Can generative synthetic intelligence methods like ChatGPT genuinely create unique concepts? A brand new examine led by Professor Karim Jerbi from the Division of Psychology on the Université de Montréal, with participation from famend AI researcher Yoshua Bengio, takes on that query at an unprecedented scale. The analysis is the biggest direct comparability ever performed between human creativity and the creativity of enormous language fashions.
The examine, revealed in Scientific Experiences (Nature Portfolio), factors to a big shift. Generative AI methods have now reached a stage the place they’ll outperform the typical human on sure creativity measures. On the identical time, essentially the most inventive individuals nonetheless present a transparent and constant benefit over even the strongest AI fashions.
AI Reaches Common Human Creativity Ranges
Researchers evaluated a number of main giant language fashions, together with ChatGPT, Claude, Gemini, and others, and in contrast their efficiency with outcomes from greater than 100,000 human contributors. The findings spotlight a transparent turning level. Some AI methods, together with GPT-4, exceeded common human scores on duties designed to measure divergent linguistic creativity.
“Our examine exhibits that some AI methods primarily based on giant language fashions can now outperform common human creativity on well-defined duties,” explains Professor Karim Jerbi. “This end result could also be stunning — even unsettling — however our examine additionally highlights an equally necessary remark: even the perfect AI methods nonetheless fall wanting the degrees reached by essentially the most inventive people.”
Additional evaluation by the examine’s co-first authors, postdoctoral researcher Antoine Bellemare-Pépin (Université de Montréal) and PhD candidate François Lespinasse (Université Concordia), revealed a putting sample. Whereas some AI fashions now outperform the typical individual, peak creativity stays firmly human.
In reality, when researchers examined essentially the most inventive half of contributors, their common scores surpassed these of each AI mannequin examined. The hole grew even bigger among the many high 10 % of essentially the most inventive people.
“We developed a rigorous framework that enables us to check human and AI creativity utilizing the identical instruments, primarily based on knowledge from greater than 100,000 contributors, in collaboration with Jay Olson from the College of Toronto,” says Professor Karim Jerbi, who can be an affiliate professor at Mila.
How Scientists Measure Creativity in People and AI
To guage creativity pretty throughout people and machines, the analysis group used a number of strategies. The first device was the Divergent Affiliation Process (DAT), a extensively used psychological take a look at that measures divergent creativity, or the power to generate various and unique concepts from a single immediate.
Created by examine co-author Jay Olson, the DAT asks contributors, whether or not human or AI, to checklist ten phrases which can be as unrelated in that means as attainable. An instance of a extremely inventive response contains phrases like “galaxy, fork, freedom, algae, harmonica, quantum, nostalgia, velvet, hurricane, photosynthesis.”
Efficiency on this activity is strongly linked to outcomes on different established creativity assessments utilized in writing, concept technology, and inventive drawback fixing. Though the duty is language-based, it goes effectively past vocabulary. It engages broader cognitive processes concerned in inventive considering throughout many domains. The DAT additionally has sensible benefits, because it takes solely two to 4 minutes to finish and could be accessed on-line by most people.
From Phrase Lists to Actual Inventive Writing
The researchers then explored whether or not AI success on this straightforward phrase affiliation activity may lengthen to extra complicated and practical inventive actions. To check this, they in contrast AI methods and human contributors on inventive writing challenges comparable to composing haiku (a brief three-line poetic kind), writing film plot summaries, and producing quick tales.
The outcomes adopted a well-recognized sample. Whereas AI methods generally exceeded the efficiency of common people, essentially the most expert human creators constantly delivered stronger and extra unique work.
Can AI Creativity Be Adjusted?
These findings raised one other necessary query. Is AI creativity mounted, or can it’s formed? The examine exhibits that creativity in AI could be adjusted by altering technical settings, significantly the mannequin’s temperature. This parameter controls how predictable or adventurous the generated responses are.
At decrease temperature settings, AI produces safer and extra typical outputs. At greater temperatures, responses change into extra diverse, much less predictable, and extra exploratory, permitting the system to maneuver past acquainted concepts.
The researchers additionally discovered that creativity is strongly influenced by how directions are written. For instance, prompts that encourage fashions to consider phrase origins and construction utilizing etymology result in extra surprising associations and better creativity scores. These outcomes emphasize that AI creativity relies upon closely on human steering, making interplay and prompting a central a part of the inventive course of.
Will AI Change Human Creators?
The examine affords a balanced perspective on fears that synthetic intelligence may exchange inventive professionals. Whereas AI methods can now match or exceed common human creativity on sure duties, they nonetheless have clear limitations and depend on human course.
“Regardless that AI can now attain human-level creativity on sure assessments, we have to transfer past this deceptive sense of competitors,” says Professor Karim Jerbi. “Generative AI has above all change into a particularly highly effective device within the service of human creativity: it is not going to exchange creators, however profoundly remodel how they think about, discover, and create — for many who select to make use of it.”
Slightly than signaling the tip of inventive careers, the findings recommend a future the place AI serves as a inventive assistant. By increasing concepts and opening new paths for exploration, AI might assist amplify human creativeness slightly than exchange it.
“By instantly confronting human and machine capabilities, research like ours push us to rethink what we imply by creativity,” concludes Professor Karim Jerbi.
Concerning the Examine
The paper titled “Divergent creativity in people and huge language fashions” was revealed in Scientific Experiences on January 21, 2026. The analysis introduced collectively scientists from Université de Montréal, Université Concordia, College of Toronto Mississauga, Mila (Quebec AI Institute), and Google DeepMind.
Professor Karim Jerbi led the examine, with Antoine Bellemare-Pépin (Université de Montréal) and François Lespinasse (Université Concordia) serving as co-first authors. The analysis group additionally included Yoshua Bengio, founding father of Mila and LoiZéro, and a pioneer of deep studying, the know-how behind fashionable AI methods comparable to ChatGPT.
