MIT Technology Review Subscribe

Five big takeaways from Europe’s AI Act

The AI Act vote passed with an overwhelming majority, but the final version is likely to look a bit different

This article is from The Technocrat, MIT Technology Review’s weekly tech policy newsletter about power, politics, and Silicon Valley. To receive it in your inbox every Friday, sign up here.

It was a big week in tech policy in Europe with the European Parliament’s vote to approve its draft rules for the AI Act on the same day EU lawmakers filed a new antitrust lawsuit against Google

Advertisement

The AI Act vote passed with an overwhelming majority, and has been heralded as one of the world’s most important developments in AI regulation. The European Parliament’s president, Roberta Metsola, described it as “legislation that will no doubt be setting the global standard for years to come.” 

This story is only available to subscribers.

Don’t settle for half the story.
Get paywall-free access to technology news for the here and now.

Subscribe now Already a subscriber? Sign in
You’ve read all your free stories.

MIT Technology Review provides an intelligent and independent filter for the flood of information about technology.

Subscribe now Already a subscriber? Sign in

Don’t hold your breath for any immediate clarity, though. The European system is a bit complicated. Next, members of the European Parliament will have to thrash out details with the Council of the European Union and the EU’s executive arm, the European Commission, before the draft rules become legislation. The final legislation will be a compromise between three different drafts from the three institutions, which vary a lot. It will likely take around two years before the laws are actually implemented.

What Wednesday’s vote accomplished was to approve the European Parliament’s position in the upcoming final negotiations. Structured similarly to the EU’s Digital Services Act, a legal framework for online platforms, the AI Act takes a “risk-based approach” by introducing restrictions based on how dangerous lawmakers predict an AI application could be. Businesses will also have to submit their own risk assessments about their use of AI. 

Some applications of AI will be banned entirely if lawmakers consider the risk “unacceptable,” while technologies deemed “high risk” will have new limitations on their use and requirements around transparency. 

Here are some of the major implications:

  1. Ban on emotion-recognition AI. The European Parliament’s draft text bans the use of AI that attempts to recognize people’s emotions in policing, schools, and workplaces. Makers of emotion-recognition software claim that AI is able to determine when a student is not understanding certain material, or when a driver of a car might be falling asleep. The use of AI to conduct facial detection and analysis has been criticized for inaccuracy and bias, but it has not been banned in the draft text from the other two institutions, suggesting there’s a political fight to come.
  2. Ban on real-time biometrics and predictive policing in public spaces. This will be a major legislative battle, because the various EU bodies will have to sort out whether, and how, the ban is enforced in law. Policing groups are not in favor of a ban on real-time biometric technologies, which they say are necessary for modern policing. Some countries, like France, are actually planning to increase their use of facial recognition
  3. Ban on social scoring. Social scoring by public agencies, or the practice of using data about people’s social behavior to make generalizations and profiles, would be outlawed. That said, the outlook on social scoring, commonly associated with China and other authoritarian governments, isn’t really as simple as it may seem. The practice of using social behavior data to evaluate people is common in doling out mortgages and setting insurance rates, as well as in hiring and advertising. 
  4. New restrictions for gen AI. This draft is the first to propose ways to regulate generative AI, and ban the use of any copyrighted material in the training set of large language models like OpenAI’s GPT-4. OpenAI has already come under the scrutiny of European lawmakers for concerns about data privacy and copyright. The draft bill also requires that AI generated content be labeled as such. That said, the European Parliament now has to sell its policy to the European Commission and individual countries, which are likely to face lobbying pressure from the tech industry.
  5. New restrictions on recommendation algorithms on social media. The new draft assigns recommender systems to a “high risk” category, which is an escalation from the other proposed bills. This means that if it passes, recommender systems on social media platforms will be subject to much more scrutiny about how they work, and tech companies could be more liable for the impact of user-generated content.

The risks of AI as described by Margrethe Vestager, executive vice president of the EU Commission, are widespread. She has emphasized concerns about the future of trust in information, vulnerability to social manipulation by bad actors, and mass surveillance. 

“If we end up in a situation where we believe nothing, then we have undermined our society completely,” Vestager told reporters on Wednesday.

What I am reading this week

What I learned this week

This week, Human Rights Watch released an in-depth report about an algorithm used to dole out welfare benefits in Jordan. The agency found some major issues with the algorithm, which was funded by the World Bank, and says the system was based on incorrect and oversimplified assumptions about poverty. The report’s authors also called out the lack of transparency and cautioned against similar projects run by the World Bank. I wrote a short story about the findings

Meanwhile, the trend toward using algorithms in government services is growing. Elizabeth Renieris, author of Beyond Data: Reclaiming Human Rights at the Dawn of the Metaverse, wrote to me about the report, and emphasized the impact these sort of systems will have going forward: “As the process to access benefits becomes digital by default, these benefits become even less likely to reach those who need them the most and only deepen the digital divide. This is a prime example of how expansive automation can directly and negatively impact people, and is the AI risk conversation that we should be focused on now.”

Advertisement
This is your last free story.
Sign in Subscribe now

Your daily newsletter about what’s up in emerging technology from MIT Technology Review.

Please, enter a valid email.
Privacy Policy
Submitting...
There was an error submitting the request.
Thanks for signing up!

Our most popular stories

Advertisement