It’s done. It’s over. Two and a half years after it was first introduced—after months of lobbying and political arm-wrestling, plus grueling final negotiations that took nearly 40 hours—EU lawmakers have reached a deal over the AI Act. It would be the world’s first sweeping AI law.
The AI Act was conceived as a landmark bill that might mitigate harm in areas where using AI poses the most important risk to fundamental rights, comparable to health care, education, border surveillance, and public services, in addition to banning uses that pose an “unacceptable risk.”
“High risk” AI systems can have to stick to strict rules that require risk-mitigation systems, high-quality data sets, higher documentation, and human oversight, for instance. The overwhelming majority of AI uses, comparable to recommender systems and spam filters, will get a free pass.
The AI Act is a significant deal in that it can introduce vital rules and enforcement mechanisms to a hugely influential sector that’s currently a Wild West.
Listed below are MIT Technology Review’s key takeaways:
1. The AI Act ushers in vital, binding rules on transparency and ethics
Tech firms like to speak about how committed they’re to AI ethics. But in the case of concrete measures, the conversation dries up. And anyway, actions speak louder than words. Responsible AI teams are sometimes the primary to see cuts during layoffs, and in reality, tech firms can determine to vary their AI ethics policies at any time. OpenAI, for instance, began off as an “open” AI research lab before closing up public access to its research to guard its competitive advantage, similar to every other AI startup.
The AI Act will change that. The regulation imposes legally binding rules requiring tech firms to notify people once they are interacting with a chatbot or with biometric categorization or emotion recognition systems. It’ll also require them to label deepfakes and AI-generated content, and design systems in such a way that AI-generated media will be detected. This can be a step beyond the voluntary commitments that leading AI firms made to the White House to easily AI provenance tools, comparable to watermarking.
The bill can even require all organizations that provide essential services, comparable to insurance and banking, to conduct an impact assessment on how using AI systems will affect people’s fundamental rights.
2. AI firms still have plenty of wiggle room
When the AI Act was first introduced, in 2021, people were still talking concerning the metaverse. (Are you able to imagine!)
Fast-forward to now, and in a post-ChatGPT world, lawmakers felt that they had to take so-called foundation models—powerful AI models that will be used for many various purposes—under consideration within the regulation. This sparked intense debate over what kinds of models must be regulated, and whether regulation would kill innovation.
The AI Act would require foundation models and AI systems built on top of them to attract up higher documentation, comply with EU copyright law, and share more details about what data the model was trained on. For probably the most powerful models, there are extra requirements. Tech firms can have to share how secure and energy efficient their AI models are, for instance.
But here’s the catch: The compromise lawmakers found was to use a stricter algorithm only probably the most powerful AI models, as categorized by the computing power needed to coach them. And it can be as much as firms to evaluate whether or not they fall under stricter rules.
A European Commission official wouldn’t confirm whether the present cutoff would capture powerful models comparable to OpenAI’s GPT-4 or Google’s Gemini, because only the businesses themselves know the way much computing power was used to coach their models. The official did say that because the technology develops, the EU could change the way in which it measures how powerful AI models are.
3. The EU will change into the world’s premier AI police
The AI Act will arrange a brand new European AI Office to coordinate compliance, implementation, and enforcement. It would be the primary body globally to implement binding rules on AI, and the EU hopes this can help it change into the world’s go-to tech regulator. The AI Act’s governance mechanism also features a scientific panel of independent experts to supply guidance on the systemic risks AI poses, and easy methods to classify and test models.
The fines for noncompliance are steep: from 1.5% to 7% of a firm’s global sales turnover, depending on the severity of the offense and size of the corporate.
Europe can even change into the certainly one of the primary places on the earth where residents will give you the chance to launch complaints about AI systems and receive explanations about how AI systems got here to the conclusions that affect them.
By becoming the primary to formalize rules around AI, the EU retains its first-mover advantage. Very similar to the GDPR, the AI Act could change into a worldwide standard. Firms elsewhere that need to do business on the earth’s second-largest economy can have to comply with the law. The EU’s rules also go a step further than ones introduced by the US, comparable to the White House executive order, because they’re binding.
4. National security at all times wins
Some AI uses at the moment are completely banned within the EU: biometric categorization systems that use sensitive characteristics; untargeted scraping of facial images from the web or CCTV footage to create facial recognition databases like Clearview AI; emotion recognition at work or in schools; social scoring; AI systems that manipulate human behavior; and AI that’s used to take advantage of people’s vulnerabilities.
Predictive policing can be banned, unless it’s used with “clear human assessment and objective facts, which mainly don’t simply leave the choice of going after a certain individual in a criminal investigation only because an algorithm says so,” based on an EU Commission official.
Nonetheless, the AI Act doesn’t apply to AI systems which were developed exclusively for military and defense uses.
One in all the bloodiest fights over the AI Act has at all times been easy methods to regulate police use of biometric systems in public places, which many fear could lead on to mass surveillance. While the European Parliament pushed for a near-total ban on the technology, some EU countries, comparable to France, have resisted this fiercely. They need to use it to fight crime and terrorism.
European police forces will only give you the chance to make use of biometric identification systems in public places in the event that they get court approval first, and just for 16 different specific crimes, comparable to terrorism, human trafficking, sexual exploitation of youngsters, and drug trafficking. Law enforcement authorities can also use high-risk AI systems that don’t pass European standards in “exceptional circumstances regarding public security.”
5. What next?
It would take weeks and even months before we see the ultimate wording of the bill. The text still must undergo technical tinkering, and needs to be approved by European countries and the EU Parliament before it officially enters into law.
Once it’s in force, tech firms have two years to implement the principles. The bans on AI uses will apply after six months, and corporations developing foundation models can have to comply with the law inside one 12 months.