EU Talks on AI Act Stall, Set to Resume Friday

Facts

• The three-way debate between the 27 EU member states, the European Commission, and the EU Parliament on world-leading comprehensive artificial intelligence (AI) regulations was paused on Thursday after nearly 24 hours of discussions to try to sign off on a deal to finalize the so-called AI Act.¹
• Negotiations will resume on Friday morning as policymakers are attempting to pass the legislation before elections in June. Bloomberg News reported, citing people familiar with the matter, that progress has been made, but it remains unclear whether a draft will be agreed on.²
• Provisional terms on foundation models for regulating generative AI systems such as ChatGPT were reportedly struck on Thursday, with Reuters suggesting that the European Commission would focus on high-risk models while requiring detailed summaries of the content used to train general-purpose AIs.³
• Yet, according to Politico, talks that started on Wednesday have stalled over differences on what types of AI should be banned under the pioneering rulebook and whether national security should be exempted from its scope.⁴
• If a final text emerges this week, it's expected to be voted on later this month in the EU Parliament. Though lawmakers had already approved a draft version in June, the legislation wouldn't pass as France, Germany, and Italy opposed its tight regulations on foundation models.⁵
• While Brussels has adopted a risk-based approach to regulating AI, the US has taken an incentive-based approach to the emerging technology in a bid to retain developers in the country.⁶

Sources: ¹Time, ²Bloomberg, ³Reuters, ⁴Politico, ⁵Tech Monitor and ⁶Washington Post.

Narratives

• Narrative A, as provided by Euronews. The EU has a moral duty to reach an agreement on the AI Act as soon as possible to protect the common good and address the risks inherent to the absurd proposal to refrain from regulating foundation models. Otherwise, one of the biggest scandals in its history will have taken place as lobbyists would have co-opted institutions to promote non-European interests.
• Narrative B, as provided by The Hill. Despite its downsides, self-regulation remains the best policy course to deal with AI, even compared to well-intentioned government regulation. The very complex nature of the controversial technology and its dynamic pace would make it challenging and costly to enforce government-imposed guidelines. Additionally, countries that over-regulate may lag behind in the global AI race.

Predictions