In the cultural sector, artificial intelligence is a divisive subject. Animated films have been attacked at festivals because some of their visual elements had been generated, and within teams people often avoid talking about it. Meanwhile, according to the 2026 Baromètre du numérique, 48% of French people already use these tools, and 76% of managers and professionals. I suggest a method drawn from the eleven months of the Forum des images monitoring committee, in 2024-2025: first ask everyone where they stand, then meet on fixed dates, give everyone a chance to speak, and keep a written record. The team can then decide what it does with these tools, including doing without them. It still needs to know what they change in work: they take over tasks, and professions evolve as a result.
A divisive subject, in the cultural sector more than elsewhere
Why does this subject divide the cultural sector so much? Because it touches what people care about most in their work: creation, the work itself, the labour of the hands and of the mind. In the report on the artificial intelligence monitoring committee of the Forum des images, I listed the questions that keep coming up: the status of authors, environmental impacts, the risk of new inequalities, trust in images, the stereotypes these tools carry, etc. To these must be added the concentration of power in the hands of a few companies. In animation, films have been attacked at festivals because a machine had produced some of their visual elements, sometimes only the textures, as I recounted in Invisible artificial intelligence. In May 2023, the Bradford Literature Festival, in England, withdrew the AI-generated images that illustrated its programme and apologised, after protests from authors and illustrators. I also described in Conscious compromise with artificial intelligence the uncomfortable position of those who criticise a technology they depend on, for example a union circulating an op-ed against the extractivism of AI models, a text that bears all the marks of a large language model.
Should we try to get everyone to agree? I don’t think so. These questions are real, and no internal memo will settle them. On the monitoring committee, “we sometimes had a good laugh, and sometimes we argued too”, as I wrote in the report, and that was a very good thing! Claude Farge, managing director of the Forum des images, for his part stressed the importance of social dialogue: in his view, imposing such changes by force is neither possible nor desirable. On 17 October 2025, at the launch of the Observatory of AI in the Arts and Media (ObsIA) at the Sorbonne Nouvelle, I had argued the same idea: a cultural institution has every interest in organising controversy rather than closing it down (Cultivating empathy and controversy : my proposals for human governance of AI in cultural institutions). So I start from disagreement.
Appropriation moves faster than the organisation
While this debate goes on, practices are spreading. According to the 2026 Baromètre du numérique, carried out by Crédoc for Arcep, Arcom, the Conseil général de l’économie and the Agence nationale de la cohésion des territoires, based on a survey conducted in June 2025 among more than four thousand people, 48% of French people say they use generative artificial intelligence, compared with 20% two years earlier, and the figure reaches 85% among 18-24 year-olds and 76% among managers and professionals. A third of users turn to it every day, and more than half among 18-24 year-olds. The most frequently cited uses are searching for information (73%), writing and translation (58%), finding ideas (57%) and help with homework (44%).
According to the same survey, 41% of users consider themselves competent, and 11% very competent. These people also work in libraries, theatres and museums. Long before artificial intelligence, the sociologist Norbert Alter studied, in companies, what happens when employees adopt something new in their own way, without waiting for their management, in L’innovation ordinaire (PUF, 2000). He distinguishes invention, the novelty that management introduces, from innovation, which would be the appropriation of that novelty by people, often on the margins of the rules. According to him, the organisation would only recognise these practices after the fact, and in the meantime the people who carry them would be seen as “deviants”, until they grow weary of it. That is why it seems a pity to me to act as if no one used these tools: we leave alone the people who have acquired these skills. They then have good reasons to keep quiet.
Assuming everyone has tried, or that no one uses it
On 16 October 2026, at the Bibliothèque publique d’information (Bpi) in Paris, I led the last afternoon of the course “Éducation du regard : images, médias, informations” (Educating the eye: images, media, information) organised by Images en bibliothèques: two and a half hours on generative AI in image education, with some fifteen professionals from libraries and the film sector. I had opened the session with a series of quick questions, asked as if everyone had already used these tools. One participant pointed out that two weeks earlier, in another training course, most of the group had never done so, whether by choice or not. He was right, and I told him so in front of the group. It is a common mistake among those who, like me, work with these tools every day, and it immediately puts a good part of the people present among the latecomers; and we do not speak to someone we think is lagging behind as we speak to the others. I argued in Not judging people’s digital practices that mediation means neither judging people’s practices nor ranking them; this applies within a team as much as with the public.
