A.I. Round Table - Artificial Intelligence: a tool for heritage?

A thorny subject in the light of recent American strikes, the role of Artificial Intelligence (AI) in cinema is becoming increasingly important. An issue addressed by the 11th MIFC during a round table discussion on Thursday, October 19. Rodolphe Chabrier, Julien Dupuy and Barbara Mutz, guided by Perrine Quennesson, discussed the use of A.I. in heritage cinema.

Explaining A.I.

First to speak, Rodolphe Chabrier began by explaining what A.I. is. There are A.I.s used in everyday life, such as those that define which e-mails are spam and which aren't, and then there are generative A.I.s, at the heart of the discussion. Co-founder of production studio Mac Guff, the speaker described the latter as something capable, from inputs, of giving outputs. For this, datasets are needed - millions of pieces of data of all kinds.

A.I. can therefore be understood as a creative tool, but it can also be used as an indexing tool. Represented by Barbara Mutz, head of its legal and regulatory affairs department, INA uses A.I. in this second way. With more than 25 million hours of archives in its catalog, using an A.I. makes it easier to identify the speakers, objects, and subjects of the preserved images. The Institut National de l'Audiovisuel can then propose relevant content for its clients, or on its own behalf. For example, it enables Arcom to assess the gender balance on television.

A.I.: creation or copy?

In addition to indexing, the question for generative artificial intelligences is how to qualify their creations. For Julien Dupuy, journalist for Trois Couleurs and columnist for "I.A. Quoi", A.I. is more about inspiration than plagiarism. For him, they can help to realize "cinephile fantasies", and thus give a different perspective on cinema: highlighting an artistic "aesthetic", questioning the cinephilia of creators, putting cinema codes into perspective.

Rodolphe Chabrier shares this vision. For him, heritage cinema can benefit from artificial intelligence, not only in terms of restoration, but also in terms of interpretation and enhancement. Works created by A.I. cannot then be recognized as the original works, but as interpretations of them.  

In this respect, Rodolphe Chabrier and Julien Dupuy are unanimous: A.I.s are partners in creation, proposing several results to achieve what the person calibrating the software with precise prompts had in mind.

Use in heritage cinema 

The co-founder of Mac Guff Studios then goes on to discuss the use of A.I. in heritage cinema, arguing that it can be beneficial to heritage works. He cites the example of L'arrivée du train en gare de La Ciotat: thanks to A.I., the image can be made more fluid (by adjusting the frame rate), avoiding jerks, and adding colorization. A.I. can then bring a distant heritage to life, bringing it closer to contemporary viewers, creating proximity between a society of yesteryear and that of today. It can also create a better-quality image, isolating and correcting visual defects such as lighting, blurred faces, or film coloring.

Artificial intelligence can also help store and distribute heritage images. Indeed, A.I. can help compress catalogs, making the conservation and distribution of heritage works less restrictive.

As far as constraints are concerned, he also explains that artificial intelligences help to reduce restoration costs: previously, studios could invoice in hours, but now with A.I., invoicing is done in minutes. Costs are divided by thirty.

Lastly, it is possible to revive a lost heritage: this is the work of Mac Guff Studios in the French TV program L'hôtel du temps, in which Thierry Ardisson welcomes deceased celebrities to redo their biographies. A feat made possible by the Face Engine technology.

Legal and financial issues

These issues also raise legal and financial questions. Indeed, A.I. raises the question of authors' moral rights. On this subject, Barbara Mutz explains that there is a beginning of legislation specific to artificial intelligence, stemming from a 2019 directive. As artificial intelligence is based on exceptions to copyright and neighboring rights, A.I. will overturn the general rule that authorization must be sought to use an image. Here, what is not forbidden is understood to be authorized: an A.I. can therefore delve into works if it has had lawful access to them, and if the authors holding the rights have not objected by means of machine-readable processes. In the opposite case, there is an opt-out. She cites the example of Sacem, which has exercised this opt-out for all its members. But if this legislation is a start, Barbara Mutz reaffirms the need for tools to help apply the opt-out in the best possible way, with effective implementation for rights holders.

This legal issue of copyright raises a financial question: should A.I. users pay royalties to the creators of the works entered in datasets? And if so, how? For Rodolphe Chabrier, A.I.-generated works are a matter of inspiration, so it's difficult to remunerate the artists from whom the software draws its inspiration. The co-founder of Mac Guff Studios is not opposed to the idea of fair remuneration, which he is already trying to apply to datasets that are traceable. However, he finds it difficult to define who should be remunerated, if not all datasets have been used by A.I.. 

Barbara Mutz adds that respect for copyright, transparency and traceability are fundamental to the use of AI. The notions of parasitism and unfair competition were also raised, as well as the problem of clearance, during an exchange with the audience.