Posted by 
BRUSSELS—The European Union is banning some “unacceptable” uses of artificial intelligence from Feb. 2, setting a world first for curbs on the nascent technology.
But critics say the bans have too many loopholes that allow European police and migration authorities to use AI — including technology that can quickly identify faces on security cameras — to help track perpetrators of serious crimes, such as terror attacks.
The European Union’s AI Act no longer allows AI to be used to profile whether someone will commit a crime, with such “predictive policing” one of seven categories of AI use now banned. Scraping images from the internet to build up a database for facial recognition or deducting a person’s emotions from biometric data is also prohibited.
While the bans enter into force this Sunday, governments still have until August to nominate the authorities that will enforce them.
The AI Act is a world-first in setting rules for how the new technology can be used. The European Commission’s tech boss Henna Virkkunen lauded the bill last year as one that will “protect our citizens” while also being an “enabler for innovation.” Creating safeguards for the nascent technology is viewed as essential to enable widespread use. Other parts of the law will be rolled out gradually over the next year and a half.
The rules set the EU apart from other parts of the world where there are few similar rules to govern the technology, underlining its position at the regulatory vanguard. U.S. President Donald Trump took the opposite tack by ripping up an executive order on AI safety from his predecessor, President Joe Biden.
Brando Benifei, an Italian lawmaker who helped to negotiate the rules, said the bans aim to avoid that AI could be used for “societal control” or the “compression of our freedoms.”
“The bans are very much concentrated on one area of concern, which is linked to the protection of our democracies,” he said.
Enforcing them could be messy. European police and migration authorities are also using several AI practices, like real-time facial recognition in public spaces, and were able to secure long lists of exemptions in the rulebook to continue doing so.
Academics and activists are concerned and watching closely how the law will be applied.
“You can even question whether you can really speak of a prohibition if there [are] so many exceptions,” said Nathalie Smuha, an assistant professor and researcher in AI ethics at KU Leuven university.
Hollywood or China
A total ban on some AI uses wasn’t on the cards when the European Commission instructed its experts to prepare an AI strategy in 2018.
Experts and top officials quickly changed their minds, said Smuha, who coordinated the expert group’s work, as they realized that existing legislation might fall short of tackling certain practices.
The Commission’s first draft for the new AI rules in April 2021 already suggested prohibiting several practices, such as those that “distort a person’s behavior” via subliminal techniques or exploit vulnerable groups.
When the bill went to the European Parliament for discussion, lawmakers added to that list as they negotiated with government officials. The final shape of the law was agreed in December 2023.
Kim Van Sparrentak, a Dutch Greens lawmaker involved in the AI Act negotiations, said the list’s bans fell into three categories.
“Things that were already existing that we knew we didn’t want, [and] things that we could imagine either from Hollywood movies to some extent or from China,” she said.
One case that influenced the drafting was a furor in the Netherlands in 2019 over Dutch tax authorities using an algorithm to spot childcare benefits fraud, a move which led to some 26,000 people being wrongly accused of fraud.
Such a practice could now be banned, falling under predictive policing.
“If you want a society that is based on the rule of law, you can’t treat people constantly like they are already potential criminals,” Van Sparrentak said.
Similarly, the ban on scraping pictures of people’s faces from the Internet stems from U.S.-based Clearview AI, a facial recognition firm investigated for pulling billions of images into its database.
The AI Act forbids the use of “AI systems that create or expand facial recognition databases through the untargeted scraping of facial images from the Internet. ”
China may have inspired some other bans, such as a prohibition on social scoring, such as the Chinese system of ranking individuals based on their behavior.
Migration
Digital rights campaigners pointed to the law’s “various grave loopholes,” particularly for policing and migration, a group of 22 organizations wrote in an open letter in January.
“The most glaring loophole is the fact that the bans do not apply to law enforcement and migrational authorities,” said Caterina Rodelli, EU policy analyst at digital rights group Access Now.
An eye-catching exemption is the ban for real-time facial recognition in public places.
In principle, law enforcers won’t be able to do that anymore, but EU countries can still allow exceptions, especially for the most serious crimes.
AI also can’t be used to detect emotions at schools or offices, but that doesn’t apply apply for law enforcement and migration. Activists have warned this could allow the use of AI lie detectors, that scan a face for any signs of deceit, at the borders
Van Sparrentak said these sorts of exemptions were a red line for EU governments during negotiation, forcing last-ditch talks to stretch to 36 hours.
“They want to be able to use all the tools at their disposal,” she said.
