Artificial Intelligence and Cultural Actors: The Laws of Creative Transformation

Cultural institutions that passively wait without embracing artificial intelligence could see their neighbors transform in just a few months. The difference would depend neither on resources nor training, but on dynamics that I believe are important to explore.

Principles that we rarely share

One could imagine that a cultural institution could completely reinvent its relationship to creation in just a few months, even if today it finds itself entangled in old methods, without a technology budget, without a trained team, without even understanding what “artificial intelligence” really means. One could say this is impossible, that institutional transformations take years, that committees must meet, that training programs must be deployed, that audits must be conducted, etc. The problem is that by the time projects are finally built, they will already be obsolete. Whereas relevant cultural projects with AI can emerge in a few weeks, while other structures will still be deliberating for years.

This might be a matter neither of chance, nor luck, nor budget, but rather of a recurring dynamic, of principles that institutions successfully navigating their transformations rarely share, perhaps because they seem counterintuitive. These principles were already evoked by Michel Serres in Petite Poucette (2012) when he described a paradigm shift where knowledge becomes accessible and externalized. AI would embody this spectacular externalization today, and it seems that those who precisely formulate what they want to do with it see transformation happen more quickly.

How can we explain that one cultural structure can remain paralyzed for years in the same theoretical debates about AI, while another launches innovative projects in just a few months? Why does one team accumulate reports, meetings and concerns, while another sees possibilities unfold as if they were just waiting for them? The difference might not depend on the size of the institution, not on the team’s degrees, nor on political connections. It could reside in something more subtle yet accessible: the capacity to transform a vague intuition into a defined intention.

Nearly 400 participants in AI filmmaking workshops that I facilitated in 2024 experienced this dynamic in practice:

• Those who entered the workshop with a vague idea, “I’d like to explore AI,” left with a precise, constructive and reproducible experience.
• Those who entered with a precise intention, “I want to create a three-minute film about family memory using AI to generate images of a lost photo album,” left with a finished work, a deep understanding of possibilities and new skills.

It seems the machine responds to the precision of creative desire. But what is it, and how do we build it? Hannah Arendt distinguished in her work on the human condition labor, work and action as three fundamental dimensions of our relationship to the world. AI might demand this same clarity about what we want to accomplish.

Defined protocol rather than participatory dispersion

A first principle emerges from cultural structures that successfully navigate their transformations: innovation with AI seems to flee vague participation and engage in defined protocol. Too many institutions launch “participatory projects with AI” without having determined which approach they’re adopting. They want to “involve audiences” and “use AI” but these formulas might contain no operative force. The project risks dispersing, the team feeling overwhelmed, and after months of effort, few substantial things emerge, it often remains anecdotal.

It might be useful to understand what I consider a fundamental distinction. There would be two approaches to participatory creation, and each would require a different protocol with AI:

• The participatory approach, where participants become interpreters of an artistic vision carried by a creator, which can be enriched by augmentative AI that enhances everyone’s capabilities.
• The cooperative approach, where the creator designs encounter protocols and accepts that their project will be transformed by collective contributions, can integrate AI as an autonomous actor in the creative process.

Wanting to mix the two without defining the framework can lead to dead ends. The protocol must be precise. We often observe this phenomenon. A cultural structure wants to create “the brilliant project where everything is participatory and AIs are everywhere.” It starts with enthusiasm, but after a few weeks, the host structure feels overwhelmed by unexpected directions. And it ultimately ends up taking back control in an authoritarian way so the project can be finalized, which potentially produces the opposite of what it initially aimed for. Paulo Freire warned in Pedagogy of Autonomy (1996) that freedom in learning must be accompanied by a framework enabling its effective exercise.

Conversely, those who precisely define their protocol before starting, “we’re creating a participatory performance where AI generates sets in real time from emotions captured in the room” or “we’re setting up a cooperative device where AI proposes three narrative directions and the group votes to choose,” seem to see their projects unfold more fluidly. Definition would precede realization. This methodological clarity would draw inspiration from the principles of democratic pedagogy developed by Célestin Freinet, where art becomes a democratic space allowing exploration of diverse perspectives, but always within a defined framework.

Defined alliance rather than fear of replacement

A false idea circulates and could paralyze cultural actors: AI would replace humans. This vague fear risks producing blockages and inaction. Structures wait, deliberate, debate their opinions, without ever confronting action, that is, the reality of experience, which is the only ground for learning. Meanwhile, other institutions have already trained their teams, launched projects, discovered that AI doesn’t replace but transforms.

AI might more closely resemble a new team member than a replacement. But to collaborate effectively with this member, one might need to precisely define their role. Creators who succeed with AI generally don’t tell it “do something creative.” They would rather define: “generate three variations of this image in 18^th-century Japanese watercolor style” or “propose five ways to transform this phrase while respecting the initial poetic rhythm.” Precision would call forth performance.

This alliance between human creator and AI might resemble what Bernard Stiegler called “pharmakon” in Technics and Time (1994): a technology is both poison and remedy, depending on how it’s used. Generative AI would be neither savior nor destroyer by essence. It becomes what creators and institutional leaders make of it, provided they define how they want to work with it. Without this definition, the tool will remain unusable or, worse, it will impose its own logic on the creator. Of course, to be precise, one must first experiment individually with numerous AI tools, and have hierarchical and financial support to do so.

Invention rather than reproduction of dominant models

A trap particularly threatens cultural institutions: adopting dominant models without questioning. Using ChatGPT because “everyone uses it,” employing MidJourney because “it makes the most beautiful images,” following training offered by big tech companies because “it’s convenient.” This apparent ease masks a strategic trap.

