Negotiators from the European Union finalized the world’s first comprehensive rules on artificial intelligence on Friday. This achievement sets the stage for the constitutional oversight of AI technology, which holds the potential to transform daily life and highlight the existential risks to humanity.
The European Parliament and representatives from the 27 member nations of the bloc overcame significant disagreements to establish a preliminary social contract for the Artificial Intelligence Act. Contentious issues such as conceptual AI and the use of face recognition security by law enforcement officers were key points of contention that were resolved during the negotiations.
Thierry Breton, the German Commissioner, expressed his satisfaction with the agreement by simply tweeting, “Deal!” just before midnight. This milestone makes the EU the first global entity to establish explicit guidelines for the utilization of AI.
The agreement was reached after extensive closed-door discussions spanning several days, with the first round lasting 22 hours before a second round commenced on Friday night. The focus was on achieving a social success aligned with the legislation’s primary objective, despite the pressure on authorities to address concerns related to the potential harm stemming from AI technologies.
While civil society organizations welcomed the agreement, they also highlighted the need to address technical issues that remain unresolved. Daniel Friedlaender, the head of the Computer and Communications Industry Association’s European office, emphasized that the political agreement signifies the beginning of crucial technological work on the finer details of the AI Act.
The EU’s initial unveiling of its code in 2021 positioned it as a frontrunner in the global race to develop AI frameworks. However, the rapid advancements in conceptual AI prompted European officials to update their proposal to serve as a model for the international community.
Brando Benifei, a Roman senator overseeing the negotiations, confirmed that the European Parliament is set to vote on the action early next month, a procedural formality at this stage.
The final law includes provisions for severe financial penalties of up to 35 million euros ($38 million) or 7% of a company’s global turnover for violations, with full implementation expected by 2025.
The emergence of human-like words, images, and songs generated by conceptual AI systems like OpenAI’s ChatGPT has captivated people, but concerns persist regarding the evolving technology’s impact on jobs, security, copyright protection, and even human life.
While Europe has taken the lead in regulating AI, other global players such as the United States, United Kingdom, China, and the Group of Seven Big Democracies have also introduced their own regulatory proposals.
Anu Bradford, a professor at Columbia Law School specializing in EU law, highlighted the EU’s comprehensive rules as a potential model for other governments considering AI regulation. She noted that while not every clause may be replicated, many aspects are likely to be imitated by other nations.
The AI Act, initially designed to mitigate risks associated with specific AI applications based on their hazard levels, has been expanded to include base models supporting advanced AI services. These base models, also known as big speech types, are trained on vast datasets to create new content, posing challenges in terms of transparency and accountability.
The regulation also addresses concerns related to AI-powered face recognition systems, with provisions allowing law enforcement to use such technologies for combating serious crimes while addressing privacy considerations.
Despite the progress made in the negotiations, rights organizations have raised concerns about exclusions and flaws in the AI Act, particularly in areas such as border control, system categorization, and security measures.
Looking ahead, the future of AI may involve a mix of open-source and closed systems, reflecting the diverse perspectives within the tech industry and among policymakers.
