Europe will be the first world power to have its own law on Artificial Intelligence. As a result of the European Commission's proposal on AI, the European Parliament has decided to pick up the baton and approve its position on the 'Report on Artificial Intelligence in the Digital Age', which will mark the main lines that this pioneering legislation will have.
Europe agrees on how AI should be regulated. It has not been easy. It has taken 18 months of work and dozens of seminars and conferences to agree on the MEPs from the different political groups. And there are still nuances, such as the regulation of facial recognition, where voting has been adjusted. The reason is none other than the complexity of the Artificial Intelligence technology itself.
Europe wants on the one hand to boost the AI-based industry, but on the other hand they are aware of the enormous risks and challenges that it entails. Achieving a balanced agreement, capable of collecting these two points that seem almost antagonistic, has been one of the main obstacles.
In the absence of large companies, Europe takes the global lead in the legal field. "Artificial Intelligence has strategic relevance. With this new legislation, Europe is a pioneer and sets the standard for global standards", explained Axel Voss, German rapporteur and responsible for carrying out this initiative.
From Europe they are aware that a lot of investment is needed to reach the level of the US or China in AI, but they are proud to lay concrete foundations to try to ensure that the AI systems that exist in Europe are under certain rules. And also in the hope that, thanks to this legal framework, many companies feel comfortable developing new AIs.
Up to 20,000 million euros of investment per year. As explained by Margrethe Vestager, Commissioner for Competition, the goal is to spend around 20,000 million euros of annual public-private investment. Currently between 12 and 14,000 million euros are invested. And if we only take into account public money, some 1,000 million euros are invested in Europe, far from the 5,100 million in the US or the 6,800 million euros in China.
A human behind the most susceptible AI systems. "AI should not scare us, but it should revolve around the human being," defends Axel Voss. The agreed legislation will have specific mechanisms for human supervision in cases of risk. In other words, in particularly sensitive sectors such as health, education or mobility, where the biases of the algorithm could lead to discrimination, it will be a human who can intervene in the system to alleviate these biases in decision-making that may affect the population. in a very important way.
Take the case of diagnostics through AI, public services, recruitment or immigration. In these areas that the different institutions of the European Union consider "high risk", several obligations will be established, such as risk analysis, traceability of results, detailed documentation, human supervision and a high level of robustness.
The IAs will be differentiated according to risk. The legislation proposed by the Commission and on which Parliament's report is based will differentiate AIs not according to their sector or the technology used, but according to the risk that they may affect fundamental rights, such as non-discrimination or privacy.
Lilian Edwards, Professor of Law and Innovation at Newcastle University.
The first level will be the unacceptable risk, those that pose a threat to security. These should be prohibited under the proposal. This includes everything from systems that encourage dangerous behavior in minors to "social scoring" systems by governments. There has been debate over whether biometric recognition systems will be included here, but the final vote is not to include a blanket ban.
There will be no major bans (including facial recognition). Except for the most extreme cases, the European Parliament has decided to opt for more pragmatic legislation. With the idea of continuing to give air to this technology, there will be no major prohibitions. Europe has wanted to avoid sending a negative message.
The case of facial recognition was one of the most debated and the amendments related to this matter have received the most votes, although none of them have been implemented. The Commission established a special category for it, its use being prohibited but only in public areas and live. However, exceptions were added such as in the cases of searching for a missing child, preventing a specific and imminent terrorist threat or to "detect, locate, identify or prosecute a perpetrator or suspect of a serious crime".
„We do not suggest to regulate the technology but to regulate the use of the technology“ – very important point by @vestager that could not be stressed enough. Straight out bans do not follow this logic 😉
– Axel Voss MdEP May 3, 2022
Miapetra Kumpula-Natri, Socialist MEP, explained that "we cannot focus on specific technologies. We must find a political balance point and avoid future frustration when seeing AI evolve." This combination of a driving vision and a moderating vision is what has allowed the European Parliament to come to an agreement.
The lack of control over weaponized AIs generates votes against. The group The Left has decided to vote against. Barrena Arza, a Basque MEP, explained that the agreement "lacks points such as a serious risk analysis; a real collaboration at a multinational level and neither facial recognition nor AI weapons have been prohibited."
Europe seeks to create an equivalent to the GDPR but for AI. The General Data Protection Regulation is considered a benchmark in the European institutions. While the rules for working with our data were created then, now we are looking at how to build a regulation with a view to 2030 and that establishes the rules on the impact of algorithms, what limitations they will have and who should be held responsible.
According to the Commission, companies or providers of these AI systems that do not comply with the legislation will be exposed to fines of up to 6% of their annual global income, the same percentage as the GDPR and slightly higher than the 4% initially proposed. .
The final legal text still has a way to go. If the European institutions have anything, it is that they must go through several processes and debates before reaching the final implementation. With the legislation of Artificial Intelligence we are still in the first phase.
The Commission was the first to propose regulating AI, but the European Parliament is the only body where the representatives have been voted on by the citizenry. And therefore it must be there where the initiatives to be taken are approved. After studying the report for a year and a half, Parliament has approved today that Artificial Intelligence deserves its own law and this must be in line with European values. Few countries, except Finland or Spain, have government initiatives around AI. But they do not fall within the legal framework.
The next step. The European Commission will now propose specific legislation, the AI Act, which should occur towards the end of September 2022. A proposal that will return to Parliament and that should be definitively approved by the beginning of 2023. After a short period, usually 6 months, it should be implemented without the need to go through transpositions in each country. It will be then when the European Union can boast of having the first legislation on Artificial Intelligence in the world.