22 September 2026 · René de Baaij
The Difference Between Artificial and Creative. And Why I'd Rather Not Have Known.
I am writing this piece with a language model open beside me. That is the joke, and at once the heart of the story. For years I have been advising leaders to examine the undercurrent beneath their own behaviour, and meanwhile I let a machine help shape my sentences without ever asking what that machine feels while doing so. The answer is simple. Nothing. And that nothing is precisely what this piece is about.
The psychologist Margaret Boden distinguishes three forms of creativity. Combinational: joining existing ideas in a new way. Exploratory: probing the limits within an existing frame of thought and using the space it offers. Transformational: breaking the frame itself, so that something becomes thinkable that was unthinkable before (Boden, via Popova, The Marginalian, 22 August 2025). Generative AI is remarkably good at the first two. It combines fluently, and it explores within a style with a precision I rarely achieve as a writer. The third type, breaking the frame itself, remains structurally out of its reach (ESCP, The Choice, 20 June 2023). Not for lack of computing power. Because there is nobody at home for whom the frame matters.
The undercurrent beneath my own enthusiasm
The tension does not lie in the technology. That is entertaining. The tension lies in what I do with its fluency. AI is never uncertain. It does not hesitate before answering, does not rewrite out of embarrassment, does not wonder whether something fits who it is. I do. And it is exactly that doubt, the uncomfortable moment just before a thought takes shape, where the real work happens for me. Boden calls it the emotional investment that precedes human creativity: something matters because we are vulnerable beings with something to lose. A language model has nothing to lose. It generates the most probable next word, not the word that hurts because it is true.
That is where the temptation lies for me as an adviser. Fluency feels like insight. A quick, coherent answer feels like a breakthrough, while it is often no more than a rearrangement of what was already known. Franceschelli and Musolesi add nuance here: they show that language models do produce combinational and sometimes surprisingly exploratory work, and that the boundary is less sharp than Boden's typology suggests (Franceschelli & Musolesi, AI & Society, 2024). I take that nuance seriously. But in their analysis too, the transformational, the moment a frame breaks, remains a human event. For me, then, the distinction is not human good, machine bad. It is more precise than that. Two different kinds of work, with a different source.
A silence no model can produce
Last year I was in a session with a management team that had been working on a reorganization issue for an hour. The arguments were exhausted. Everyone had made their point three times in three different ways. Then came a silence of perhaps twenty seconds, which in a meeting feels like two minutes. Nobody filled it. And out of that silence came a sentence from the finance director that reframed the entire issue: the structure was not the problem; the problem was that nobody in the room was willing to let their own unit become smaller. That was not a combination of earlier remarks. It was a break in the frame, born of a discomfort nobody had gone looking for.
Compare that with how I often start a blog or a proposal myself. I ask a language model for five angles on a theme, choose the best combination and work it out. That is useful. It saves me the going round in circles I used to do alone. But it never produces those twenty seconds of silence. It produces the average of what has already been thought, excellently phrased. Both kinds of work have their place. The problem only arises when I confuse them, and mistake the second for the first because it is quicker and sounds better.
What this asks of my own leadership
For myself, this means a discipline, not a ban. I use the model without hesitation for the combinational and the exploratory: organizing research, setting out a first structure, phrasing a counter-argument more sharply than I would myself. What I no longer allow myself is to let the model share in a decision at the moment when it ought to hurt. A strategic choice that affects a colleague. A judgement about my own failure. A sentence in which I have to admit that I misjudged something. Those moments call for no fluent answer. They call for the doubt that belongs to me, and that I must not outsource to something that never has to live with the consequences.
With regard to others, the task is different, and perhaps heavier. I see in teams that AI-polished input makes a conversation seem faster while the undercurrent remains untouched. A proposal that looks flawless discourages the very question it ought to provoke. Whoever leads therefore protects not the speed of the output but the room for the hesitation that comes before it. That sometimes means asking quite literally: which part of this piece comes from you, and which part only sounds smooth? Not out of distrust, but because a team that never has to doubt again will never think transformationally again either. It becomes very good at rearranging yesterday.
The question that remains
I am not sure whether I would have written this piece as it now stands without the model beside me. Probably not, and that is exactly why I mention it rather than hide it. The question I leave you with is not about technology. When did you last have a thought that surprised yourself: not your team, not a language model, but you? And what would have to change in your diary to give that surprise a chance again?
Sources
- Margaret Boden, discussed by Maria Popova, "Decoding the Mystery of Intuition: AI Prophet Margaret Boden on the Three Elements of Creativity," The Marginalian, 22 August 2025. themarginalian.org
- "AI can replicate human creativity in two key ways – but falls apart when asked to produce something truly new," The Choice by ESCP (originally The Conversation), 20 June 2023. escp.eu
- Giorgio Franceschelli & Mirco Musolesi, "On the creativity of large language models," AI & Society, 2024. link.springer.com