Europe’s AI startup founders shouldn’t “overreact” in response to the legislation being prepared by the EU to regulate the technology, Spain’s AI and digitalisation minister tells Sifted.
The AI Act draft, which is still undergoing tough negotiations, would place a number of obligations on developers of high-risk AI systems, including an assessment on whether they would undermine fundamental rights.
The regulation would class all other AI systems as either “low-risk” or “unacceptable risk” — with the latter being applied to use cases that will be outright banned, such as those using manipulative or deceptive techniques to distort someone’s behaviour such as deepfakes, and facial recognition technology in public places.
Founders have told Sifted that they fear the EU rules could put Europe even further behind the US and China in the AI race by stifling innovation. Cedric O, the former French government minister who’s been leading generative AI startup Mistral’s public affairs efforts, has even warned that the EU AI Act could “kill” the Paris-based company.
But in an interview with Sifted, Carme Artigas says “nobody should overreact” against the draft legislation before it is agreed, arguing Europe can be “competitive and innovative” and still protect people’s rights and freedoms. The EU act will give companies two years to adapt, and help them through a network of national sandbox initiatives in the meantime, she adds.
“Founders should be very calm, because the big goal of this legislation is not to stifle innovation but, on the contrary, to introduce many measures to boost it, including sandboxes all over the EU,” she says. “We haven’t improvised [the rules] six months ago, we’ve been thinking about this for three years. I can assure you that there’s no dimension we have not considered.”
Spain has played an important role in shaping the legislation because it is hosting the rotating presidency of the EU until the end of the year.
The fifth round of trilogue negotiations among the three EU law-making institutions — Parliament, Commission and Council — concluded last week and were “very positive,” Artigas says, because all three bodies reached a consensus on what will be considered a “high-risk” AI system.
Unlike with most other pieces of legislation, the EU is this time negotiating all the articles of the AI Act simultaneously, in a massive exercise of give and take among the three EU bodies. But the minister is confident a final text can be agreed by the end of this year, perhaps after one or two more rounds of talks in November and December.
“I am convinced that this legislation will become a reference at the international level, as it was the case with the EU’s GDPR,” she says.
The biggest stumbling blocks at the moment remains how to decide which systems will fall in each category and whether consensus can be built between Parliament and Council, which have requested that some AI applications such as facial recognition technology may be used for some law-enforcement purposes — outraging parliament members.
Artigas adds there are still many proposals on the table on how to regulate foundational models. Negotiators are yet to discuss whether there should be a two-tiered model for companies developing foundational models, which would see those working on more advanced systems being subject to more requirements than smaller companies, like startups.
A key test for the EU AI Act will be keeping up-to-date with the fast progress of AI. “We are trying to come up with a law that doesn’t become obsolete in six months,” she says. “We are trying to add mechanisms allowing the Commission to update key parts, like those on high-risk models, through executive orders, for instance.”
While wrangling on the AI Act continues, Spain has launched the Spanish Agency for the Supervision of Artificial Intelligence (AESIA) — the first AI regulatory body within the EU.
The agency will be tasked with implementing the EU rules and running Spain’s AI sandbox, aimed at helping AI companies adapt to the new requirements. The agency will also create a “voluntary stamp of quality” to certify AI algorithms. “It’s a unique experience, nobody has ever done it before,” says Artigas.
AESIA is recruiting members for its council, whose names will be announced by mid-November, Artigas says. The appointment of the agency’s director, however, will have to wait until a new Spanish government is formed.
Artigas has also been chosen to represent Spain in a new UN AI body to build a global scientific consensus on risks and challenges. Officials and experts from Albania, Estonia, Germany, France, Switzerland and the UK will also sit at the table.
The move is likely to be seen as a response to AI discussions in the G7 group of the seven most industrialised nations and Britain’s AI Safety summit, taking place on November 1-2, which will be attended by representatives from about 25 countries invited by the UK government.
Ahead of this week’s gathering, the British prime minister Rishi Sunak confirmed plans to create a UK-based AI Safety Institute, to evaluate new types of AI to help the world understand their risks. He also intends to pitch fellow leaders a new AI research network formed by international experts and modelled on the Intergovernmental Panel on Climate Change, tasked with publishing an annual state of AI science report.
Artigas, who will attend the UK summit, says the AI Safety Institute is “complementary” to other initiatives developed elsewhere, and that instead of creating “parallel fora”, all these various bodies should “converge into a global governance mechanism” — with more details on its exact shape potentially emerging by the end of this year.
“I think the right place to lead that international discussion is the United Nations,” she says. “We need to sum [our] efforts, we don’t need to reinvent the wheel. The problems are identified.”
Scaleup Act not needed
Artigas says Spain aims to become “the friendliest country” for startups with its brand new Startup Act, which introduced a number of incentives and facilitations for the creation of new tech companies, including a certification system for startups, a visa for so-called digital nomads, tax benefits for VC firms and an extension of the residence permits for entrepreneurs from one to three years.
But only 10 months after the act came into force, the Spanish tech ecosystem is pushing for a second piece of legislation tailored to scaleups’ needs. Artigas, however, says there’s no need for more laws and that the impact of the new incentives will be felt within one or two years.
“I think laws exist to remove barriers to entry, not to shape the dynamic of the market,” the minister says. “Scaleups’ great limitation is access to financing, and we solved that over a year ago with the Next Tech public-private fund.
“We should ask ourselves why a large company dedicated to making software should have more advantages than a winery that has digitalised all its production and is assisted by a generative AI. All sectors are going to be digitalised, are going to use AI and it is going to be very difficult to segment.”
The Next Tech fund, set up as a fund of funds in July 2021, aims to fuel the emergence of tech unicorns in Spain and has received €2bn of Spanish taxpayers’ money. Its investment ticket is set at €3m, but Artigas says the government could soon lower this to just €1m to benefit generative AI startups and others in the deeptech sector.
VCs in Spain were traditionally too risk-averse to invest in tech startups, but that’s changing in parallel with founders’ mentality, she argues, because they’re realising that it is not necessary to leave Europe to access the capital they need. “We are changing the structural fabric of the country,” she says.