On Thursday 13 July, MEPs debated with the European Commission on the global convergence of generative artificial intelligence (AI) systems. These discussions come at a time when the co-legislators began inter-institutional negotiations (trilogues) on 14 June – a second trilogue is scheduled for 18 July (see EUROPE 13219/12) – and when the Commission is aiming to set up a ‘global deal for AI’ (see EUROPE 13194/3).
Some of the MEPs who spoke in the Chamber began by stressing the importance of guaranteeing the European Parliament’s involvement in the process of drawing up global standards for AI.
“We must reiterate our request within the EU-US Trade and Technology Council. The parliamentary dimension must be taken into account. We cannot accept decisions being taken behind closed doors. You must react as quickly as possible”, insisted Brando Benifei (S&D, Italian), who is also co-rapporteur on AI legislation.
The European Commission plans to base its ‘AI deal’ on a voluntary basis. This future deal was presented by Internal Market Commissioner Thierry Breton on 5 June as “the antechamber to the ‘AI Act’ “, designed to enable companies to prepare before the legislation comes into force. However, some MEPs, such as Kim van Sparrentak (Greens/EFA, Dutch), have voiced their concerns about the approach the Commission has opted for.
“Voluntary international efforts are essential, but by their very nature they are not very transparent. They must not be used to weaken legislation. Can you promise us that the negotiations on AI will be linked to obligations and that we will not be told that there are already voluntary agreements at international level on generative AI?”, she said, addressing the European Commissioner for Values and Transparency, Věra Jourová.
On this point, the Commission has given assurances that the work being carried out will not interfere with the work of the co-legislators. “The deal does not prejudge the work of the co-legislators. The deal will have to comply once the AI Act comes into force. The text will have to be adapted according to the text that will be adopted at the end of the procedure”, declared Ms Jourová, insisting at the same time on the need to avoid legal loopholes.
The OECD as a potential working base
While the European Parliament, in the position it adopted on 14 June (see EUROPE 13201/1), adopted the OECD definition, some MEPs felt that the Commission should follow suit in its work with third countries. “There is already international convergence and this is an excellent basis for continuing the debate” said Kosma Złotowski (ECR, Polish).
The Polish MEP also called for “not creating too many administrative obstacles”. The importance of providing for labelling of AI-generated content and provisions to protect intellectual property also came up several times during the discussions. For others, finally, like the co-rapporteur of the AI legislation, Dragoș Tudorache (Renew Europe, Romanian), it would also be “necessary to consider authoritarian uses of AI”.
The aim is to reach a political agreement with the Council of the EU before the end of the parliamentary term on the AI Act. The inter-institutional negotiations will take place against a backdrop of rapid and constant developments in artificial intelligence. The emergence of systems such as ChatGPT even prompted MEPs to reconsider (see EUROPE 13167/6). More and more players are embarking on the development of generative AI systems.
Within hours of the MEPs’ debate in Strasbourg, digital giant Google announced the launch of its own generative AI system, Bard. (Original version in French by Thomas Mangin)