AI Matches and Surpasses Average Human Creativity
A groundbreaking study by researchers from the Université de Montréal, in collaboration with major institutions including Concordia University and the University of Toronto, reveals that generative AI systems are now capable of outperforming the average human in specific creativity tests. The study, published in Scientific Reports, highlights a pivotal moment in artificial intelligence development: AI models like GPT-4 are not only mimicking creativity but occasionally exceeding the imaginative output of most people.
Despite these impressive advances, the findings also underscore a critical limitation. While AI may rival the average person, the most imaginative individuals—especially those in the top 10% of creativity—still produce work that AI cannot match. This distinction is most evident in rich, expressive tasks like poetry and storytelling, where emotional nuance and depth remain uniquely human.
Largest Comparative Study of AI and Human Creativity
Led by Professor Karim Jerbi from the Department of Psychology at the Université de Montréal and featuring contributions from AI pioneer Yoshua Bengio, this research marks the largest direct comparison of human and AI creativity to date. Over 100,000 human participants were tested alongside leading language models such as ChatGPT, Claude, and Gemini.
“Some AI systems based on large language models can now outperform average human creativity on well-defined tasks,” said Professor Jerbi. “This may be surprising—even unsettling—but our study also shows that top human creators still have the edge.”
Co-first authors Antoine Bellemare-Pépin and François Lespinasse conducted detailed analyses, revealing that while AI can outperform the average, it lags significantly behind the creative elite. In fact, the top 50% of human participants consistently outscored all AI systems tested. The gap widened even further among the top 10% of human performers.
How Creativity Was Measured
To evaluate creativity across both humans and AI fairly, researchers employed the Divergent Association Task (DAT), a psychological tool designed to assess divergent thinking. Originally developed by co-author Jay Olson, the DAT challenges participants to list ten words that are as semantically unrelated as possible.
For example, a highly creative response might include words like “galaxy, fork, freedom, algae, harmonica, quantum, nostalgia, velvet, hurricane, photosynthesis.” This kind of output reflects the ability to generate broad and original associations—a key indicator of creative thinking.
Importantly, the DAT is efficient and accessible, taking just two to four minutes to complete online. Its concise format makes it ideal for large-scale studies involving both humans and AI models.
Beyond Tests: Real-World Creative Tasks
Researchers extended their testing to more complex creative activities, including composing haikus, summarizing movie plots, and writing short stories. In these domains, the pattern held true: while AI could produce competent and sometimes impressive results, it fell short when compared to the most skilled human creators.
This suggests that while AI can generate content with surface-level creativity, deeper and more emotionally resonant work remains a human stronghold.
Can AI Creativity Be Enhanced?
The study also explored whether AI creativity is fixed or adjustable. Researchers discovered that creativity in AI systems can be influenced by tweaking technical settings such as “temperature,” a parameter that controls the randomness of AI-generated responses. Higher temperatures lead to more adventurous and unexpected outputs, while lower settings produce safer, more conventional results.
Additionally, the way prompts are written significantly impacts AI performance. Instructions that encourage etymological thinking or unusual associations tended to produce more creative responses. This highlights the role of human guidance in shaping AI output, emphasizing that prompt engineering is essential for unlocking generative potential.
Implications for the Future of Human Creativity
While the findings may stoke fears of AI replacing creative professionals, the researchers suggest a more nuanced future. Rather than supplanting human ingenuity, AI is likely to become a powerful tool that augments it. “Generative AI has become an extremely powerful tool in the service of human creativity,” said Jerbi. “It will not replace creators, but it will transform how they imagine, explore, and create.”
This collaborative future envisions AI as a creative assistant—one that expands possibilities and helps individuals explore new ideas, rather than replacing the imaginative spark that defines humanity.
About the Study
The paper, titled “Divergent creativity in humans and large language models,” was published in Scientific Reports on January 21, 2026. The research team included scientists from Université de Montréal, Concordia University, the University of Toronto Mississauga, Mila (Quebec AI Institute), and Google DeepMind. Professor Karim Jerbi served as lead author alongside co-first authors Antoine Bellemare-Pépin and François Lespinasse. Notably, Yoshua Bengio, a pioneer of deep learning, was also part of the team.
This article is inspired by content from Original Source. It has been rephrased for originality. Images are credited to the original source.