A landmark study spearheaded by Professor Karim Jerbi from the Department of Psychology at the Université de Montréal, featuring the insightful participation of world-renowned AI pioneer Yoshua Bengio, has delivered unprecedented clarity on the capabilities of generative artificial intelligence systems like ChatGPT in the realm of original idea generation. This comprehensive research represents the largest direct comparison ever undertaken between human creativity and the creative output of large language models (LLMs), offering a nuanced perspective on a topic of intense public and scientific debate.
Published in the esteemed journal Scientific Reports, part of the Nature Portfolio, the findings indicate a pivotal shift in the landscape of AI capabilities. Generative AI systems have demonstrably advanced to a stage where they can now surpass the average human on specific metrics of creativity. Concurrently, the study rigorously confirms that the most exceptionally creative individuals continue to exhibit a distinct and consistent advantage, outperforming even the most sophisticated AI models tested. This dual discovery reshapes our understanding of AI’s creative potential and its interplay with human ingenuity.
AI Ascends to Average Human Creativity Levels
The research meticulously evaluated several prominent large language models, including but not limited to OpenAI’s ChatGPT and GPT-4, Google’s Gemini, and Anthropic’s Claude. Their performances were benchmarked against a massive dataset derived from over 100,000 human participants, creating a robust statistical foundation for comparison. The results unequivocally point to a significant turning point in AI development: certain advanced AI systems, notably GPT-4, successfully exceeded the average human scores on tasks specifically engineered to gauge divergent linguistic creativity.
"Our study definitively demonstrates that some AI systems, powered by large language models, are now capable of outperforming average human creativity on well-defined tasks," explained Professor Karim Jerbi, lead author of the study and an associate professor at Mila – Quebec AI Institute. He acknowledged the potential for surprise, even apprehension, that such a finding might elicit. However, he was quick to underscore an equally critical observation: "Despite these advancements, our study also highlights that even the most advanced AI systems currently fall short of the levels consistently achieved by the most creative humans." This statement frames the findings not as a competition won by AI, but as a new benchmark in its evolving capabilities.
Further in-depth analysis, conducted by the study’s co-first authors, postdoctoral researcher Antoine Bellemare-Pépin from Université de Montréal and PhD candidate François Lespinasse from Concordia University, illuminated a compelling pattern. While the average person’s creative output can now be matched or even surpassed by certain AI models, the zenith of creative expression remains firmly within the human domain. Specifically, when the researchers isolated the most creative half of the human participants, their average scores consistently eclipsed those of every AI model subjected to testing. This gap became even more pronounced and statistically significant when focusing on the top 10 percent of the most creative individuals, illustrating a distinct upper echelon of human creativity that AI has yet to penetrate.
Professor Jerbi elaborated on the methodology: "In collaboration with Jay Olson from the University of Toronto, we painstakingly developed a rigorous framework that enabled us to directly compare human and AI creativity using identical tools and metrics. This framework, underpinned by data from over 100,000 participants, ensures the validity and reliability of our comparative analysis." This emphasis on a unified evaluation standard for both biological and artificial intelligence was critical to the study’s groundbreaking nature.
A Deep Dive into Measuring Creativity: The Divergent Association Task
To ensure a fair and equitable assessment of creativity across both human and machine intelligence, the research team employed a multi-faceted approach. The cornerstone of their evaluation was the Divergent Association Task (DAT), a well-established psychological test widely recognized for its efficacy in measuring divergent creativity. This form of creativity refers to an individual’s capacity to generate a diverse array of novel and original ideas in response to a single prompt, moving beyond conventional thought patterns.
The DAT, originally conceived by study co-author Jay Olson, is deceptively simple yet profoundly revealing. It challenges participants, whether human or AI, to list ten words that are as semantically unrelated as possible. The goal is to maximize the conceptual distance between each chosen word. As an illustrative example of a highly creative response, the study cited a list such as "galaxy, fork, freedom, algae, harmonica, quantum, nostalgia, velvet, hurricane, photosynthesis." Such a combination demonstrates a wide-ranging cognitive search and an ability to connect disparate concepts, hallmarks of divergent thinking.
