2023-05-11 09:33:22
As usual, the European Union (EU) is seeking to carve out a “third way” between China and the United States. These days, it is the vast field of artificial intelligence (AI) that the European institutions intend to regulate, avoiding both Chinese-style authoritarianism and American ultra-liberalism. “The EU is going to be the first jurisdiction in the world to legislate on AI and it can’t afford to do it wrong,” warns Ursula Pachl, Deputy Director General of the European Consumers’ Bureau (BEUC).
Watched very closely by tech lobbies but also by associations for the defense of freedoms and human rights, the draft regulations currently being drawn up in Brussels, the “AI Act” in English, will have given a thread to trick the negotiators. Fifty meetings, hundreds of amendments and several postponements later, the European Parliament must adopt its version of the text in committee, Thursday, May 11.
It should be voted on in plenary session by the summer, before the start of final negotiations (trialogue phase) between Parliament, the European Commission and the Council of the EU, for final adoption before the end of the year. Two and a half years after the initial Commission proposal, in April 2021.
Four levels of risk
The particularity of the AI Act is to regulate artificial intelligence according to its uses, according to four levels of risk. This ranges from unrestricted permission (video games, for example) to outright prohibition (Chinese-style social rating). Between these two poles, two intermediate levels – “high risk” et “moderate risk” – will require various transparency and security obligations from designers.
Compared to the compromise text of the Member States (gathered within the Council of the EU) of December 2022, the Parliament has extended the list of prohibited systems. This now includes the “emotion recognition” used for law enforcement or border control, as well as software for “predictive policing”.
Facial recognition, a real catalyst for political tensions, has also joined this list: MEPs want these biometric identification systems not only to be banned in real time, but also to resolve investigations a posteriori, unless a judge gives his authorisation.
A Divided Parliament
That being said, in parallel with the extension of this official “black list”, the right wing of Parliament obtained the introduction of a filter which should limit the real scope of these obligations. To be truly considered as being at “high risk” (not prohibited, but very supervised), these systems will not only have to appear in the corresponding list: it will be necessary to prove that they present a “significant risk” for health, safety or fundamental rights.
“With this new filter, the Parliament adopts a hybrid position”, analyzes Clément Pérarnaud, associate researcher in political science at the Vrije Universiteit Brussel (VUB). The two co-rapporteurs having had the greatest difficulty in finding a majority, this text is the result of a political compromise. “However, faced with a divided Parliament, the Council of the EU seems much more united: the Member States are above all concerned not to slow down innovation and to allow secure applications of AI, continues the specialist. It is likely that, during the trilogue phase, the Council’s positions will prevail over the fragile compromise found in Parliament. »
Lobbying
The companies concerned, for their part, criticize regulations likely to ” to choke ” a sector of the future. “The AI Act holds the developers of AI systems responsible for the use that will be made of their tools: but how could they anticipate it? It is the user who should be made responsible”, believes Thibault de Tersant, vice-president of the digital professional federation Numeum.
In Brussels, the lobbying of tech companies now far exceeds that of the pharmaceutical, financial and chemical sectors. Its budget was 97 million euros in 2021: ten times more than the powerful automobile lobby.
——
ChatGPT, a risky technology?
“General-purpose” AI systems were included in the AI Act last year, at the instigation of France, President of the Council of the EU in early 2022. Among them, text generators and ‘image, which can be used harmlessly or maliciously (disinformation, etc.). ChatGPT, Midjourney, but also their future successors are therefore concerned. The text which will be voted on in committee on Thursday, May 11 in the European Parliament, plans in particular to oblige the designers of this software to make public the data with which they have been trained and which are protected by copyright. This will allow rights holders to act if necessary.
#Brussels #intense #negotiations #regulate