The explosion of generative AI with the advent of ChatGPT has prompted questions within the cultural and creative industries (CCIs) and the entire French Touch ecosystem regarding the challenges of AI deployment. Creative processes, data access, standardization, intellectual property, and of course, the representation of companies and the presence of French talent on platforms and algorithms—these are just some of the many issues and challenges addressed during this roundtable discussion on "The Use of Generative AI: Opportunities and Threats for Content Creation," moderated by Johanna Tircazes, CCI Support Manager at Bpifrance.

Generative artificial intelligence is booming. How to explain it?

Mélanie Lopez-Mallet (MLM): Imagine a black box, the algorithm. This black box stores a collection of information, some more complex than others, which it uses to learn how to perform a task. If you input, for example, a photo of your pet, this more or less complex algorithm will translate that data into information, such as recognizing whether your pet is a cat or a dog. Its boom is mainly linked to the launch of ChatGPT and its conversational aspect, which gives the impression of being able to talk and interact with the machine. This anthropomorphization is at the heart of a new revolution in our use of computers.

How can we use AI to push the boundaries of creativity without crystallizing possible threats?

Michael Turbot (MT): In the music industry, our philosophy is to see how this interaction with this "black box" is most relevant. It allows for highly technical expertise in the production of new sounds, which we can then make more playful and productive. This word "productive" perfectly captures the idea of going faster (...) Through working with artists, we've realized that their creative processes have changed: they spend much more time on the creative aspect thanks to this new ability to create new sounds.

MLM: When a machine creates, it doesn't mean it replaces the creative person. The process at work isn't the same; there's no creative approach or intention involved, since the intention comes from the human who will use the machine. So it's not the democratization of generative AI that can kill creativity. For example, photography: just because you can capture a moment with a click doesn't make you a photographer, and just because we all have a camera on our phones doesn't mean we're all creating art. It's much more complex. Does the machine debase creation? Or is it up to artists to ennoble machines by doing something interesting with them?

What will tomorrow’s content look like?

Adrien Basdevant (AB): We are undeniably moving towards a world where more and more so-called "synthetic data" will be generated. This data will itself become a new source of feedback for the model. In a world of hybrid experiences, the question is: what will be the quality of the content feeding the algorithms? Generative AI has existed for over 75 years, and what we are currently experiencing with ChatGPT is a democratization of this technology. Should we then be talking about the homogenization of results? If today we cannot yet detect which generative AI tool was used to generate an image, will this still be the case tomorrow? Also, to what extent will it be necessary to disclose the source of information? It's clear that these questions are no longer just technical; it's a societal issue. For example, in your company, will you be allowed to use this type of tool? Without informing your superiors? This is the question of "shadow IT." Instead of taking divisive or simplistic positions by responding in a dichotomous manner, let's avoid prejudging these tools. Let's keep an open mind and participate in this co-creation.

How can we avoid this trap of homogenization of content, for the benefit of diversity?

MLM: The key is to include creatives. And the first step is to address their questions, doubts, and legitimate fears. Gaining perspective on this technology, through their viewpoints, is healthy. This allows us to gradually develop an understanding that transforms fear into comprehension. Some creatives will make the "informed" choice not to use generative AI, while others will say, "That interests me, it gives me an idea." The best moment is when the artist themselves expresses their interest. Because from that point on, we can build a project, a way to use these technologies. That's how we achieve quality content.

AB: If some content publishers decide they no longer want to be listed on ChatGPT, they can choose to disable the indexing of their content. This is called "opting out," and it was recently done by the New York Times and various publishers, groups, and media outlets in Europe, particularly France. It's important to understand that currently, in the pre-training of these algorithms, French-language content, for example, is already underrepresented. If everyone disables this indexing, you can be certain that, in terms of generated results, there will be very little content trained on French content. This raises a very broad question about content diversity—linguistic, cultural, and so on.

Hélène Nguyen-Ban (HNB): At Docent, we chose not to use "collaborative featuring" algorithms, as some cultural platforms or social networks do. Instead, we started research from scratch with the help of art historians to build algorithms capable of reconstructing clusters of visually similar artworks, as well as algorithms capable of screening millions of texts and identifying the words that define an artist's practice, in order to then make profound connections between artists and their practices. The goal is to recommend personalized content that doesn't rely on "since so-and-so liked Picasso, we'll recommend so-and-so," but goes far beyond this " echo chamber " which, when recommendation algorithms operate in this way, ultimately creates a form of cultural isolation. Finally, to encourage discovery—and not just a standardized kind of discovery—we partnered with Goldsmiths, a London-based school for art historians and contemporary art curators. They create online content for us, which we can use to communicate in English with artists from regions of the world who lack English-language content, in order to promote them alongside or in conjunction with Western artists.

How to reconcile access to data and intellectual property?

AB: For the user, there's first a very practical answer: benchmark the available tools to choose the one that best suits their use case, and then review the terms of service. It might seem trivial, but some platforms and tools allow you to own the generated output, while others don't. Some give you ownership of the generated result, others grant a license to use it, and in both cases, for a fixed or indefinite period. Then there's a more theoretical question: is content generated by generative AI protected by copyright? From Hollywood to Paris, the debate plays out in various ways. In France, for example, an illustrator who was supposed to publish a comic book created with the help of "MidJourney" had to cancel its publication because the publishing house feared repercussions for his image. So, it's clear that the debate isn't solely legal. It's a complex and multifaceted one. And that is why, before embarking on this, it is essential to fully understand the value chains.

When an artist picks up a tool, there is generally a back and forth between the tool and his own creative process. In the end who is the creator? What value do we attribute to the data used to train the model?

MT: Making music with AI is just as complicated as making music with a computer. Is the computer only used to combine tracks? We need to remain vigilant about this issue of intellectual property. Because there's a big difference between using AI to generate a track in someone else's style—which is copyright infringement—and using AI as a sampler.

AB: Regarding value sharing, the answer isn't solely legal. It's multidisciplinary because there's an economic aspect and a governance aspect. From the user's perspective, you can already learn to configure the tool: in the settings, you can choose whether or not your interactions with a chatbot, prompts, or results generated by your interaction will be reused for a subsequent retraining phase. In the creative process, it's different: there will almost certainly be less standardization if the prompts are more sophisticated. From a macro perspective, I think we need to facilitate discussion spaces to find new approaches, and ensure that tool developers don't have to approach each content publisher individually, and vice versa. Otherwise, only the most organized will be represented. These issues are not unique to the CCI sector: in healthcare, when the AP-HP (Paris Public Hospitals) produces health data used by Siemens, for example, to create a digital twin of the heart, the question also arises as to who owns the data and the algorithms used in its transformation. Managing to examine these questions dispassionately and avoid getting bogged down would allow us to benefit from a promising innovation.

Recently a bill was tabled to integrate AI into intellectual property. Should we regulate or let the actors do their thing as in the Metaverse?

AB: Often, as soon as there's a new innovation, people think it's the Wild West. But a robust arsenal already exists. This doesn't mean it shouldn't evolve, but we first need to succeed in applying what already exists, especially in areas where real innovation is needed. In the United States, for example, with " Fair Use ." Should we create " Fair Learning "? And consider the boundary between a temporary monopoly on exploitation and the dissemination of knowledge. I encourage those who are more involved in politics and law to be creative and understand technology and its uses. Don't just ban it...