(Credit: Santiago Urquijo/Getty Images)
AI regulation is moving forward in Europe.
The European Union's Internal Market and Civil Liberties Committees voted overwhelmingly in favor of supporting the Artificial Intelligence Act on Tuesday, an important step toward codified AI laws in Europe and the first of its kind globally.
The bill previously saw a provisional agreement back in September, and has since undergone a round of negotiations, according to a European Parliament announcement.
The latest version of the AI Act, if passed into law, would ban a range of different uses of AI, providing civilians with more legal protections and privacy around some of AI's more dystopian potential use cases. Biometric categorization, "untargeted" scraping of facial images from the internet or surveillance footage, AI-powered "emotion recognition," social scoring, and predictive policing tools (à la Minority Report) would become illegal in the EU under the AI Act.
Any type of "manipulative" AI that could be used to exploit human behavior or vulnerabilities would also be prohibited under the Act.
While police deploying AI for biometric purposes is "prohibited in principle," according to the European Parliament's announcement, EU lawmakers have carved out a "real-time" exemption for police to use AI tools that tap into biometric data under a few restricted use cases. Police would be able to conduct real-time, AI-powered biometric analyses for specific urgent cases, like trying to find missing people or "preventing a terrorist attack," to name the two listed examples.
Law enforcement's use of AI-powered biometric tools after events have occurred are deemed "high-risk" under the AI Act and would only be possible if connected to bringing a criminal to justice. A judge or other legislator would have to grant explicit permission for such use cases.
The AI Act also sets guardrails for AI use in "high-risk systems" like education, healthcare, banking, law enforcement, immigration, and elections. The Act would also give civilians a way to submit formal complaints about how AI might be used in these fields.
AI tools determined to be "General-Purpose AI systems" (GPAI) would have to comply with existing EU copyright laws and meet transparency requirements. The bill also lets lawmakers impose stricter requirements on any GPAI models deemed higher-risk.
It's unclear to what extent copyright laws might be applied to AI, though, and whether images created "in the style of" another artist, for example, could be considered in violation of copyright.
The AI Act would also make any unlabeled deepfakes illegal, meaning victims of AI-manipulated or generated images, audio recordings, or videos could have legal recourse under the Act.
If passed into law, the legislation would take some time to fully come into effect. Some parts of the bill would apply within six months, while other parts of the bill would take two years to 36 months to become fully enforceable under EU law.
While the EU is among the first to push AI regulation forward, the US is taking some steps of its own. Last week, US President Joe Biden announced an AI Safety Consortium inspired by a previous executive order that brings together over 200 different companies and institutions like Microsoft, Nvidia, Meta, and others, with the goal of mitigating AI's risks and ensuring the US remains competitive in AI development at a global scale.