The European Union is set to create the world鈥檚 first broad standards for regulating artificial intelligence. As well as determining how the technology affects the lives of almost 450 million citizens in the 27 countries of the EU, the rules are likely to influence how AI is used elsewhere in the world.
鈥淭he idea is that you have a harmonised system, which is really good,鈥� says at the University of Oxford.
The EU is working towards finalising the text of the in 2023. Its focus is on AI systems that could detrimentally affect citizens. Some uses, such as government-run social scoring, could be banned. Others would be labelled as high risk and strictly regulated. This is likely to include systems involved in accessing essential services, such as loans or benefits, judicial systems and workplace hiring decisions. But most AI uses, such as chatbots that assist with orders for pizza in phone apps, aren鈥檛 expected to be categorised as high risk.
Advertisement
The act would require all AI systems to notify people if they are interacting with a computer, not a person. 鈥淭his could change how platforms like Facebook and TikTok look, since their content-recommendation algorithms should fall under this requirement,鈥� says at the University of Bologna in Italy.
Although many policy experts and tech giants have welcomed the idea of regulation, there is no widespread agreement on how some applications of AI, such as facial recognition in public spaces, should be categorised.
Coming up with definitions that cover AI uses in many areas will also be tough. 鈥淭hat creates a really difficult scenario where you鈥檙e trying to include fairly simple algorithms that are doing really important stuff, and exclude fairly technically complicated algorithms that are doing trivial stuff,鈥� says at the Brookings Institution in Washington DC.
Another outstanding issue is that law-makers must clarify how the act will be enforced, says at the European Center for Not-for-Profit Law in the Netherlands, who wants to see the final regulations adopt a 鈥渞obust and independent EU AI board with concrete monitoring powers鈥�.
AIs used for national security or military purposes are likely to be excluded from the rules, says Skoric, despite an opinion poll suggesting that .
Whatever the final AI Act looks like, compliance will be mandatory for any company delivering AI-powered services to people in the EU. Some platforms with an international customer base may adopt the AI Act standards for all users, whereas those with more isolated and localised algorithms may not, says Engler.
The AI Act could lead to similar regulations worldwide, says Hine. Countries such as Brazil and Canada have already adopted parts of the draft act in their own proposals. 鈥淎lmost every part of the world is agreeing that we need to do something,鈥� says Wachter.
