Creative AI, disregarding copyright?

As artificial intelligence (AI) revolutionizes the field of intellectual creation (music, images, or generated texts), it also raises pressing legal questions. How can copyright law, originally designed to protect human intellectual creation, adapt and organize itself in the face of this new challenge?

From paintings to musical compositions, as well as literature and design, AI has demonstrated its ability to create works that challenge the traditional boundaries of creativity.

Creation of "Original" Works

One of the many examples of artificial intelligence's creative power is the "Portrait of Edmond de Belamy," a painting created by the Obvious collective using AI, which sold for $432,500 in October 2018.

Portrait of Edmond de Belamy - Christie's

Other tools are capable of creating artworks from a prompt (textual description). This is the case with the artificial intelligence Midjourney.

Prompt: Draco Malfoy and Hermione Granger, backs to photo, stained glass, detailed. - Midjourney

Another notable example is the song "Heart on my Sleeve", entirely generated by artificial intelligence, replicating the voices of singers Drake and The Weeknd. This work is attributed to the anonymous Ghostwriter977. Posted in a video on TikTok, it accumulated over 15 million views in 48 hours. The song was even declared eligible for Grammy Awards, according to the CEO of the Recording Academy (the organization that runs the ceremony).

In response, Universal Music Group, the label representing these two artists, requested that music platforms (Apple Music, Spotify, etc.) prevent AI systems from training using music protected by copyright.

These examples illustrate AI's ability to create art, but they also raise significant questions regarding the protection of authors' rights. This is because AI does not truly create; it draws inspiration from existing content, created by humans and often protected by copyright.

So, what are the risks for human authors? Can these AI-generated creations be considered as works in the sense of copyright law?

What is a Work in the Sense of Copyright Law?

Article 112-1 of the Intellectual Property Code (CPI) states that "all works of the mind, regardless of their genre, form of expression, merit, or purpose" are eligible for copyright protection.

However, another criterion applies: the originality of the work. Even though this condition is not explicitly mentioned as such in the CPI, it is widely accepted in both European and French law. Indeed, originality is defined as the mark of the intellectual contribution of the author.

For a work to be considered protectable, it must express the personality of its author. The author of any protected work then has moral and proprietary rights.

In the case of a work created by AI, this condition of originality seems difficult to demonstrate, given the way these systems are trained.

AI Training and Risks for Authors

Artificial intelligence heavily relies on data, often including works protected by copyright: texts, books, images, music, or newspaper articles. Developers collect this data to train AI models. Therefore, the question arises: does this fall under fair use, or does it require a specific license?

The unauthorized use of these data, most often protected by copyright, poses a problem and can thus give rise to legal disputes.

Getty Images, an online image bank, filed a lawsuit on February 3, 2023, against the company Stability AI for the misuse of more than 12 million photos to train its image-generating AI system, Stable Diffusion.

Another case highlighting the debate on AI and copyright is that of The New York Times, which sued Microsoft and OpenAI for unauthorized use of its articles to train their AI models. The newspaper claims that "millions" of articles, some of which are "only accessible by subscription," were used without its permission to improve ChatGPT's performance. This would allow the tool to compete with the newspaper as a reliable source of information.

Moreover, The New York Times accuses AI of reciting the content of its articles verbatim, summarizing it, and even mimicking the writing style, as demonstrated by numerous examples. AI also allegedly falsely attributes information to the newspaper.

Given these concerns about copyright, a legislative response becomes more than necessary.

Legislative Developments on AI and Copyright

In response to the challenges posed by AI, national and international legislators have begun to rethink existing laws and take steps to adapt them. France and the European Union aim to play a pioneering role in regulating the use of AI in the field of copyright.

Article 53 of the European AI Act establishes specific obligations for providers of general-purpose AI models, particularly regarding copyright:

Respect for Copyright: Providers must implement a policy to respect copyright as defined in the EU. This includes identifying and respecting expressed rights reservations¹ in accordance with Article 4(3) of Directive (EU) 2019/790².
Summary of Content Used for AI Development: Providers must prepare and make available to the public a sufficiently detailed summary of the content used for the development of the general-purpose AI model.

¹A rights reservation means that if a copyright holder does not want their content to be used by an AI system, they can express this reservation [Editor's note: opt-out]. This rights reservation must be made using machine-readable methods, including metadata and the terms of use of a website or service.

In other words, if an artist or creator does not want their work to be used by an AI, they can specify this so that the AI can "understand" this refusal and respect the rights reservation. This will help protect copyright.

²Directive (EU) 2019/790 of April 17, 2019, on copyright and related rights in the digital single market.

In French domestic law, a bill was introduced to the National Assembly on September 12, 2023, to "regulate artificial intelligence through copyright." This bill proposes, in its first article, to subject the use of protected works by AI to the prior authorization of the authors or rights holders.

This bill also aims to:

Transparency for AI developers with a requirement to mention that the work was generated by AI and to acknowledge the authors of the works from which the AI draws its inspiration.

However, it is important to ask whether it will be possible to prove that content is derived from a work, and by what means. Will checks be carried out by a "special IT brigade"?

Would this not mean requiring AI developers to disclose the computer code, in violation of industrial secrecy? A balance will undoubtedly be necessary between protecting this secrecy and respecting copyright.

A taxation project on the company that developed the tool for works whose origin can be determined.

While this would encourage and protect authors, would it not, on the other hand, hinder innovation?

These initiatives reflect a growing awareness of the need to adapt the legal framework to technological changes. However, this (commendable) legislative intent is likely to encounter several challenges, the resolution of which could determine the future of AI.

Nevertheless, one of the next challenges for the legislature will likely be determining whether AI creations can be protected by copyright. In this case, who will hold the rights: the AI provider, the tool user, or the author of the work that inspired the AI? Time will tell...

ChatGPT