After a marathon 36-hour negotiation session, European Union member states and lawmakers reached an agreement on "historic" rules governing artificial intelligence (AI), including models like ChatGPT. The deal, known as the "AI Act," was finalized in Brussels, marking the first comprehensive set of regulations for AI usage on a global scale. Thierry Breton, the EU's internal market commissioner, hailed the agreement as a pivotal moment, emphasizing its significance beyond being a mere rulebook, envisioning it as a catalyst for EU startups and researchers to lead the global race for trustworthy AI.
The rushed legislative process, triggered partly by the emergence of the mass-market chatbot ChatGPT in late 2022, addressed concerns about the potential misuse of AI technology. Despite the remarkable advancements demonstrated by ChatGPT in generating coherent essays and poems, critics raised alarms about ethical considerations. Generative AI, which encompasses tools like Google's chatbot Bard, Dall-E, Midjourney, and Stable Diffusion, enables the rapid creation of text, images, and audio based on simple language commands.
Negotiations faced challenges, with an initial 22-hour session on Wednesday resulting in a temporary impasse before talks resumed on Friday. There was no strict deadline, but EU leaders aimed to secure a deal before the year-end. The AI Act, proposed by the European Commission in 2021, aims to regulate AI systems by assessing the risk posed by software models, with greater obligations for higher-risk applications.
While the law still requires formal approval from member states and the parliament, the recent political agreement is viewed as a significant milestone. The EU's focus on AI aligns with global concerns, as evidenced by US President Joe Biden's executive order on AI safety standards in October. China has also implemented legislation specifically regulating generative AI.
A crucial aspect of the negotiations was balancing regulations for general-purpose AI systems like ChatGPT without hindering the growth of European champions in the sector. The agreed-upon two-tier approach involves transparency requirements for all general-purpose AI models, with stricter regulations for more powerful models. Issues surrounding remote biometric surveillance, particularly facial recognition in public places, led to restrictions on real-time facial recognition with limited exceptions for law enforcement and national security.
Despite the agreement, not all stakeholders are entirely satisfied, with concerns raised about potential negative impacts on the European economy and the risk of deterring European champions in the AI sector. The EU AI office, a new regulatory body, will monitor and sanction violations, imposing fines of up to seven percent of a company's turnover or 35 million euros, whichever is larger.
Â