A new study has found that the most advanced AI language models can now match or beat the average person at creative thinking tasks. The catch: the most creative humans still pull ahead, and by a meaningful margin.
Researchers at Université de Montréal, led by AI pioneer Yoshua Bengio, tested leading AI systems—OpenAI's GPT-4, Google's Gemini Pro 1.5, and Meta's Llama models—against 100,000 human participants using a standard psychological test called the Divergent Association Task. The test is straightforward: generate 10 words that are as semantically distinct from one another as possible. It measures a specific type of creative thinking: the ability to make unexpected connections.
The results were clear. GPT-4, Gemini Pro 1.5, Llama 3, and Llama 4 all outperformed the median human participant. But when researchers looked at the top performers—the most creative 50%, 25%, and 10% of humans—the AI systems fell noticeably behind. The gap widened further when the researchers asked both AI and humans to create more complex outputs: haikus, movie plot synopses, flash fiction. In these tasks, human creativity consistently outpaced the models by a significant measure.
We're a new kind of news feed.
Regular news is designed to drain you. We're a non-profit built to restore you. Every story we publish is scored for impact, progress, and hope.
Start Your News DetoxWhat's interesting is that the gap isn't fixed. The researchers discovered they could boost AI creativity with simple adjustments. Increasing the "temperature" setting on GPT-4—a parameter that adds randomness to outputs—pushed it above the 72nd percentile of human performance. Carefully rewritten prompts improved performance on GPT-3.5 and GPT-4 as well. This suggests the models' creative ceiling might be higher than their default settings reveal.
The reassurance and the caveat
Karim Jerbi, who led the study, framed the findings as somewhat reassuring for creative professionals. The persistent gap between top human performers and even the best AI models suggests that genuine creative mastery—the kind that separates exceptional creators from the rest—remains a distinctly human strength. But Jerbi also urged creators to take these models seriously as collaborators, not competitors. AI systems, he argued, could "profoundly transform how [creators] imagine, explore, and create."
The study adds to a growing body of research that's beginning to answer a fundamental question: Is creativity uniquely human? This research doesn't settle the debate, but it offers one of the most rigorous attempts yet to measure it. The answer appears to be more nuanced than either "AI will replace all creativity" or "AI can't be creative." Instead, it's something closer to: AI can now do the creative work that the average person does. The exceptional work—the ideas that surprise us, that shift culture—still belongs to the humans who can think in ways machines haven't learned to replicate.
