In two years’ time, if everything goes to plan, EU residents will be protected by law from some of the most controversial uses of AI, such as street cameras that identify and track people, or government computers that score an individual’s behaviour.
This week, Brussels laid out its plans to become the first global bloc with rules for how artificial intelligence can be used, in an attempt to put European values at the heart of the fast-developing technology.
Over the past decade, AI has become a strategic priority for countries around the world, and the two global leaders — the US and China — have taken very different approaches.
China’s state-led plan has led it to investing heavily in the technology, and quickly roll out applications that have helped the government boost surveillance and control the population. In the US, AI development has been left to the private sector, which has focused on commercial applications.
“The US and China have been the ones that have been innovators, and leading in investment into AI,” said Anu Bradford, EU law professor at Columbia University.
“But this regulation seeks to put the EU back in the game. It is trying to balance the idea that the EU needs to become more of a technological superpower and get itself in the game with China and the US, without compromising its European values or fundamental rights.”
EU officials hope that the rest of the world will follow its lead, and claim that Japan and Canada are already taking a close look at the proposals.
While the EU wishes to rein in the way that governments can wield AI, it also wants to encourage start-ups to experiment and innovate.
Officials said they hoped the clarity of the new framework would help give confidence to these start-ups. “We will be the first continent where we will give guidelines. So now if you want to use AI applications, go to Europe. You will know what to do and how to do it,” said Thierry Breton, the French commissioner in charge of digital policy for the bloc.
In an attempt at being pro-innovation, the proposals acknowledge that regulation often falls hardest on smaller businesses, and so incorporate measures to help. These include “sandboxes” where start-ups can use data to test new programmes to improve the justice system, healthcare and the environment without fear of being hit with heavy fines if mistakes are made.
Alongside the regulation, the commission published a detailed road map for increasing investment in the sector, and pooling public data across the bloc to help train machine-learning algorithms.
The proposals are likely to be fiercely debated by both the European Parliament and member states — two groups that will need to sanction the draft into law. The legislation is expected by 2023 at the earliest, according to people following the process closely.
But critics say that, in trying to support commercial AI, the draft legislation does not go far enough in banning discriminatory applications of AI-like predictive policing, migration control at borders and the biometric categorisation of race, gender and sexuality. These are currently marked as “high-risk” applications, which means anyone deploying them will have to notify people on whom they are being used, and provide transparency on how the algorithms made their decisions — but their widespread use will still be allowed, particularly by private companies.
Other applications that are high-risk, but not banned, include the use of AI in recruitment and worker management, as currently practised by companies including HireVue and Uber, AI that assesses and monitors students, and the use of AI in granting and revoking public assistance benefits and services.
Access Now, a Brussels-based digital rights group, also pointed out that outright bans on both live facial recognition and credit scoring only address public authorities, without affecting companies such as the facial recognition firm Clearview AI or AI credit scoring start-ups such as Lenddo and ZestFinance, whose products are available globally.
Others highlighted the conspicuous absence of citizens’ rights in the legislation. “The entire proposal governs the relationship between providers (those developing [AI technologies]) and users (those deploying). Where do people come in?” wrote Sarah Chander and Ella Jakubowski from European Digital Rights, an advocacy group, on Twitter. “Seems to be very few mechanisms by which those directly affected or harmed by AI systems can claim redress. This is a huge miss for civil society, discriminated groups, consumers and workers.”
On the other hand, lobby groups representing the interests of Big Tech also criticised the proposals, saying they would stifle innovation.
The Center for Data Innovation, a think-tank part whose parent organisation receives funding from Apple and Amazon, said the draft legislation struck a “damaging blow” to the EU’s plans to be a global leader in AI and that “a thicket of new rules will hamstring technology companies” hoping to innovate.
In particular, it took issue with the ban on AI that “manipulates” people’s behaviours and with the regulatory burden for “high-risk” AI systems, such as compulsory human oversight, and proof of safety and efficacy.
Despite these criticisms, the EU is concerned that if it does not act now to set rules around AI, it will allow the global rise of technologies that are contrary to European values.
“The Chinese have been very active in applications that give concern to Europeans. These are being actively exported, especially for law enforcement purposes and there is a lot of demand for that among illiberal governments,” Bradford said. “The EU is very concerned that it needs to do its part to halt the global adoption of these deployments that compromise fundamental rights, so there is definitely a race for values.”
Petra Molnar, associate director at York University in Canada, agreed, saying the draft legislation has more depth and focuses more on human values than early proposals in the US and Canada.
“There is a lot of hand waving around ethics and AI in the US and Canada but [proposals] are more shallow.”
Ultimately, the EU is betting on the fact that development and commercialisation of AI will be driven by public trust.
“If we can have a better regulated AI that consumers trust, that also creates a market opportunity, because . . . it will be a source of competitive advantage for European systems [as] they are considered trustworthy and high quality,” said Bradford of Columbia University. “You don’t only compete with price.”