Alexandra Bensamoun is a copyright specialist and lecturer at Paris-Saclay University. After participating in the report of the Commission on Artificial Intelligence (AI), presented to the government in March, the Ministry of Culture entrusted her – through the High Council of Literary and Artistic Property – with a mission on the application of a much-debated provision of the regulation of the European Law on AI: the obligation, for AI manufacturers, to provide a detailed summary of the data (articles, books, photographs, etc.) used to train their models.
Its report on this point, which had sparked opposition from Paris and companies such as Mistral and Meta, is expected at the end of 2024, before the international summit on AI to be held in Paris in February 2025.me Bensamoun was also tasked with giving his opinion on business models between AI manufacturers and content owners (press, publishing, music, etc.), another sensitive topic, while the first paid agreements were signed between OpenAI and the financial times, News Corporation (The Wall Street Journal…), Rush Media (The Country) or, in France, The world.
PWhy is AI revolutionizing copyright?
Copyright was born from a technical revolution: printing. And every technological revolution has shaken it up: the Internet, the mobile Web, streaming… Today, it is AI’s turn to counter it. Indeed, generative AI, which produces texts, images or sounds, must be trained with quality data. And cultural and media content is quality data. Copyright is therefore part of the value chain of AI design and must be valued.
How can we reconcile the two contradictory principles of the AI Act: the obligation of AI manufacturers to provide a detailed summary of training data and their right to invoke trade secrets to preserve their know-how?
The important thing is to get out of the way and recognize that there is a value chain. We pay for all the inputs: computer chips, talent, electricity, etc. The content used to train AI also has a price. The difficulty is that the AI Law requires that this summary of data be available to everyone. It would have been interesting to entrust it to an intermediary. [par exemple, le Bureau européen de l’IA] preserve, if necessary, confidentiality.
That said, we cannot invoke trade secrets to exempt ourselves from applying the law. Moreover, this concept is mentioned in the explanatory recitals [des annexes] of the text, but not in the provision itself, which has the force of law.
You have 79.01% of this article left to read. The rest is reserved for subscribers.
