Researchers from the Université de Montréal conducted what’s considered one of the largest comparative creativity studies ever, asking both humans and generative AI models to take part in standardized divergent thinking tests. In these tasks, participants were asked to list unrelated words under time pressure — a common psychological method for measuring a type of creativity where the distance between ideas reflects creative thinking. The research aimed to see how advanced language models stack up against a very large pool of human participants.
The results showed that AI — including leading generative models — often scored above the average human participant on these creativity benchmarks. However, when looking at the top performers, highly creative humans still outshone the machines. Around half of the human participants generated more divergent creative responses than the AI’s outputs, and the very best human lists were markedly more creative than what the AI could produce, highlighting that AI’s “creativity” is tied closely to the way these tests are defined.
Researchers emphasize that measuring creativity — especially in machines — is inherently tricky because creative thinking encompasses a wide range of subjective and context-dependent qualities that aren’t fully captured by simple tests. The study’s lead scientists argue that rather than framing the question as AI versus humans, it’s more useful to view AI as a tool that can serve human creativity, assisting people in generating ideas, exploring possibilities, and expanding creative expression without fully replacing human ingenuity.
In additional testing involving creative writing tasks such as short stories and haikus, the most creative humans again tended to outperform AI models, especially when human judgement, emotional nuance, and unique stylistic expression were involved. Importantly, the study also suggested that AI tends to show its strongest creative results when guided by human direction, reinforcing the idea that collaboration between people and generative AI — rather than competition — may be the most productive model for creative work going forward.
