A groundbreaking new study, spearheaded by Professor Karim Jerbi from the Department of Psychology at the Université de Montréal and featuring the insights of renowned AI pioneer Yoshua Bengio, has delivered the most extensive direct comparison to date between human creativity and that of large language models (LLMs). Published in Scientific Reports (Nature Portfolio) on January 21, 2026, the research provides a nuanced answer to the increasingly pertinent question: Can generative artificial intelligence systems like ChatGPT truly originate novel ideas? The findings indicate a significant inflection point, showing that while AI can now outperform the average human on certain measures of creativity, the most exceptionally creative individuals retain a clear and consistent advantage over even the most sophisticated AI models.
Unprecedented Scale and Rigor in Assessing AI Creativity
The scale of this investigation is unprecedented, involving the evaluation of several leading large language models—including notable systems like GPT-4, Claude 3, and Gemini 1.5 Pro—against the performance of more than 100,000 human participants. This monumental comparative dataset allows for a statistically robust and comprehensive assessment, moving beyond anecdotal evidence or smaller-scale experiments that have characterized much of the earlier discourse on AI and creativity. The sheer volume of human data, meticulously collected and analyzed, provides a solid benchmark against which the evolving capabilities of AI can be accurately measured.
Professor Karim Jerbi, the study’s lead author, underscored the dual nature of these findings. "Our study demonstrates that some AI systems based on large language models can now surpass average human creativity on well-defined tasks," he explained. "This outcome might be surprising—even unsettling for some—but our research also illuminates an equally crucial observation: even the best AI systems currently fall short of the levels achieved by the most creative humans." This statement encapsulates the core paradox of AI creativity: impressive progress yet persistent limitations at the zenith of human ingenuity.
The meticulous analysis, conducted by co-first authors postdoctoral researcher Antoine Bellemare-Pépin from the Université de Montréal and PhD candidate François Lespinasse from Concordia University, further clarified this pattern. While a subset of advanced AI models now demonstrably surpasses the average person in creative tasks, the pinnacle of creativity remains firmly within the human domain. Specifically, when researchers isolated the most creative half of the human participants, their average scores consistently outstripped those of every AI model tested. This gap widened considerably when focusing on the top 10 percent of the most creative individuals, illustrating a profound distinction that AI has yet to bridge.
"We developed a rigorous framework that enabled us to compare human and AI creativity using identical tools, leveraging data from over 100,000 participants, in collaboration with Jay Olson from the University of Toronto," Professor Jerbi, who is also an associate professor at Mila (Quebec AI Institute), elaborated. This emphasis on a standardized, large-scale methodology is critical to the study’s credibility and its potential to reshape public and scientific understanding of AI’s creative capacities.
Methodology: Measuring the Unmeasurable with the Divergent Association Task
To ensure a fair and equitable evaluation of creativity across both human and machine intelligence, the research team employed a multi-faceted approach. The cornerstone of their methodology was the Divergent Association Task (DAT), a widely recognized psychological test specifically designed to measure divergent creativity. Divergent creativity refers to an individual’s ability to generate a broad range of unique and varied ideas from a single starting point or prompt—a hallmark of innovative thinking.
The DAT, originally conceived by study co-author Jay Olson, presents a deceptively simple challenge: participants, whether human or AI, are asked to list ten words that are as semantically unrelated as possible. The power of this task lies in its ability to quantify the breadth and originality of an individual’s conceptual associations. For instance, a highly creative response might include words such as "galaxy, fork, freedom, algae, harmonica, quantum, nostalgia, velvet, hurricane, photosynthesis." The disparate nature of these terms reflects a mind capable of traversing vast conceptual landscapes, making connections where none are immediately apparent.
The efficacy of the DAT extends beyond its straightforward format. Performance on this task has been robustly correlated with results from other established creativity tests, including those used in creative writing assessments, idea generation challenges, 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 fundamental to creative thinking across a multitude of domains, from scientific discovery to artistic expression. Furthermore, the DAT offers practical advantages: it typically takes only two to four minutes to complete and is readily accessible online, facilitating large-scale data collection.
Beyond Wordplay: AI’s Performance in Complex Creative Endeavors
A crucial aspect of the study involved investigating whether AI’s success on the relatively structured word association task would translate to more complex and realistic creative activities. To this end, the researchers pitted AI systems against human participants in a series of creative writing challenges. These included composing haiku, the concise three-line poetic form demanding brevity and evocative imagery; crafting movie plot summaries, which require narrative coherence and imaginative hooks; and generating short stories, demanding character development, plot progression, and emotional resonance.
The results from these advanced tasks mirrored the pattern observed with the DAT. While AI systems occasionally managed to exceed the performance of average human writers and storytellers, the most skilled and imaginative human creators consistently produced work that was deemed superior in originality, depth, and overall creative impact. This finding reinforces the idea that while AI can capably handle the mechanics of creative output, the spark of truly exceptional, human-level originality remains largely elusive for current models.
The Modulability of AI Creativity: Temperature and Prompt Engineering
The study also delved into a pivotal question concerning AI’s creative potential: Is AI creativity an immutable characteristic, or can it be influenced and refined? The research unequivocally demonstrated that creativity in AI systems is indeed adjustable, primarily through the manipulation of technical settings, most notably the model’s "temperature" parameter.
