet the needs of the market, democratic governments will play a crucial role in infusing principles of trust and longer-term economic thinking.
Just as Apple created the smartphone using tools that emerged from high-risk government investments like GPS, the internet and touch screens, European governments can set a clear direction for AI development, which private companies can then advance.
There is no time to waste. In the US and China, large-scale public and private investments are flowing into advanced AI and the semiconductors their AI models are trained on. A cost of between €30bn and €35bn is significant, but without massive improvements in AI, the European economy risks missing the next tech boom, just like it missed the last one.
It’s time for a CERN for AI. This is the EU's best chance to get a seat at the table of global AI development and usher in an era of repowered European economic competitiveness.