The opposite mistake is just as common: acting as if no one used these tools, because no decision has been taken, because management is reluctant, or because the subject is divisive and people prefer not to raise it. What happens then? Everyone keeps to themselves what they do, or refuse to do, for fear of being judged, and the team benefits neither from the experience of some nor from the vigilance of others. For several months I have observed this silence in the intellectual and cultural circles I move in, where people criticise in public the tools they use in private, as I recounted in Conscious compromise with artificial intelligence. In both cases, we make assumptions about other people’s practices and positions instead of asking them.
Four positions in every team
When you take the time to ask, you find at least four positions in any team, and each calls for different support:
- People who have never tried, for lack of opportunity or curiosity. During the TUMO x AI event, which preceded the Forum des images monitoring committee, nearly 500 TUMO Paris students made 200 short films in small groups with these tools, after an information session and before a debate. That is the kind of collective setting I would wish for a first experience.
- People who refuse, and who say so. Their reasons are ecological, social, professional, moral, etc.
- People who use these tools without knowing it, for example when their search engine displays a generated answer above the results. The sociologist Susan Leigh Star speaks of “infrastructure” for what serves constantly without being noticed, and I drew the consequences for artificial intelligence in Invisible artificial intelligence and in Who uses artificial intelligence without knowing it ?. Asked “who has already used it?”, many people therefore no longer know what to answer.
- People who use these tools a lot and don’t say so. Alter calls them innovators. I am one of them, and I recounted in Contribution cannot be commanded how I work every day with language models.
Around the table of the Forum des images monitoring committee sat a philosopher, an inspector general of social affairs, a visual artist, managers from TUMO Paris, and representatives of Salesforce and the City of Paris. This diversity was, as I wrote in the report, “a great richness”.
Why use it, and what for
Since December 2025, I have written eight articles on what these tools change in work. I summarise them here:
- Tasks and professions. It seems useful to me to distinguish, as I proposed in Artificial Intelligence, Work and Professions, tasks (summarising a document, writing minutes, translating a text, etc.) from professions, which bring tasks together around a body of know-how. On my website, for example, the English version of each article, including this one, is translated with a language model; I choose what is published, I review it, and I take responsibility for it. These tools take over tasks, and professions therefore evolve towards higher qualification. I drew the consequences for an organisation in Transforming Your Organization in the Age of Artificial Intelligence.
- Possibilities. A small team can now build its own working tools, in free software and with the help of these models: a shared contact database, project tracking, festival management, etc. Until recently, you had to know how to code or pay someone who did. I explained how this is done in RAG, or how to give an AI the power of a business tool and in Building professional tools with the teams.
- Ways of working. Working with these tools forces you to make explicit what you want, for whom, in what context, and having to say it already changes the work, as I analysed in Context : The Key to Artificial Intelligence in Cultural Organizations and in Artificial Intelligence and Cultural Actors : The Laws of Creative Transformation.
- Ways of collaborating. A person placed in the position of an executor executes; invited to contribute, they give much more. In Contribution cannot be commanded, I took the example of FAVI, the foundry in the Somme that Jean-François Zobrist reorganised from 1983 onwards into mini-factories of around twenty people, each responsible for its own production, and I showed that the same law holds with language models. These tools also change the way a team produces its narratives and its communication: this is the subject of the one-day training course I offer in Crafting cultural narratives with artificial intelligence.
- Mediation and the public. The people cultural organisations welcome use these tools, or fear them, or refuse them, and a team that has not worked through the question for itself will approach it badly with them.
