Where is the battle between the major manufacturers of artificial intelligence (AI) models and content rights holders on the regulatory front? On Wednesday June 19, the first meeting of the AI Committee (AI Board) created by the European AI Act regulation took place in Brussels to involve the member states of the European Union (EU) in the implementation of this text. which they approved in early February.
But, for the moment, discussions on the application of the most contested measure during the negotiation of this text – transparency on the content used to train AI models – are still far from being concluded. And the standoff between AI manufacturers and content creators continues to play out on several fronts.
The description of the type of information that the manufacturer will have to provide on the texts or images used to train its models is not expected before the “spring 2025”for implementation of the measure expected in July or August, specifies a European Commission official. “Reflection is underway”we explain while noting that this point is not at this stage the “priority number one” : the Commission has just officially created the “AI office”, placed within the management of the Internal Market Commissioner, Thierry Breton, and responsible for supervising this vast text.
Thirty of its agents have been recruited and 140 are expected by the end of 2024. The Commission is also busy with the first measures to be implemented: within six months, the ban on certain uses of AI, including identification real-time biometrics except for law enforcement, under certain conditions; or the standards to use to evaluate the performance and risks linked to large AI models.
Data transparency
The issue of transparency of training data is closely scrutinized by AI manufacturers and rights holders. Indeed, it was the subject of a tough battle lasting several months, before reaching a political compromise: the AI Act provides that manufacturers must provide a “sufficiently detailed summary” of the content used, but also that their right to “business secret” will be taken into account. Indeed, AI manufacturers, including French start-ups like Mistral, supported by France, had argued that too much transparency would force them to reveal their manufacturing methods to their competitors.
How to put this provision into practice? “A detailed summary is an oxymoron. I wonder about the operational nature of the implementation of the measure”explains Pierre Petillault, general director of the Alliance for the general information press, who, without delay, requested at the beginning of June the opening of negotiations on the remuneration of articles and images used for training models, in a letter addressed to twenty-five manufacturers, including OpenAI (ChatGPT), Google, Mistral or Meta (Facebook, Instagram).
You have 35.93% of this article left to read. The rest is reserved for subscribers.