What the study found

The article "Divergent Creativity in Humans and Large Language Models" was published in Scientific Reports — part of the Nature Portfolio — on January 21, 2026. The study represents an exceptional collaboration among researchers from Université de Montréal, Université Concordia, University of Toronto Mississauga, Mila (Quebec AI Institute), and Google DeepMind. The research was led by Professor Karim Jerbi, and the author team also includes Yoshua Bengio — founder of Mila and one of the world's leading pioneers of deep learning, the technology underpinning modern AI systems such as ChatGPT.

The study tested the creativity of several large language models — including ChatGPT, Claude, Gemini, and others — and compared their performance with that of 100,000 human participants. The results showed that some advanced AI systems, including GPT-4, now overtake the average human score on specific measures of linguistic creativity. Before this study, AI creativity was always seen as inferior to human originality. The new evidence suggests that, at least on structured tasks, AI has crossed a threshold that has always been unique to humans. ‍

"Our study shows that some AI systems based on large language models can now outperform average human creativity on well-defined tasks," explains Professor Karim Jerbi.

What the test actually measured — and why it matters

The researchers used the Divergent Association Task — a standard tool in cognitive psychology designed to measure a narrow slice of creative ability. Participants were instructed to produce ten words that are as unrelated to one another as possible. Responses are scored by measuring the semantic distance between the meanings of those words. The farther apart the meanings, the higher the creativity score.

This is the detail that changes everything about how to read the headline.

That study matters because it immediately exposes what the task can and cannot tell us. It rewards the ability to generate variation under a fixed instruction. It does not measure whether an idea is useful, persuasive, defensible, or strategically sound. It does not ask whether the idea should exist.

AI is extraordinarily good at generating variation. Given a prompt, it produces a vast spread of outputs — divergent, distant, technically unpredictable. On the specific task the study measured, this is exactly the skill being tested. And AI passed. But passing a divergent thinking test and possessing the judgment to know which idea is worth pursuing — those are entirely different capacities. The study measured the first. It cannot touch the second.

AI tends to repeat the same "safe" ideas, while people naturally vary their responses. The creative humans who scored highest did not win by being more expressive. The people who outperformed AI did not do so by being more imaginative in some mystical sense. Instead, they rejected obvious combinations and pushed beyond adequacy. They did not stop when the instruction had been met. ‍

That distinction — between producing ideas and deciding which ideas deserve to survive — is the most important line in the entire study. And it is a line every school should tape to its wall.

What the full data actually shows

Although AI models performed strongly, the most creative half of human participants outperformed every AI system tested, and the top 10% of individuals demonstrated a particularly wide advantage. This pattern suggests that while AI can match or exceed average performance, exceptional creativity remains firmly human. The results highlight creativity as a spectrum rather than a single benchmark, with machines clustering around the middle while humans dominate the upper range.

The researchers also investigated how AI models compared with humans when it came to creative writing tasks, including haikus, film synopses, and short stories. Once again, the most creative humans outperformed the machines — even if LLMs overall scored better than the average participant. And it is worth noting that the LLMs expressed the most creativity when they were guided well — by humans.

Read that last sentence again. AI is most creative when guided well by humans. Not when left alone. Not when given a blank prompt. When a person who knows what they want — who has taste, conviction, and judgment — directs the model with precision, the output improves. The human is not redundant in the creative process. The human is the variable that determines whether the AI produces something ordinary or something extraordinary.

What this means for every school building its AI strategy

The headline from this study — "AI beats humans at creativity" — will circulate widely. It already has. And the schools that read only the headline will draw the wrong conclusion in one of two directions.

The first wrong conclusion is resignation: if AI can now out-create the average human, creative education is less important. Why develop children's creative capacity if AI already does it faster? This conclusion is not supported by the data. It is contradicted by it. The top 10% of human participants left every AI model behind. The pathway to that upper range is not effortless. It is built through years of being asked to think originally, to reject the obvious, to push past the point where the instruction has been met. That is what creative education develops. And the study shows that the humans who have developed it are precisely the ones AI cannot touch.

The second wrong conclusion is complacency: because the top 10% still leads, schools do not need to change anything. This conclusion ignores what is happening to the other 90%. If average human creativity is now matched or exceeded by AI, the middle of the distribution — the students who go through school producing competent, adequate, unremarkable work — is no longer employable on the basis of competence alone. Adequate is what AI does cheaply. What remains valuable is what average schooling rarely produces: genuine originality, taste, the capacity for judgment, and the courage to decide that an idea is not good enough.

This is the reset AI Ready School's philosophy describes. The industrial model of education was designed to produce consistent, adequate outputs — students who could follow instructions, complete defined tasks, and produce work that passed a standardised bar. AI has now crossed that bar. The students who will matter in the economy this study describes are not the ones who scored above average. They are the ones whose schools gave them the conditions to develop the capacities that AI, even at its most creative, still cannot replicate.

Cypher does not ask students to produce answers. It asks them to push past the obvious — to question their first response, examine their reasoning, and arrive somewhere more interesting than where they started. This is not a philosophical preference. It is a direct response to the world the creativity study describes. The students who learn to reject adequacy — who are trained to push beyond the point where the instruction has been met — are the students who will remain irreplaceable in a world where AI handles the average.

NEO exists to put real creative problems in children's hands — problems that do not have a single correct answer, where the quality of what they produce depends not on whether they followed the right steps but on whether they made the right judgments. Robots that need to be programmed. AI tools that need to be directed. Projects that only work if the child's thinking is genuinely original. These are not extracurricular activities. They are the core curriculum of the era this study has just described.

The finding worth reading slowly

Generative AI systems have now reached a level where they can outperform the average human on certain creativity measures. At the same time, the most creative people still show a clear and consistent advantage over even the strongest AI models.

The gap between average and exceptional creativity has always existed in schools. What changes now is that the gap has consequences. Average is what AI does. Exceptional is what schools must build. The question is whether the schools your children attend have understood that — and whether they are building for it.

‍