Is AI More Creative Than Humans?

A new large-scale study demonstrates that when it comes to creativity, AI already outperforms the average human, but not the most creative among us.

“The persistent gap between the best-performing humans and even the most advanced LLMs indicates that the most demanding creative roles in industry are unlikely to be supplanted by current artificial intelligence systems,” wrote AI deep learning pioneer Yoshua Bengio, along with the study’s co-first authors Antoine Bellemare-Pépin and François Lespinasse, and co-authors Philipp Thölke, Yann Harel, Kory Mathewson, Jay A. Olson, and Karim Jerbi, who led the study.

Are Creative Jobs Safe From AI?

The creative economy accounts for roughly 4% of the U.S. gross domestic product or an estimated USD 1.2 trillion in 2023, according to the National Endowment for the Arts. Approximately 500 million jobs globally are in the creative economy according to United Nations estimates.

Creativity is essential for many industries such as digital media, information technology, publishing, music, performing arts, advertising, marketing, film, television, radio, design, video games, software, computer, visual arts, crafts, design goods, jewelry, culinary arts, cultural institutions, photography, design, fashion, architecture, dance, entertainment, and more.

Years before the generative AI chatbot ChatGPT was released to the public by OpenAI in November 2022, futurists posited that the industry areas where AI would be adopted initially would be in professions where the jobs are repetitive, predictable, and rote. Look at any list prior to 2022, and the jobs that would be considered safe included many professions in the creative industry. Fast forward to 2026, and it seems like AI is a looming threat to many creative professions as well.

AI in the Creative Economy: Threat or Opportunity?

AI and technological automation are rapidly transforming creative industries. By 2030, activities that make up 30% of hours currently worked across America’s economy could be automated, and generative AI is accelerating the pace, according to McKinsey Global Institute’s “Generative AI and the future of work in America” report.

According to one survey of over 2,500 creative professionals conducted by Adobe shows that 83% reported using generative AI tools for work. The rising adoption rate of using generative AI tools in the creative economy necessitates a closer look at the creative abilities of large language models compared to humans.

It seems natural that humans will always be more creative than machines. Yet artificial intelligence (AI) machine learning is rapidly improving and rapidly encroaching on the decidedly human realm of creative abilities. Just how creative are generative AI models?

How to Objectively Measure Creativity

How to take a scientific approach in measuring a highly subjective and fuzzy attribute? The answer lies in a methodology developed by a co-author Jay A. Olson in prior research. Olson introduced the Divergent Association Task (DAT) method in a different research study when he was a Postdoctoral Fellow at Harvard University with the publication of “Naming unrelated words predicts creativity” in The Proceedings of the National Academy of Sciences (PNAS), along with co-authors Margaret Webb, Johnny Nahas, and Denis Chmoulevitch. DAT was developed with over 8,900 study participants spanning 98 countries who were asked to complete tests of creativity.

To conduct the new study led by Professor Jerbi from the Department of Psychology at the Université de Montréal, the team scientifically compared the performance of a wide range of AI large language models (LLMs) versus humans using the Divergent Association Task (DAT) method.

“We benchmark performance on the Divergent Association Task (DAT) and across multiple creative-writing tasks (haiku, story synopses, and flash fiction), using identical, objective scoring,” the researchers wrote.

To adapt DAT for evaluating LLMs, instructions were converted into prompts. The prompt asks the LLMs to output 10 words that are as different from each other as possible. Then DAT scores were calculated from the first seven words that are valid from the 10 words produced by the LLMs. The researchers measured a wide variety of LLMs against a massive dataset of 100,000 human participants split evenly between men and women.

“We found evidence that LLMs can surpass average human performance on the DAT, and approach human creative writing abilities, yet they remain below the mean creativity scores observed among the more creative segment of human participants,” the researchers wrote.

The LLMs evaluated include StableLM by Stability AI, Vicuna by Nous Research, Pythia by EleutherAI, Claude 3 by Anthropic, RedPajama by Together AI, Gemini Pro by Google, and OpenAI’s GPT-4-Turbo, GPT-4, and GPT-3. The researchers discovered that GPT-4 outperformed the human population-average DAT scores the most, followed by Gemini Pro.

“A key finding is that several LLMs, including GPT-4, surpass the population-average DAT score from our sample of 100,000 humans; however, even the best-performing models do not exceed the mean of the top 50% of human responses, and the upper human deciles still define a clear gap,” the researchers reported.

The study suggests that highly creative people still have the performance advantage over LLMs for now.

“Despite widespread concern that AI could imminently replace creative professionals (like writers, for instance), our results suggest that such fears remain premature,” the researchers reported.

Copyright © 2026 Cami Rosso. All rights reserved.