I recently moderated a panel on AI and creativity at the INSEAD AI Forum Europe in Paris. Station F, where the event was held, has two dinosaur sculptures sitting above the main hall. When I walked in that morning, I found myself thinking about “Jurassic Park”, specifically the three types of characters in the film. There is the John Hammond type, who will build the dinosaurs regardless of whether they eat society. There is the Ian Malcolm type, who thinks we should not be doing any of this. And then there are the people in the middle: terrified of the dinosaurs, but also desperately excited to touch them. That range of emotions feels like an accurate map of where we are with AI.

I was joined by Samuel Demont, VP of AI Transformation at LVMH; Gérard Assayag, Research Director at IRCAM and Principal Investigator of the European Research Council’s REACH project on co-creativity in music; and Axel Unger, CEO of Kaiser X Labs. Between them, they represent a wide breadth of creative industries, from music to fashion and design. 

We started with a “true or false” statement for the audience: AI will increase the total amount of creativity in the world. The room was split almost fifty-fifty. The case for “true” was supported by Demont, as he described young fashion designers whose “idea boxes” – collections of half-formed inspirations they aren’t ready to use yet – are more full than ever. Unger argued that we’re living through the most creatively exciting era he’s experienced, and that AI is part of what’s making it possible. More people can make more things with fewer technical barriers than at any point in history.

Assayag pushed back, asking whether humans are more creatively proficient now than we were a thousand years ago, given that art has existed since the caves. The answer, he said, is no. The human brain hasn’t changed. What changes is the form that creative expression takes, and the meaning we attach to the word “creativity” itself. Additionally, more output isn’t the same as more creativity. It’s an important distinction that I’d argue gets lost when we frame AI’s success in the creative domain in terms of volume.

More friction, please

Creativity has always involved friction: the struggle, the failure, the moment when it seems all is lost. AI is specifically designed to remove friction. So, what happens to creative work when the struggle isn’t there?

Unger stated his biggest fear about AI isn’t the technology itself but human laziness – our unwillingness to do the hard work. Picasso learnt to paint like Raphael before he invented cubism, while Thom Yorke and Jonny Greenwood of Radiohead both classically trained before they developed their signature alt-rock sound. Virtuosity in anything, whether in art, music, entrepreneurship or marketing, comes from struggling and failing until you reach mastery – only then can you make something completely different. This process didn’t have a shortcut until the arrival of AI. 

Assayag brought in Margaret Boden’s taxonomy of creativity: combinatorial creativity, where you recombine existing things in new configurations; exploratory creativity, where you find new paths within an existing mental map; and transformational creativity, where you remap the mental map itself. Most of us, most of the time, operate in the first two modes. The third is genuinely rare and is precisely what AI can’t do, because transformational creativity requires being embedded in the world, in history, in a specific social and sensory moment. It can’t come from a system trained on past data and cut off from embodied reality.

To illustrate this point, Assayag raised the example of Marcel Duchamp and the readymades – ordinary, mass-produced objects that the artist selected, titled and named as art. Would AI have come up with this? No, not because the idea is technically complex, but because it was born from an acute sensitivity to a specific set of social conditions. The readymade was necessary at the moment it appeared. That necessity isn’t something a large language model (LLM) can generate, because LLMs have no relationship to the real world in that sense.

We don’t respond to creative works the way we respond to the story behind them. If you tell someone that one piece of music was written by AI and another was written by a person on their deathbed and dedicated to their child, you know which one will resonate, regardless of which is technically superior. We don’t really respond to the artefact but to the meaning it carries and the context it emerged from.

What good creative AI use looks like

What happens when creative people engage with AI on their own terms rather than accepting it as given? Assayag’s work at IRCAM provides a useful example. The AI system he’s developed for musical improvisation doesn’t look like anything in the mainstream. In his system, the prompt isn’t text but other musicians playing live. The AI listens, learns and responds within that continuous flow. What emerges isn’t a tool but something closer to a collaborator: an agent with a kind of presence that musicians can feel and play off. Neither the human nor the machine is fully accountable for what comes out. That ambiguity, Assayag suggested, isn’t a flaw, but the most interesting thing about it.

Unger has taken a similar approach in his own practice by designing friction into the tool. He works with a virtual team of AI collaborators with distinct personalities – one that never gives an answer but only asks questions, another that plays devil’s advocate, and so on. He argues that because of the well-documented tendency of LLMs to agree with you, you have to make a conscious, sustained effort to work around it. 

Demont then made an observation that cut to the heart of the risks. When creative people take AI and build something genuinely custom for their own practice, the results tend to go in directions that surprise even the people who built the underlying technology. But when people reach for general-purpose LLMs off the shelf, the danger is different: people become a little less creative and settle for a flat AI answer.

The question worth asking

The audience exercise at the end of the session asked which future worried people most: AI making creative work economically worth less, AI making average work so good that people stop developing mastery, or AI becoming the primary creator while humans become curators. Again, the room was split evenly across the three options. 

Assayag’s answer was that none of those futures quite captures what’s actually possible. What he’s observed at IRCAM is that the interaction between human and machine can produce something that goes beyond what either could achieve alone – a collaboration where genuinely new things emerge. But getting there requires building AI differently, with intentionality about what you want the interaction to achieve, and with humans at the helm.

Unger argued for what he called “humans in the lead, not humans in the loop”. The latter positions people as a checkpoint in someone else’s process. The former insists on agency from the start. The difference between the two determines whether the friction that produces genuine creative work gets preserved or designed away.

The ultimate question, which remained unanswered, isn’t whether AI will increase the total amount of creativity in the world. It’s whether that output will carry the meaning that comes from struggle, and if we will still do the hard work required to make any of it matter.