Recent developments and the deployment of new artificial intelligence (AI) systems – such as ChatGPT – have prompted MEPs working on the future AI Act to reconsider and postpone the vote on the report in the European Parliament committee, scheduled for 26 April (see EUROPE 13155/5).
Discussions are continuing and the political groups aim to reach an internal agreement by 11 May, when the report by Ioan-Dragoş Tudorache (Renew Europe, Romanian) and Brando Benifei (S&D, Italian) could be voted on in the committee of the European Parliament.
The text currently being discussed by MEPs responsible for artificial intelligence legislation provides for a distinction to be made between general-purpose AI systems and basic models, which are generative technologies, based on data from across the internet, and include systems such as ChatGPT.
These basic models would be subject to additional requirements, in particular before they are placed on the market, in order to limit potential risks to sensitive sectors and areas, such as health, the environment or fundamental rights and the Rule of law.
In addition to consistent high levels of performance, safety and the possibility of corrective action throughout the lifecycle of the AI systems concerned, the text provides that the providers of these basic AI models shall make “relevant” data and documents available for 10 years after placing them on the market.
Users should also be able to access certain information through an obligation on providers to make publicly available a summary of the data used to train the AI system.
The EU Council for Artificial Intelligence - whose precise role and remit is still being discussed in the European Parliament - could, in this context, be the interlocutor of the suppliers of AI systems described as ‘basic models’. The exchanges would include, as well, the compliance of the AI systems concerned. (Original version in French by Thomas Mangin)