European Union officials have reached a provisional deal on a legal framework for the development and use of artificial intelligence within Europe, calling for greater transparency as well as setting parameters for high-risk AI.
The political agreement, which has yet to be detailed and came together following 37 hours of debates within the European Commission, highlights what is prohibited when it comes to AI, key requirements for using high-risk AI and penalties.
For the entertainment industry, the most crucial aspect when it comes to AI is the requirement for transparency when using generative AI or copyright-protected work.
The official text states that users should be informed when interacting with a chatbot, and it requires AI systems that generate or manipulate text, image, audio or video content (such as a deep-fake tool) to disclose that the content has been artificially generated or manipulated. This also applies to “artistic, creative, satirical, fictional, analogous work,” in which case the “transparency obligations are limited to disclosure of the existence of such generated or manipulated content” in a way “that does not hamper the display or enjoyment of the work.”
The AI Act also prohibits emotion recognition systems at work and in education, as well as the use of AI to exploit people’s age, disability, behavioral manipulation and circumvention of free will; untargeted scraping of images for facial recognition; and biometric systems using sensitive characteristics. The text also gives proposed penalties of up to 7% of global annual turnover or €35 million for prohibited Al violations and up to 3% of global annual turnover or €15 million for most other violations.
The AI Act has been met with a mix of relief and concerns by European authors’ guilds and producers.
Juliette Prissard, the general delegate of Eurocinema, a guild of film and TV producers, laments that the transparency requirements don’t apply to generative AI used internally, notably for R&D purposes.
“The transparency guidelines only concern tools that are commercialized but not software that are being used in-house by companies,” says Prissard, who has been tracking all the debates on AI at the European Union. “As such, streamers will continue quietly using these generative AI tools to gather data on tastes, etc., without communicating it and that’s problematic.”
GESAC, the org representing 32 European authors’ societies, said “the actual wording of the text resulting from the political agreement is to be fine-tuned at forthcoming technical meetings.” The org suggested that a “new AI office will be tasked with conducting further work to formulate practical arrangements for its implementation.”
Véronique Desbrosses, GESAC’s managing director, indeed called for a “a robust implementation” of “transparency requirements and respect for EU copyright rules applicable to all models.”
France has been one of the driving forces behind the AI act. French lawmakers were initially pushing to have restrained limitations on its use, claiming it would slow down innovation, but they triggered a strong pushback from creative industries emphasizing the importance of authors’ rights and the danger of deep fakes.
“French lawmakers initially argued that we shouldn’t over-regulate AI in Europe because we needed to compete with the U.S.,” says Jean-Pierre Giansilj, who is in charge of institutional and judicial relations at the UPC, the French producers guild. “While we understand that it’s important to create the perfect conditions for France to become champions of AI, we also know it can’t happen without a solid foundation that respect sthe basis of French authors’ rights.”
Plissard says people in the film and TV industry started talking about AI only last year with the arrival of ChatGPT. “These discussions around this AI Act have been in the work for much longer, so we came in very late but thankfully we’ve been able to make a significant impact in a relatively short time,” she says.
The set of proposed regulations will be voted on at the European Parliament early next year, while the compliance grace period has been set “between six to 24 months.”