ROUND TABLE - A.I and the audiovisual industry: Towards what French and international regulatory framework?

Continuing on from the discussions on artificial intelligence held on Thursday October 17, this Friday morning's MIFC featured a discussion focused on French and international regulations around AI, moderated by Perrine Quennesson.Valérie-Laure Benabou, Juliette Prissard and Justine Radel-Cormann were responsible for outlining the legal framework.

AI raises two major issues

A new study by the European Audiovisual Observatory, published this Friday, October 18, lays the groundwork for the use of artificial intelligence (AI) in classic cinema in Europe. Presented remotely by its legal analyst Justine Radel-Cormann, the study points to a dual application in the heritage sector, in the service of indexing and restoration. This dual use is not without consequences: Justine Radel-Cormann lists the film formats that are difficult for AI to process, the costs that such use can generate, and the risks of over-dependence on the machine, which can lead to the loss of skills.

She then raises the various legal and regulatory issues raised by AI, which can be summed up in two points: the question of copyright and the remuneration of outputs. Faced with these issues, the legal analyst concedes that there is still a vagueness surrounding the European regulatory framework on AI, with software regulated by only two specific texts, not always adapted to generative AI. 

Find the full version of the European Audiovisual Observatory's publication on AI regulation by clicking here. 

On copyright

On the question of copyright, Juliette Prissard, General Delegate of EUROCINEMA, began her speech by reminding that output applies to generative AI, allowing data to be searched unless contraindicated. Valérie-Laure Benabou, Professor of Private Law at the Université de Paris-Saclay / UVSQ, points out that there are limitations, but they are still modest, because the output chain has not been regulated. Technical systems still need to be created to say no to the robot on all the catalogs to be searched, which would imply changes in companies' internal regulations.

AI regulations also need to be reviewed, in that they are not adapted to generative AI with outputs competitive with conventional cinema, leaving countries to regulate on their own. In France, copyright is a special case, authorizing reproduction for reproduction's sake. It remains to question the exception and adapt it as best we can.

On remuneration

The issues surrounding copyright raise others concerning the remuneration of outputs. How can we remunerate outputs if we don't know who the authors are? And according to what model? Juliette Prissard explains that agreements have already been signed with press institutions, without the details being revealed, which maintains the vagueness surrounding AI remuneration. This raises two questions: how these agreements will be distributed, and the value of the data provided. Without transparency on the economic model used, lump sums are allocated, which may not remunerate the authors sufficiently.

The remuneration of inputs also raises questions. Valérie-Laure Benabou adds that we don't yet know what this involves, or how to pay for it. She explains that with AI, there are several phases, and therefore several types of data. Some are used for learning the tool, for exercises, for validation and so on. The question then arises as to which inputs should be remunerated according to the phase in which they are used. 

A common question: origin

Finally, the main question raised by this round table is that of the origin of the works produced by generative AI. For Valérie-Laure Benabou, it is necessary to know to what extent humans remain in control of these tools, at a time when synthetic prompts can produce outputs without any human supervision.

To encourage people to claim the origin of their creations, a common solution seems to be emerging: that of placing unclaimed outputs in the public domain, within a framework similar to that of the paying public domain. They would then be subject to a royalty, to encourage human rather than synthetic creation. An idea also put forward by Juliette Prissard, for whom remuneration should favor human creation.

She also advocates the identification of prompts as proof of originality. She gives the example of a Chinese court for which human originality is proven even if it was a machine that produced the output, because a human guided it in its manufacture.

A final point, which echoes the need to regulate the use of AI without restricting it, is that of the invisibilization of cultural diversity in AI-generated projects: if we don't pay attention to this, there will be an imbalance in content between AI-friendly countries and the others.