Creativity has long been considered one of the last bastions of human intelligence. Imagining, connecting unrelated ideas, and producing originality seemed beyond the reach of machines. However, a large-scale scientific study, published in early 2026 in *Scientific Reports* (Nature group), has turned this assumption on its head. By comparing the creativity of more than 100,000 human participants with that of the best generative artificial intelligence models, the researchers show that AI has crossed a symbolic threshold—that of average human creativity—while still lagging significantly behind the most creative individuals.
A study of unprecedented scope
This research, led by Professor Karim Jerbi (University of Montreal), with the participation of researchers from Concordia University, the University of Toronto, Mila, and Google DeepMind, is the largest comparative study to date on human and artificial creativity1. Among the co-authors is Yoshua Bengio, a global pioneer in deep learning.
The researchers evaluated several large language models, including GPT-4, Claude, and Gemini, by comparing them to a large sample of human participants recruited online. The goal was not to make a subjective judgment about creativity, but to rely on established psychometric tools that have been used for decades in cognitive psychology.
When AI surpasses average human creativity
The results mark a turning point. In linguistic divergent creativity tasks, certain AI models—particularly GPT-4—achieve scores higher than the average of human participants. In other words, on well-defined exercises, AI is now capable of generating associations of ideas that are considered more original and varied than those of the average human.
This observation is far from trivial. It suggests that generative AI systems no longer merely reproduce existing patterns, but can explore broad semantic spaces—a central feature of creative thinking2.
The most creative people maintain a clear lead
However, the study strongly qualifies any notion that human creativity has been surpassed overall. When looking at the most creative half of the participants, their performance remains superior to that of all the AI models tested. The gap becomes even more pronounced for the top 10% of the most creative individuals, who far outperform artificial systems.
This hierarchy reveals a fundamental truth. While AI today matches average human creativity, it cannot rival the highest forms of creativity—those that rely on a complex combination of experience, intuition, sensitivity, and cultural context.
How to Measure Creativity: Humans and AI on Equal Footing
To make this comparison, the researchers used the divergent association test, a widely validated tool in psychology. The principle is simple: participants—whether human or AI—are asked to generate ten words that are as different from one another as possible, starting from the same prompt. The further apart the words are semantically, the higher the creativity score.
This test, developed by researcher Jay Olson, has two major advantages. It is quick—taking just two to four minutes to complete—and its results correlate with other established creativity tests used in writing, idea generation, and creative problem-solving3. In other words, even though it is language-based, it taps into general cognitive mechanisms of creativity that go far beyond mere vocabulary.
More complex creative tasks confirm the gap
To take things a step further, the team also evaluated humans and AI on tasks more closely related to actual creative work, such as writing haikus, movie synopses, and short stories. Once again, the results confirm the observed trend. While AI systems can, in some cases, surpass average human creativity, the works deemed the most rich, coherent, and original are still predominantly produced by humans.
These experiments show that AI creativity performs well within well-defined frameworks, but that it still struggles to match the narrative depth and overall coherence of the most accomplished human works.
Adaptable artificial creativity
A key finding of the study concerns the plasticity of AI creativity. The researchers show that the creative performance of the models varies significantly depending on their settings. The temperature parameter, which controls the degree of randomness in the responses, plays a central role. At low temperatures, the AI produces cautious and predictable responses. At higher temperatures, it takes more risks and generates more original associations.
Similarly, the way instructions are phrased has a significant impact on the results. Prompting strategies based on word etymology or structure significantly improve creativity scores. These observations highlight a key point: AI creativity is not autonomous; it depends heavily on how humans guide it4.
Ethical Issues and the Future of Creative Professions
These findings inevitably reignite the debate over the potential replacement of human creators. The study, however, urges us to move beyond a strictly competitive perspective. While AI can rival average creativity in certain tasks, it does not eliminate human originality or the value of lived experience. Instead, the authors advocate for a collaborative approach, in which AI becomes a tool for creative exploration, capable of expanding the realm of possibilities without replacing human intent.
This prospect, however, raises ethical questions. As AI advances, how can we recognize and value human contributions? How can we avoid the standardization of creativity, dictated by models trained on existing datasets? Perhaps the central question is no longer whether AI is creative, but how to redefine creativity in the age of machines5.
A Redefinition of Creativity Is Underway
Ultimately, this study marks an important milestone. It shows that artificial intelligence has crossed a symbolic threshold, without, however, supplanting human creativity in its most unique form. The future of creation does not seem to lie in a head-on clash between humans and machines, but in new forms of co-creation, where AI acts as an amplifier of ideas, rather than their ultimate source.
Learn more
This exploration of artificial creativity is part of a broader debate on the cognitive capabilities of generative AI systems. On a related topic, check out our article“Vibe hacking: when users manipulate the behavior of generative AI, ” which analyzes the mechanisms of co-creation between humans and machines, as well as the limitations and potential misuse of creative models.
References
1. Jerbi, K. et al. (2026). Divergent creativity in humans and large language models. Scientific Reports (Nature).
https://www.nature.com/articles/s41598-025-25157-3
2. Runco, M. A. & Jaeger, G. J. (2012). The standard definition of creativity. *Creativity Research Journal*.https://doi.org/10.1080/10400419.2012.650092
3. Olson, J. A. et al. (2021). Measuring creativity with divergent association tasks. Psychological Assessment.
https://doi.org/10.1037/pas0000971
4. OpenAI. (2024). Temperature and sampling in large language models. Technical documentation.
https://platform.openai.com/docs
5. Boden, M. A. (2016). AI and creativity: Human and computational perspectives. Artificial Intelligence.
https://doi.org/10.1016/j.artint.2016.01.001
