Belgium pitches EU agency to screen algorithms

BRUSSELS — Belgium thinks the European Union needs an agency with technical expertise in algorithms — and it will push for one during its stint at the Council presidency next year.

EU legislators are currently in the final negotiations on the bloc’s rules on artificial intelligence. Other EU rulebooks, like the one on content moderation, will also necessitate scrutinizing algorithms, especially those used by very large online platforms. Belgium now shows a concern that the bloc won’t have sufficient — or else, too fragmented — technical comprehension to underpin the governance of AI and enforcement of other rules.

"We have to understand how an algorithm works before we’re able to identify whether certain things are acceptable or not," Mathieu Michel, the country’s junior minister for digitalization (and brother of European Council President Charles Michel), told POLITICO in an interview. "Having a structure that allows [us] to come up with an objective analysis of the algorithms, that’s something that is important."

Europe’s scattered response to TikTok, with several countries and the EU institutions all evaluating the app’s risk on their own, has also shown why there’s a need for EU-wide technical capacity to screen algorithms, he argued.

A fully-fledged European Algorithms Agency could be such a structure, Michel said, referring to the European Medicines Agency or the bloc’s cybersecurity agency (ENISA) as a blueprint for such an endeavor.

Michel doesn’t want to start from scratch with the new agency. He suggested upgrading the status of the recently opened European Centre for Algorithmic Transparency (ECAT) in Seville, which will provide technical expertise on AI-powered systems to spot infringements of the bloc’s content-moderation rulebook, the Digital Services Act (DSA). ECAT is currently managed by the Commission’s Joint Research Centre, an in-house think tank, and the Commission’s digital department.

In anticipation of the EU’s AI rulebook, some of the bloc’s data protection authorities have already taken on new algorithm-surveillance powers — raising the question of whether Belgium is not too late with its pitch. But Michel warned that in the enforcement of the EU data protection rules, Europe is left with authorities that are "fragmented," adding that AI governance should "not take the same route."

Belgium has already discussed the idea with about 10 fellow EU countries and will put it on the agenda when it takes over the Council presidency on January 1, 2024.