Of these shifts, the first seems to me the most important, because the place we take depends on it. On the artificial intelligence monitoring committee of the Forum des images, Louis-Charles Viossat, inspector general of social affairs, suggested that we reason in terms of tasks rather than professions: we cannot say that the profession of screenwriter will disappear, but some of the research tasks that make it up will be transformed, as I reported in Webinar presenting the findings of the AI Monitoring Committee of the Forum des images. Mathematics offers a precise example. In November 2025, Terence Tao and three colleagues, Bogdan Georgiev, Javier Gómez-Serrano and Adam Zsolt Wagner, published the results of their work with AlphaEvolve, a Google DeepMind system, on sixty-seven problems in analysis, combinatorics, geometry and number theory. They were the ones who wrote the programs that evaluated the machine’s proposals, guided the search and checked some of the proofs by hand, and they note that the system gives much better results in the hands of a specialist in the field. In the other sciences, Yian Yin and his colleagues at Cornell University showed in Science, in December 2025, that researchers who write with these tools publish more and cite a wider range of work, but also that many of the articles written this way are brilliant in form and weak in substance; I discussed this in Intelligence as a Resource.
I believe we gain by defining ourselves, as human beings, not as cognitive agents doing the same tasks as the machine, only more slowly, but as responsible people who orchestrate the work and answer for what it produces, like these four mathematicians who write the evaluation programs and check the proofs. That, in my view, is the best we have to offer. If we do not offer it, we leave power to the machines, and that would be serious.
Does that mean accepting everything? Of course not. In 2023, the Bradford Literature Festival decided to stop commissioning generated images. An organisation can also ban these tools for certain tasks, or reserve them for trained staff: that too is a way of exercising this responsibility.
Six principles to move from the corridor to the committee
I offer the teams I work with six principles, which I put into practice at the Forum des images, where I co-designed and led this monitoring committee for eleven months from 13 November 2024, as I recounted in Forum des images : AI watch committee (2024-2025) and in the findings webinar:
- Start by asking, and in the right order. That is what I missed at the Bpi. You ask the questions as a sit-stand exercise, and everyone stands up when their answer is yes: Who has never used one of these tools? Who has used one without choosing to? Who has chosen not to? Who uses them every day? It takes five minutes, and everyone sees who stands up at the same time as they do.
- Provide a regular framework. At the Forum des images, we had planned six working meetings, from November 2024 to September 2025, around four themes: artistic creation, ethics, the evolution of professions, and education.
- Give everyone a chance to speak. I suggest exchanges in pairs. For example, on the subject “artificial intelligence in our library”, the first person spends a minute and a half explaining to the other what these tools can bring, then listens for the same length of time. In the next round, with another person, everyone looks for what is dangerous about these tools. It is not about pretending, but about finding real arguments, for and then against, and the shyest people speak as much as the others!
- Keep a record. For the Forum des images committee, I published the objectives and the list of participants from the launch session on 13 November 2024, then the synthesis of the eleven months of work after the findings webinar. In a team, minutes after each session are enough: the uses described, the questions, the disagreements, the decisions taken and those still open, etc. Those who were absent can follow, and management can show its funders that it has worked on the question.
- Give a role to those who refuse. Explicitly entrust them with a task: making sure that ecological, social, legal and professional questions stay on the agenda. At the ObsIA, I called this making disagreement “the engine of collective intelligence”.
- Arrive at rules, and date them. At the findings webinar, Claude Farge announced several commitments for the Forum des images: making the committee permanent, exploring a local AI server running on free software and using less energy, training mediators, and creating a kind of “Nutri-Score” telling the public how AI has been used in cultural offerings. A charter of use can also draw on the four modes I distinguished in Participatory design and Artificial Intelligence (augmentative, creative, as a subject, or interactive AI). It carries a date and a review deadline, because it will be out of date within the year.
So it takes very little: one person to facilitate, an hour and a half a month, and management that accepts experiences it did not decide on and opposition it did not wish for. In return, everyone knows where the others stand, and the team can decide what it wants to do with these tools, including doing without them. At least it will know why!