The selection of the DAT was not arbitrary. Performance on this task has been empirically linked to strong results on various other established creativity assessments, including those used in creative writing, idea generation workshops, and complex problem-solving scenarios. While the task is inherently language-based, its demands extend far beyond mere vocabulary recall. It actively engages broader cognitive processes central to creative thinking across a multitude of domains, making it an ideal measure for a broad comparison. Furthermore, the DAT offers practical advantages, requiring only two to four minutes to complete and being easily deployable online, which facilitated the collection of data from such a vast human participant pool.
Beyond Word Lists: AI’s Foray into Complex Creative Writing
Recognizing that the DAT primarily assesses a foundational aspect of linguistic creativity, the researchers extended their investigation to ascertain whether AI’s success on this relatively simple word association task could translate to more intricate and realistic creative activities. To address this, they pitted AI systems against human participants in a series of creative writing challenges. These included the composition of haiku, the ancient Japanese poetic form comprising three lines with a 5, 7, 5 syllable structure, the crafting of compelling movie plot summaries, and the generation of short stories.
The results from these advanced creative tasks largely mirrored the pattern observed with the DAT. While AI systems demonstrated the capacity to occasionally exceed the performance of average human writers in these domains, the most skilled and imaginative human creators consistently produced work that was not only stronger in narrative quality and emotional resonance but also exhibited a higher degree of originality and conceptual depth. This reinforced the notion that while AI can mimic and synthesize, genuine groundbreaking creativity, particularly in complex narrative structures, remains a human forte.
The Modulable Nature of AI Creativity: Temperature and Prompt Engineering
These compelling findings naturally prompted another critical inquiry: Is AI creativity an immutable characteristic, or can it be deliberately shaped and adjusted? The study provided clear evidence that creativity in AI systems is, in fact, highly modulable. This adjustability is primarily achieved by altering technical settings, most notably the model’s "temperature" parameter. This parameter effectively dictates the degree of predictability versus adventurousness in the AI’s generated responses.
At lower temperature settings, AI models tend to produce outputs that are safer, more conventional, and closely aligned with the most probable linguistic sequences found in their training data. This leads to predictable, often generic, creative attempts. Conversely, when the temperature is increased, the AI’s responses become significantly more varied, less predictable, and markedly more exploratory. This allows the system to deviate from familiar patterns and venture into less common, more novel conceptual spaces, thereby boosting its apparent creativity scores.
Beyond internal parameters, the researchers also discovered that AI creativity is profoundly influenced by the precision and intent embedded within the input instructions, a practice known as "prompt engineering." For instance, prompts that explicitly encouraged models to delve into the etymological origins and structural nuances of words led to a greater number of unexpected associations and, consequently, higher creativity scores on the DAT. These results underscore a fundamental truth: AI creativity, far from being autonomous, remains heavily dependent on sophisticated human guidance. This makes the art of interaction and prompt formulation a central, indispensable component of the AI-assisted creative process.
Historical Context: The Evolution of AI and the Creativity Question
The debate surrounding AI’s capacity for creativity is not new, but it has gained unprecedented urgency with the advent of generative AI. For decades, AI systems were largely confined to rule-based logic and pattern recognition, struggling with tasks requiring genuine innovation or imagination. Early attempts at creative AI were often limited to generating variations on existing themes or following strict algorithmic constraints.
The breakthrough came with the development of deep learning, particularly transformer architectures in the late 2010s, which paved the way for large language models. Trained on colossal datasets of text and code, these models learned to identify complex patterns, understand context, and generate coherent, human-like text. The public release of models like GPT-3 and later ChatGPT in the early 2020s sparked widespread fascination and alarm. Suddenly, AI could write poetry, compose music, and generate unique images, leading many to question the very definition of creativity and humanity’s unique claim to it.
