A groundbreaking study, spearheaded by Professor Karim Jerbi from the Department of Psychology at the Université de Montréal and involving distinguished AI luminary Yoshua Bengio, has undertaken the most extensive direct comparison to date between the creative capabilities of artificial intelligence and human cognition, analyzing over 100,000 individuals. This extensive research, published in the esteemed journal Scientific Reports under the Nature Portfolio, signifies a pivotal moment in our understanding of machine intelligence, demonstrating that certain generative AI systems have advanced to a point where they can demonstrably outperform the typical human on specific metrics of creative output. Concurrently, however, the study emphatically underscores that individuals exhibiting the highest levels of creativity continue to maintain a substantial and consistent advantage over even the most sophisticated AI models currently available.
The investigation meticulously evaluated several prominent large language models, including widely recognized systems such as ChatGPT, Claude, and Gemini, alongside other advanced AI architectures. Their performance was benchmarked against the creative responses of more than 100,000 human participants. The results unequivocally signal a significant inflection point: some AI systems, notably GPT-4, have achieved scores that surpass the average human performance on tasks specifically designed to gauge divergent linguistic creativity. Professor Karim Jerbi elaborated on this finding, stating, "Our research clearly indicates that certain AI systems powered by large language models are now capable of exceeding average human creativity on precisely defined tasks." He further acknowledged the potential impact of this revelation, noting, "While this outcome might elicit surprise, perhaps even a degree of unease, our study concurrently highlights an equally critical observation: the most advanced AI systems still fall short of the creative benchmarks set by the most exceptionally creative human minds."
Further in-depth analysis, conducted by the study’s co-first authors, postdoctoral researcher Antoine Bellemare-Pépin of the Université de Montréal and PhD candidate François Lespinasse from Concordia University, illuminated a particularly striking pattern. While it is now evident that certain AI models can outperform the general population on creativity assessments, the apex of human creative achievement remains distinctly within the human domain. The researchers observed that when they focused their analysis on the most creative half of the human participants, their average scores not only surpassed those of every AI model subjected to testing but the disparity widened considerably when examining the top 10 percent of individuals identified as most creative. Professor Jerbi emphasized the robustness of their methodology, stating, "We developed a stringent framework that enabled us to compare human and AI creativity using identical evaluative tools, drawing upon data from an immense cohort of over 100,000 participants, in close collaboration with Jay Olson from the University of Toronto."
To ensure a fair and comprehensive assessment of creativity across both human and artificial intelligence, the research consortium employed a multi-faceted approach. The cornerstone of their methodology was the Divergent Association Task (DAT), a well-established psychological instrument commonly used to measure divergent creativity – the capacity to generate a wide array of unique and novel ideas from a single conceptual starting point. The DAT, conceived by study co-author Jay Olson, presents participants, whether human or AI, with a singular prompt and requires them to generate a list of ten words that are as semantically distinct from each other as possible. An illustrative example of a highly creative response provided in the study features words such as "galaxy, fork, freedom, algae, harmonica, quantum, nostalgia, velvet, hurricane, photosynthesis." The study posits that performance on this specific task correlates strongly with outcomes on other recognized creativity assessments, including those related to creative writing, idea generation, and innovative problem-solving. Although the DAT is fundamentally a language-based task, its cognitive demands extend significantly beyond mere vocabulary recall, tapping into broader cognitive processes integral to creative thinking across a multitude of disciplines. A notable practical advantage of the DAT lies in its efficiency, requiring only two to four minutes to complete and its accessibility to the general public via online platforms.
Building upon the insights gleaned from this foundational word-association task, the researchers extended their investigation to ascertain whether AI’s proficiency in the DAT could translate to more intricate and realistic creative endeavors. To this end, they subjected both AI systems and human participants to a series of creative writing challenges. These included tasks such as composing haiku, a concise three-line poetic form; crafting synopses for movie plots; and generating short fictional narratives. The outcomes of these more complex tasks mirrored the pattern observed in the DAT. While AI systems occasionally achieved performance levels superior to those of the average human, the most accomplished human creators consistently produced work that was both more robust and demonstrably more original.
A crucial question arising from these findings pertains to the malleability of AI creativity: is it a fixed attribute, or can it be dynamically influenced? The study’s findings indicate that AI creativity is indeed adjustable, significantly impacted by alterations to specific technical parameters, most notably the model’s ‘temperature’ setting. This parameter functions as a control for the predictability versus the adventurousness of the AI’s generated outputs. At lower temperature settings, AI tends to produce responses that are more conservative and conventional. Conversely, higher temperature settings encourage a greater degree of variation and unpredictability in the AI’s outputs, enabling the system to explore concepts and generate ideas that deviate from well-trodden paths. Furthermore, the research team discovered that the nature of the instructions provided to the AI plays a substantial role in shaping its creative output. For instance, prompts that specifically encourage the AI to consider word origins and structural elements through etymology have been shown to elicit more unexpected associations and, consequently, higher creativity scores. These results underscore a critical dependency: AI creativity is heavily influenced by human direction, positioning human interaction and precise prompting as integral components of the creative process.
The study offers a nuanced perspective on widespread anxieties regarding the potential for artificial intelligence to supplant human creative professionals. Although AI systems have now demonstrated the capacity to match or even exceed average human creativity on certain tasks, they still exhibit discernible limitations and remain fundamentally reliant on human guidance. Professor Karim Jerbi articulated this point by stating, "Even though AI can now achieve human-level creativity on specific tests, it is imperative that we move beyond this potentially misleading sense of competition." He further posited, "Generative AI has, above all, evolved into an extraordinarily powerful instrument in service of human creativity; it will not displace creators but rather profoundly transform the very modalities through which they conceive, explore, and produce – for those who choose to leverage its capabilities." Rather than heralding the demise of creative professions, the research findings suggest a future where AI acts as a sophisticated creative collaborator. By facilitating the expansion of ideas and opening novel avenues for exploration, AI holds the potential to amplify human imagination rather than diminish it. Professor Jerbi concluded by remarking, "By directly confronting the capabilities of both humans and machines, studies such as ours compel us to re-evaluate and refine our very definition of creativity."
The foundational paper, titled "Divergent creativity in humans and large language models," was officially published in Scientific Reports on January 21, 2026. This significant research effort represents a collaborative endeavor involving esteemed scientists from the Université de Montréal, Concordia University, the University of Toronto Mississauga, Mila (the Quebec AI Institute), and Google DeepMind. The project was under the direct leadership of Professor Karim Jerbi, with Antoine Bellemare-Pépin from the Université de Montréal and François Lespinasse from Concordia University serving as co-first authors. The esteemed research team also included Yoshua Bengio, a foundational figure at Mila and LoiZéro, widely recognized as a pioneer in deep learning, the underlying technology that powers contemporary AI systems like ChatGPT.
About the Author
