European Union lawmakers line up to defend world's first AI Act
Rules were not bent for Big Tech, politicians say
The European Parliament has enacted the world's first legislation designed specifically to address the risk of artificial intelligence, including biometric categorization and manipulation of human behavior, as well as stricter rules for the introduction of generative AI.
In a vote this morning, Members of European Parliament approved the final text of the law, which is designed to protect the public in the fast-developing field of general-purpose AI (GPAI) models – a term used in the law to encompass generative AI such as ChatGPT. AI models will also have to comply with transparency obligations and EU copyright rules. The most powerful models will face additional safety requirements.
The law will require online content using AI to fake real people and events to be clearly labeled rather than duping people with "deepfakes."
While critics argue that the rules were watered down at the last minute, even suggesting lobbying from US tech giants via EU partners to alter the legislation, Forrester principal analyst Enza Iannopollo said it was a necessary compromise to get the laws enacted.
"There is the opportunity for the EU to go back and try to review some parts in the annexes. I think it is a compromise. Could it have been better? Yes. Was it a good idea to wait longer? I really don't think so," she told The Register.
According to Bloomberg, the French and German governments intervened in the stricter regulations to protect homegrown companies Mistral AI and Aleph Alpha. Others noted that Mistral has accepted a €15 million ($16.3 million) investment from Microsoft.
Campaign group Corporate Europe Observatory raised concerns about the influence that Big Tech and European companies had in shaping the final text.
The European Data Protection Supervisor said it was disappointed in the final text, labeling it a "missed opportunity to lay down a strong and effective legal framework" for protecting human rights in AI development.
However, in a press conference held before the vote, politicians responsible for negotiating the text said they had achieved a balance between protecting citizens and allowing companies to innovate.
Brando Benifei, from Italy's Socialists and Democrats party, said the legislators stood up to lobbyists. "The result speaks for itself. The legislation is clearly defining the need for safety of the most powerful models with clear criteria. We delivered on a clear framework that will ensure transparency and safety requirements for the most powerful models."
- Europe probes Microsoft's €15M stake in AI upstart Mistral
- Google wants regulators to take Microsoft down a notch before it stifles AI
- Meta to build election operations center in Europe to inspect AI content
- British businesses told: Compliance with EU AI law will satisfy UK guidance
Benifei said that at the same time, the concept of sandboxing [PDF] allows businesses to develop new products under a regulator's supervision and would aid innovation.
"In our commitment as Parliament to having a mandatory sandbox in all member states to allow businesses to experiment and to develop, we have in fact chosen a very pro-innovation. If you look at the polls, too many citizens in Europe are skeptical of the use of AI and this is a competitive disadvantage and would stifle innovation. Instead, we want our citizens to know that thanks to our rules, we can protect them and they can trust the businesses that will develop AI in Europe. That, in fact, supports innovation."
Dragoş Tudorache, from the Renew party, Romania, said the legislators had stood up to pressure, particularly in copyright infringement.
In September, the Authors' Guild and 17 writers filed a class-action lawsuit in the US over OpenAI's use of their material to create its LLM-based services.
"Clearly, there were interests for all of those developing these models to keep still a black box when it comes to the data that goes into these algorithms. Whereas we promoted the idea of transparency particularly for copyrighted material because we thought it is the only way to give effect to the rights of authors," Tudorache said.
Forrester's Iannopollo said: "This is a very complex piece of legislation. There are many areas where the legislation could have been improved. One, definitely around the requirements for general purpose AI that was added at a later stage and definitely feels much less strong that the risk based approach.
"But we have to be realistic. The technology is evolving very quickly so it is very difficult to create a piece of legislation that is going to just be perfect... There is more risk in delaying the legislation in the attempt to make it better [than the imperfections]."
There is an appetite among European politicians to revisit and strengthen the legislation, particularly in terms of copyright protection, she said. ®