Prior to the Université de Montréal study, many comparisons were anecdotal or based on smaller, less rigorous experiments. What sets Professor Jerbi’s research apart is its "unprecedented scale" – over 100,000 human participants – and its "direct comparison" methodology, employing the same standardized tools for both humans and machines. The involvement of Yoshua Bengio, a Turing Award laureate and one of the "Godfathers of AI" for his foundational work in deep learning, further solidifies the study’s credibility and highlights its significance in the broader AI research community. This study thus provides a robust, data-driven answer to questions that have long been speculative, offering a definitive snapshot of current AI creative capabilities.
Will AI Replace Human Creators? A Balanced Perspective
The study offers a thoughtfully balanced perspective on the pervasive fears that the rapid advancement of artificial intelligence could inevitably lead to the replacement of creative professionals across various industries. While it is now clear that AI systems can indeed match or even exceed average human creativity on specific, well-defined tasks, the research unequivocally highlights their persistent limitations and their inherent reliance on human direction and refinement.
"Even though AI can now achieve human-level creativity on certain tests, we must move beyond this often misleading sense of competition," urged Professor Karim Jerbi. He articulated a vision where generative AI is not a competitor but an enabler: "Generative AI has, above all, emerged as an extraordinarily powerful tool in the service of human creativity. It is not poised to replace creators but rather to profoundly transform how they imagine, explore, and bring their ideas to fruition – for those who choose to leverage its capabilities."
Rather than portending the obsolescence of creative careers, the findings strongly suggest a future where AI functions as an indispensable creative assistant. By efficiently expanding the range of initial ideas, generating diverse permutations, and opening up entirely new avenues for creative exploration, AI has the potential to significantly amplify human imagination, rather than merely replicate or supplant it. For artists, writers, designers, and innovators, AI could become a powerful co-pilot, handling routine generation while freeing human minds for higher-level conceptualization, emotional depth, and truly original leaps of thought.
The implications extend beyond individual creators to entire industries. From marketing and advertising to game design and architectural rendering, AI tools are already streamlining processes, generating prototypes, and personalizing content at scales previously unimaginable. This shift necessitates a re-evaluation of educational curricula for creative fields, emphasizing prompt engineering, critical evaluation of AI outputs, and the unique human touch that AI cannot yet replicate. Ethical considerations, such as intellectual property rights for AI-generated content and the potential for creative monopolies, will also become increasingly central to public discourse.
"By directly confronting and rigorously comparing human and machine capabilities, studies like ours compel us to critically rethink and redefine what we truly mean by creativity," concluded Professor Karim Jerbi. This ongoing dialogue is crucial as humanity navigates the complexities and opportunities presented by increasingly intelligent machines.
About the Study: A Collaborative Endeavor
The seminal paper, titled "Divergent creativity in humans and large language models," was officially published in Scientific Reports on January 21, 2026. This comprehensive research was the product of a collaborative effort, bringing together leading scientists and institutions from across the academic and AI landscape. Participating entities included the Université de Montréal, Concordia University, the University of Toronto Mississauga, Mila (Quebec AI Institute), and Google DeepMind, representing a confluence of expertise in psychology, artificial intelligence, and cognitive science.
Professor Karim Jerbi served as the principal investigator and lead author, guiding the multidisciplinary team. Antoine Bellemare-Pépin from Université de Montréal and François Lespinasse from Concordia University were recognized as co-first authors, acknowledging their significant contributions to the research design, data analysis, and interpretation of the findings. The research team was further bolstered by the involvement of Yoshua Bengio, the esteemed founder of Mila and LoiZéiro, whose pioneering work in deep learning laid much of the theoretical and practical groundwork for modern AI systems like ChatGPT. His participation underscores the study’s foundational importance in the field of artificial intelligence research.
