The AI Act vote passed with an overwhelming majority, but the final version is likely to be a little different
Stephanie Arnett/MITTR | Envato, Midjourney
This article is from The Technocrat, MIT Technology Review’s weekly tech policy newsletter on power, politics, and Silicon Valley. To receive it in your inbox every Friday, sign her upe.
It’s been an important week for European tech policy as the European Parliament voted to approve draft rules for the AI law on the same day that EU lawmakers filed a new antitrust lawsuit against Google.
The AI Act vote passed overwhelmingly and was hailed as one of the world’s most significant developments in AI regulation. European Parliament President Roberta Mezzola said it was “a law that will undoubtedly set global standards for many years to come”.
Don’t settle for half the story.
Get access to the latest tech news without a paywall.
subscribe now
Already a Subscriber? Login
MIT Technology Review provides an intelligent, independent filter for large amounts of information about technology.
subscribe now
Already a Subscriber? Login
But don’t hold your breath for immediate clarity. The European system is a little more complicated. Next, members of the European Parliament will have to work out the details with the European Union Council and the EU’s executive body, the European Commission, before the draft regulation becomes law. The final bill is a compromise of three different drafts from three agencies, with very different content. It is expected to take about two years before it is actually implemented.
What Wednesday’s vote achieved was to approve the position of the European Parliament in the upcoming final negotiations. The AI law, which has a similar structure to the EU’s Digital Services Act, the legal framework for online platforms, introduces restrictions based on how dangerous parliamentarians expect her AI applications to be. We take a risk-based approach. Businesses will also need to submit their own risk assessment of their use of AI.
Some applications of AI will be banned entirely if lawmakers deem the risks “unacceptable,” while technologies deemed “high-risk” will have new restrictions on their use and transparency. There will be gender requirements.
Some of the main impacts are listed below.
- Banning Emotion Recognition AI. A European Parliament draft bans the use of AI to try to recognize people’s emotions in police, schools and workplaces. Emotion-recognition software makers claim their AI can tell when a student doesn’t understand certain content, or when a car driver might be dozing off. The use of AI to do face detection and analysis has been criticized for inaccuracy and bias, but drafts from his two other agencies aren’t banning it, suggesting political strife will ensue. are doing.
- Prohibit real-time biometrics and predictive policing in public spaces. This will be a major legislative battle as various EU institutions will have to consider whether and how the ban will be enforced in law. Law enforcement groups are not in favor of banning the real-time biometric technology needed for modern policing. Some countries, such as France, are actually planning to increase the use of facial recognition.
- No Social Scoring. Social scoring by public agencies and the use of data about people’s social behavior to create generalizations and profiles is illegal. That said, the social score outlook often associated with China and other authoritarian governments is actually not as simple as it sounds. The practice of using social behavioral data to evaluate people is common in mortgage lending and insurance rate setting, as well as hiring and advertising.
- New limits for Gen AI. This draft is the first to propose a way to regulate generative AI and prohibit the use of copyrighted material in the training sets of large language models such as OpenAI’s GPT-4. OpenAI is already under scrutiny by European lawmakers over data privacy and copyright concerns. The bill also requires AI-generated content to be labeled as such. That said, the European Parliament now needs to pitch its policies to countries likely to face lobbying pressure from the European Commission and the tech industry.
- New restrictions on recommendation algorithms on social media. The new draft assigns the recommendation system to the “high risk” category, which is escalated further than other bills. This means that if the bill is passed, how recommender systems on social media platforms work will be subject to greater scrutiny, and tech companies could be held more accountable for the impact of user-generated content. It means that there is a gender.
The risks of AI, as described by Margrethe Vestager, Vice-President of the EU Commission, are far-reaching. She highlighted concerns about the future of trust in information, vulnerability to social manipulation by bad guys, and mass surveillance.
“If we find ourselves in a situation where we can’t believe anything, our society will be completely undermined,” Vestager told reporters on Wednesday.
what i’m reading this week
- Russian soldiers surrendered to Ukrainian military drones, according to video footage released by The Wall Street Journal. The surrender took place in May in the eastern Ukraine city of Bakhmut. The drone operator saw the soldier’s plea on video and decided to save his life in accordance with international law. Drones are extremely important in warfare, and surrendering presents an interesting perspective on the future of warfare.
- Many Redditors are protesting changes to the site’s API that remove or reduce functionality for third-party apps and tools used by many communities. As a protest, these communities were “privatized”. This means the page is no longer publicly accessible. Reddit is known for empowering its user base, but Casey Newton’s astute assessment suggests the company may regret doing so now.
- The contractors who trained Google’s large-scale language model, Bard, say they were fired after expressing concerns about working conditions and the safety of the AI itself. Contractors say they were forced to meet unreasonable deadlines, raising accuracy concerns. Google claims responsibility lies with Appen, the contract agency that hires the workers. If history tells us, the race to dominate generative AI will come at a human cost.
what i learned this week
This week, Human Rights Watch released a detailed report on algorithms used to distribute welfare benefits in Jordan. The agency found some serious flaws in the World Bank-funded algorithm and said the system was based on false and oversimplified assumptions about poverty. The report’s authors also condemned the lack of transparency and cautioned against similar projects undertaken by the World Bank. I wrote a short story about the discovery.
On the one hand, there is a growing trend to use algorithms in government services.Elizabeth Reniris, author Beyond Data: Restoring Human Rights at the Dawn of the Metaversewrote to me about this report, highlighting the future impact of this kind of system: “As the process of accessing perks becomes digital by default, the people who need these perks the most will It’s even less likely to reach the ‘digital divide’. This is a prime example of how widespread automation can directly and adversely affect people, and is the AI risk discussion we should be focusing on now. ”
