EU AI Act ‘cannot turn away from foundation models’, Spain’s state secretary says

With Europe’s AI rulebook at a delicate stage of negotiations, Carme Artigas, the Spanish Secretary of State for Digitalisation and Artificial Intelligence, spoke to Euractiv about the state of play of the inter-institutional discussions.

The AI Act is a legislative proposal to regulate Artificial Intelligence based on its capacity to cause harm. The file is at the last phase of the legislative process, so-called trilogues, whereby the EU Commission, Council and Parliament hash out the final provisions.

Last week, the negotiations broke down on foundation models, powerful types of Artificial Intelligence on which other AI systems can be built, like ChatGPT is powered by OpenAI’s GPT-4.

Following mounting opposition from France, supported by Italy and Germany, the Spanish presidency, representing EU countries in the negotiations, asked for a rethink of the approach to these models, prompting parliamentary officials to leave the room, in a bid to send a political message.

On this hot issue, the state secretary admitted that they are not covered in the EU Commission or Council’s mandates because what she calls the ‘ChatGPT eureka moment’ was yet to show its full potential.

“We cannot just say we don’t see that because it is here. We cannot turn our heads away from it,” Artigas said. “There is a consensus that, even though foundation models are deployed in non-high-risk use cases, they can propagate a systemic risk along the value chain. We are all aware of this.”

At the same time, the Spanish official sees regulating this sort of model as a trade-off that risks jeopardising the risk-based and technology-neutral approach of the AI law.

“We are now in a landing zone in which we all feel comfortable: not regulating the foundation models as high-risk, with all the compliance that follows that,” Artigas continued, explaining that instead, they are fine-tuning the legal text with codes of conduct and transparency obligations.

Asked by Euractiv, the state secretary further elaborated that there is now a common ground that some transparency obligations will be agreed upon. In contrast, some additional obligations are still under discussion, including how to deal with copyrighted content.

“Copyright is a matter of concern. We all understand that copyright, as any other law, cannot be infringed. We are trying to find the right balance in making that possible without damaging Europe’s competitiveness,” she added.

On the tiered approach to have stricter obligations for the ‘high-impact’ foundation model, on which originally there seemed to be a consensus at the last political meeting in mid-October, Artigas said that it is still ‘under discussion’ at the political level.

The Spanish presidency is expected to come up with a compromise proposal that could be acceptable to at least some of the reluctant countries, as well as the European Parliament. A discussion is expected to take place in both internal meetings and at the political level between the main EU institutions next Tuesday (21 November).

“We must be very open-minded and humble because we have no clue of the next potentiality of this technology in six months,” the state secretary added, noting that her door was open also to hear the views of the industry and academia.

“We cannot just regulate ex-ante as the technology will evolve. Asking for transparency for some controls makes sense, but this is a law, and it must guarantee accountability and enforcement. That should be our focus,” she continued.

In that regard, the state secretary highlighted the need not just to have a beautiful legal text – but also something actionable. Spain has been a frontrunner in setting up an AI agency and regulatory sandbox to keep a permanent dialogue with the industry.

In terms of international governance, Artigas is also involved in the discussions at the UN level as co-chair of the High-Level Advisory Body on Artificial Intelligence that will come up with a proposal by the end of the year that might range from models like the Intergovernmental Panel on Climate Change to the International Atomic Energy Agency.

At the same time, the Spaniard argued that the AI Act should not put an excessive bureaucratic burden that prevents new players, especially in Europe, from competing with incumbents.

Regarding the ongoing negotiations, Artigas said they are looking at the foundation model aspect holistically rather than following a step-by-step approach.

“The best negotiation is the one that leaves everyone equally unsatisfied,” she said, stressing that she is confident a consensus can be reached but foreseeing that the trilogue on 6 December “will be a long one”.

Asked by Euractiv whether she would consider the Spanish presidency a success even if they did not reach an agreement on the AI Act, Artigas replied affirmatively. Still, she admitted that an agreement on the AI law would be the ‘cherry on the cake’.

For Artigas, the point is not a success for the Spanish presidency, but the fact that the clock is ticking. “Nobody’s waiting for Europe. Everybody’s moving forward.”

[Edited by Nathalie Weatherald]