The challenges of AI and opt-out: no time to withdraw

The recent book The AI Revolution: the International Legal Perspective, co-edited by Anne-Marie Pecoraro and Alexander Ross, offers a global reflection on the legal challenges posed by artificial intelligence in the creative industries.  In this ‘In Fact’ section, let’s take a step back and put the law at the centre of the debate.

Artificial Intelligence (AI) is all over the news. Is it a formidable tool at the service of society? Or a threat to the world of work? Or even a dormant danger? These are dizzying questions, given the speed of progress in this field. To avoid any headache (or an existential crisis), let’s take some time to consider something concrete and stable: the law. 

When it becomes generative, the ability of AI to create « new » content (such as music, images, or text) raises even more questions about the protection of copyright in pre-existing works. To learn and progress, AIs meticulously examine huge databases: this is text and data mining (TDM). The debate is never-ending. 

The introduction of Directive 2019/790 « on copyright and related rights in the Digital Single Market« , known as the DSM, lays down exceptions and limitations to the extraction of data needed to train AI models. It allows rightsholders to explicitly prevent such activities through opt-out clauses. 

The right of withdrawal at European level. In accordance with the DSM Directive, Member States must establish limitations or exceptions to data mining for certain rights, including the author’s exclusive right to authorise or prohibit the temporary or permanent reproduction, direct or indirect, in whole or in part, of their work. This exception must apply unless the use of the works has been expressly reserved by the rightsholders, in particular by machine-readable means in the case of content placed online and accessible to the public (Article 4 of the Directive).

The right of withdrawal in France. The transposition of the directive into the French Intellectual Property Code (CPI) defines TDM as the application of an automated analysis technique to texts and digital data in order to extract information (patterns, trends, correlations). The CPI introduces two regimes distinguishing between searches for scientific research purposes and those that go beyond this strict framework (Article L122-5-3). 

Thus, in principle, the digital reproduction of legally accessible works may be made without authorisation when it is carried out for the purposes of scientific research by authorised organisations (libraries, laboratories, archives, museums, etc.) or at their request for other persons. However, the penultimate paragraph of the article of the CPI sets out an exception to TDM in the form of a right of withdrawal up to the author: « [...] unless the author has objected in an appropriate manner, in particular by machine-readable processes for content made available to the public online« .

European and French regulations on the right of withdrawal with regard to TDM thus seek to strike a balance between the interests of rightsholders and those of artificial intelligence developers. But there are many challenges. 

The challenges associated with the right of withdrawal.

In practice, implementing and enforcing the right of withdrawal remain complex tasks. The first challenge is certainly balancing power between rightsholders and AI companies. The former will have to negotiate with the latter, thereby reducing control over their work and their potential compensation in the event of use. 

From the point of view of specialist companies, the limitations placed on TDM by authors and their rightsholders may raise questions with regard to the development of AI technologies. 

Another major problem is the lack of transparency, particularly when databases are private. Added to this is the difficulty of implementing the concept of « machine-readable processes » in practice. 

The AI Act, published on the 1st of August 2024, provides a partial solution to this challenge by imposing a negotiated concept of transparency, but the mechanism leaves many questions open. 

Practical solutions to overcome these challenges.

To address these uncertainties, we can list a number of practical solutions. Standardised withdrawal clauses may be designed by industry bodies such as the Syndicat national de l’édition (SNE) in France. Machine-readable limitation protocols, such as the TDM Reservation Protocol, may be introduced. Solutions by start-ups such as Spawning’s Do Not Train Registry, could also enable artists to assert their right to withdraw via a list of their intellectual property rights.

Don’t withdraw.

The right of withdrawal framework in the European Union and France still presents significant challenges. Although the DSM Directive and its transposition in France are moving in the direction of better protection for rightsholders, practical difficulties remain. Continued cooperation between rightsholders and AI companies remains essential to achieve the right balance between innovation and intellectual property protection.

This article is based on the book The AI Revolution: the International Legal Perspective, presented at the Amsterdam Dance Event 2024 and edited by Anne-Marie Pecoraro and Alexander Ross.