When you generate images via Midjourney for your advertising campaigns, have you ever wondered if your competitor could take them over? ? Can I sell a book written entirely by Bard and receive royalties? ? If I am an artist on Leonardo.ai, can I sue for plagiarism ? The boundaries of creativity are being redefined by artificial intelligence, and with this emerge crucial questions about intellectual property.
I prompt a poem, who does it belong to ? To me, because I promptly ? To Open AI because it trained the language model ? To the authors of the texts which allowed the model to train ? To no one ? Often, when I ask these questions, people realize that they do not have the answers. Then, those who are a little in the know answer me frequently : “The AI Act will fix this. »
You have a document
The AI Act is a draft legislation carried out by the European Union since 2021 to regulate the artificial intelligence ecosystem. It mainly aims to measure “the dangerousness of AI” via a pyramid, ranging from minimal risk (spam filter, video games, etc.) with unacceptable risks, at its peak, such as social scoring, facial recognition or mass surveillance. Like what the Europeans did with the GDPR (General data protection regulations), Europe's objective is to legislate.
2024, the year of the trials
We are currently in a legal limbo ! Currently, few people have answers, and specialized law firms receive many requests. To reassure their customers and users, companies such as Microsoft, Open AI, or Canva are committed to offering legal protection in the event of a dispute. But in this legal vagueness, 2024 promises to be the year of trials. The challenges surrounding intellectual property linked to artificial intelligence go hand in hand with the emergence of new cases, requiring courts and lawyers to decide in areas where the rules are not yet clear.
Opt-out and ethical questions
The emergence of generative artificial intelligence also poses crucial ethical questions, particularly with regard to consent to data mining. Without data, you cannot train a language model or a diffusion model (AIs that generate texts or images to put it simply). L’Opt-out, a practice where websites (for example those of the press) insert code so that robots do not use it in its archives, etc., becomes a central subject. However, a major concern is that many AI models have already been trained without the explicit consent of content creators. Massive datasets were used, often including content created by creators without them having the opportunity to refuse. This raises questions about the legitimacy of past use of data : should creators be consulted retroactively for consent ?
At the end of the day, the challenge is to find a balance between technological innovation and respect for individual rights. Artificial intelligence is not just a technological breakthrough, it is also a break in the way of legislating.
