(Credit: Bloomberg/Getty Images)
The European Union's Parliament on Wednesday approved the AI Act, which will regulate generative AI and broader AI technology in the region. The vote tally was 523 members in favor, 46 against, and 49 abstained.
AI Act co-lead Brando Benifei, of Italy, tweeted that the act, which took two years to pass, is designed to put humans before AI. "Thanks to Parliament, unacceptable AI practices will be banned in Europe and the rights of workers and citizens will be protected," according to Benifei, who added that a European Artificial Intelligence Office will be set up to help companies comply with the rules before they come into force.
EU Commission President Ursula von der Leyen responded to the news with a note that the AI Act will help Europe's workforce and "set a blueprint for trustworthy AI throughout the world."
The AI Act's passage means that once the law is finalized and put into effect, AI systems will be categorized into one of four risk categories and be regulated accordingly. Any AI tools deemed "unacceptable" would be subject to an outright ban, such as biometric categorization, social scoring, emotion recognition, and predictive policing tools powered by AI.
Any AI tools being used for transportation, critical infrastructure, education, safety, employment, law enforcement, immigration, or court records purposes, for example, would be deemed "high-risk" and regulated (but not necessarily banned) under the law. Only AI applications deemed "minimal risk" or "no risk," such as AI-powered video games, would escape EU oversight,
High-risk applications won't be allowed to go public in the EU until they have provided adequate documentation to ensure they're safe and use "high quality" data, the commission says.
When people encounter AI tools like chatbots online, the law will require companies to disclose to EU users that they're interacting with an AI, not a real person. "Humans should be made aware that they are interacting with a machine so they can take an informed decision to continue or step back," the commission says.
The European Parliament also notes that general-purpose AI systems (GPAI) will have to comply with EU copyright laws and publish "detailed summaries of the content used for training." Copyright holders are allowed to protect their work from text or data mining, meaning that as long as their work has been opted out from such uses, generative AI developers will require permission from the copyright holders to use their work.
Once it becomes law, the EU's AI Act will likely have massive implications for generative AI models trained on vast quantities of often copyrighted web content. It's currently unclear how AI firms will respond to the law. AI companies could choose to adapt and make their tools comply with the Act's provisions, start a new legal battle, or leave the EU entirely.
It could take up to three years for the AI Act's provisions to be fully implemented. "Prohibitions will take effect after six months, the governance rules and the obligations for general-purpose AI models become applicable after 12 months and the rules for AI systems...will apply after 36 months," the commission says.