GAFAMs don’t develop AI tools out of cultural philanthropy. Their business model relies on data capture, creating dependencies, locking down ecosystems. Each time an institution uses their services, it potentially transmits sensitive data to them, trains its team on their interfaces, anchors its practices in their logics. This dependence can become problematic when conditions change, prices increase, services evolve.

There’s an alternative: developing skills with open source tools, collaborating with public structures that create sovereign infrastructures, inventing uses that escape dominant commercial logics. This path requires more effort initially, but it’s capable of building lasting autonomy, real awareness of these tools, from concrete uses. Cultural institutions could play a historic role by experimenting with emancipatory uses of AI rather than reproducing patterns imposed by tech giants.

Yochai Benkler showed in The Wealth of Networks (2006) how digital technologies can serve either the concentration of power or its distribution. With generative AI, this choice arises even more acutely. Cultural actors who develop their own creative protocols today are building the foundations of future digital cultural sovereignty.

Cultural nourishment rather than invisibility

There’s a common illusion among “engaged” cultural actors: believing one can ignore AI without consequences. Some institutions think they can wait, observe, let others experiment. Meanwhile, AI models feed on cultural data, learn from existing creations, integrate artistic productions into their training databases. What is not defined potentially disappears into the undifferentiated flow of data. If we don’t actively concern ourselves with it, we fall behind the world, and we harm those we supervise and the audiences we welcome, toward whom we have duties.

Creators and institutions who actively work with AI contribute to defining what it becomes. They feed databases with their aesthetic choices, their ethical positions, their formal innovations. Those who remain passive see their creations absorbed without being able to guide their use. The difference between active participation and passive observation could determine what culture AI will reproduce and amplify tomorrow.

In the musical field for example, composers currently using AI to develop new forms of improvisation or collective composition create precedents that will influence future models. Those who refuse this exploration let tech giants alone define what AI-assisted musical creativity means. This choice could have lasting consequences on what future generations will consider possible or desirable in terms of creation.

Immediate action rather than sterile deliberation

How many cultural structures have created working groups, seminars on AI without ever taking action? How many accumulate reports that sleep in drawers while other institutions are already transforming their practices? The difference between those who advance and those who stagnate might be summed up in one word: experimentation. Not vague experimentation, but defined, framed, methodical experimentation.

An effective action-training could begin with a half-day where participants don’t just talk about AI but create with it. They would discover environmental, economic and political issues not through abstract presentations but through concrete creative gestures. They would keep traces of their experiences, share their discoveries, collectively build an embodied understanding. This playful and lively approach would join the cooperative pedagogies developed by Élise and Célestin Freinet, where learning happens through collective experimentation.

Then would come workshops to map present skills, explore personal desires, build a collective vision, all the richer for being respectful of diversity of viewpoints. But beware: these workshops must not become new occasions to deliberate indefinitely. They must serve to define precise actions that each person commits to undertaking in the following weeks. Tim Ingold emphasizes in Making (2013) the importance of attention to processes, while reminding us that processes without concrete results risk remaining without effect.

Sharing experiences can constitute a solid defense against the many risks that poor uses of AI contain. Facing geopolitical issues, questions of control, dangers of dependence, I believe it’s essential to dialogue collectively, and this beyond preconceptions, from lived, real experiences. But these approaches and dialogues must not remain confined to expert circles. Speculative fiction writing, artistic creation, public debates, playful experimentation, all these means allow us to explore technologies without falling into technophobia or naive enthusiasm. Enlightened action replaces passive waiting.

Defined experimentation as a possible path of transformation

We have thus explored several principles:

• The principle of precise definition against vague intuition.
• The principle of protocol against dispersion.
• The principle of defined alliance against paralyzing fear.
• The principle of invention against reproduction.
• The principle of cultural nourishment against future invisibility.
• The principle of immediate action against sterile deliberation.

These principles are not abstract theories, but observations drawn from experiences I’ve conducted. Those who entered with precise intentions emerged transformed, carriers of concrete projects, aware of possibilities and limits. Those who entered with vague questions emerged with more precise questions, certainly, but without profound transformation of their practices.

The formation of critical thinking might not happen through theoretical discourse but through strong, personal, autonomous experiences. Participants would progress from AI-assisted writing to image generation, then to final assembly, while always preserving the human dimension through narrative voice. This methodology would allow development of an embodied understanding that seems more effective than any PowerPoint presentation. Paulo Freire affirmed it in Pedagogy of Autonomy (1996): no one educates anyone else, no one educates themselves alone, human beings educate themselves together through the intermediary of the world.

The question I now propose we ask is: Is this text simply an inspiring read, before returning to usual deliberations? Or could it stimulate the definition of a precise experimentation that would be launched within the next two weeks? The cultural actors who transform their practices might not be those who have the biggest budgets, the best prior training or the most influential connections. They might be those who precisely formulate what they want to experiment with, then take action.

Cultural institutions could bear the responsibility of creating experimentation spaces where ethical and creative uses of AI are collectively constructed. This methodological position stems neither from naive enthusiasm nor defensive rejection. It would recognize the anthropological scope of this transformation and choose to participate actively rather than suffer it passively. Cultural actors could thus become democratic laboratories where new forms of collaboration between humans and machines are invented.

The question is perhaps no longer whether AI will transform the cultural sector. It’s already transforming it. The question is rather who will participate in this transformation by shaping it according to their values, and who will suffer it as a spectator. Between these two positions, the difference might not depend on chance, not on luck, nor on resources. It might depend on only one thing: the precise definition of what one wants to do, followed by action to realize it. This principle seems to apply today as it applied yesterday, as it will probably apply tomorrow. Each must decide whether to make use of it.