In the context of generative AI, "temperature" is a crucial setting that controls the predictability and adventurousness of the generated responses. At lower temperature settings (e.g., 0.1-0.5), AI models tend to produce more conservative, conventional, and predictable outputs, often sticking closer to established patterns in their training data. This leads to safer, but potentially less original, results. Conversely, when the temperature is increased (e.g., 0.7-1.0), the models become more varied, less predictable, and significantly more exploratory in their output. This higher temperature allows the system to deviate further from familiar ideas, experiment with novel associations, and venture into less conventional conceptual territory, often leading to higher creativity scores.
Beyond technical parameters, the researchers also discovered that the way instructions—or "prompts"—are formulated profoundly influences AI’s creative output. For instance, prompts that explicitly encouraged models to consider the origins and structural nuances of words through etymology led to a notable increase in unexpected associations and, consequently, higher creativity scores. This highlights a critical insight: AI creativity is not an autonomous function but rather a highly interactive process, heavily dependent on human guidance and the art of prompt engineering. The quality and specificity of human input directly correlate with the creative potential unlocked in AI systems, cementing the role of human-AI collaboration in the creative process.
Implications for the Future of Human Creativity and Industry
The study offers a thoughtfully balanced perspective on the pervasive anxieties surrounding artificial intelligence’s potential to displace creative professionals. While acknowledging AI’s newfound capacity to match or even exceed average human creativity on specific, well-defined tasks, the research emphatically underscores AI’s inherent limitations and its persistent reliance on human direction.
Professor Karim Jerbi articulated a vision that transcends a zero-sum competition: "Even though AI can now reach human-level creativity on certain tests, we need to move beyond this misleading sense of competition. Generative AI has above all become an extremely powerful tool in the service of human creativity: it will not replace creators, but profoundly transform how they imagine, explore, and create—for those who choose to use it." This statement signals a paradigm shift from replacement to augmentation, positioning AI not as a rival but as a sophisticated co-pilot in the creative journey.
Rather than portending the demise of creative careers, the findings strongly suggest a future where AI functions as an invaluable creative assistant. By efficiently generating a wide array of preliminary ideas, suggesting unconventional avenues for exploration, and even handling iterative tasks, AI can significantly amplify human imagination. It can free up human creators to focus on higher-order conceptualization, refinement, and the injection of unique human insights, emotions, and cultural understanding that AI still struggles to replicate. This collaborative model could democratize creativity, empowering more individuals to explore their artistic and innovative potentials by leveraging AI as a powerful extension of their own minds.
The broader implications for various creative industries are substantial. In areas like advertising, design, content creation, and even scientific research, AI could accelerate brainstorming phases, generate diverse prototypes, and identify novel connections that human minds might overlook. For example, a graphic designer could use AI to generate hundreds of logo variations in minutes, then apply their human judgment and artistic sensibility to refine the most promising concepts. A writer might use AI to overcome writer’s block by generating plot twists or character backstories, then weave these elements into a compelling narrative with their unique voice. This evolution could lead to a surge in creative output and innovation, but also necessitates a redefinition of roles and skills within these fields.
Ethical and Philosophical Considerations
Beyond the practical applications, this study also reignites profound ethical and philosophical discussions about the very nature of creativity, originality, and authorship in the age of AI. If AI can generate ideas indistinguishable from or even superior to average human output, where does the line between machine-generated and human-inspired blur? Questions of intellectual property, copyright, and accountability for AI-generated content become increasingly complex. The involvement of AI pioneer Yoshua Bengio, founder of Mila and LoiZéiro, and a seminal figure in deep learning, underscores the gravity of these considerations. His participation reflects the deep engagement of the AI research community with these critical societal questions.
"By directly confronting human and machine capabilities, studies like ours push us to rethink what we mean by creativity," Professor Jerbi concluded. This re-evaluation is not merely academic; it has tangible implications for how society values creative work, educates future generations of artists and innovators, and constructs legal frameworks around digital creations. It forces a deeper introspection into what constitutes true originality and whether the process of creation, rather than just the output, is what fundamentally defines human ingenuity.
About the Study and Research Team
The seminal paper, titled "Divergent creativity in humans and large language models," was officially published in Scientific Reports on January 21, 2026. This collaborative research effort brought together leading scientists and institutions from across the academic and industry landscape, including the Université de Montréal, Concordia University, the University of Toronto Mississauga, Mila (Quebec AI Institute), and Google DeepMind.
Professor Karim Jerbi, a distinguished neuroscientist and expert in cognitive processes, led the study from the Université de Montréal. He was supported by co-first authors Antoine Bellemare-Pépin, a postdoctoral researcher at the Université de Montréal, and François Lespinasse, a PhD candidate at Concordia University, who were instrumental in the methodological development and data analysis. The research team also notably included Yoshua Bengio, a Turing Award laureate, founder and scientific director of Mila, and a driving force behind the deep learning revolution that underpins modern AI systems like ChatGPT. His involvement lends significant weight to the study’s findings and implications, bridging cutting-edge AI development with fundamental psychological inquiry. The collaboration with Jay Olson from the University of Toronto, creator of the Divergent Association Task, ensured the psychological validity and robustness of the primary creativity measure employed. This interdisciplinary approach underscores the complexity of understanding AI’s creative capacities, requiring expertise from psychology, computer science, and neuroscience.
