The discovery finds a strange flatness: AI is now outperforming the average human on standard creativity tests. We knew something like this was coming. We’ve been debating this for years. And yet, now that the data is actually here, it’s less clear-cut than expected—because the more you read the paper, the more the title unravels into a much more interesting question.
The research, which was published in Scientific Reports In January 2026, this is the largest direct comparison between human and artificial intelligence creativity. Led by Université de Montréal professor Karim Jerbi, with contributors including Yoshua Bengio, one of the architects of modern deep learning, the study evaluated GPT-4, Claude, Gemini, and other leading models against the creative output of more than 100,000 human participants.
The primary tool was the Divergent Association Task—a validated measure of creativity that asks humans (and now AI systems) to list ten words that are semantically as far apart as possible. The wider the conceptual range, the higher the creativity score. AI systems, especially GPT-4, achieved above-average human performance on this task. They also achieved above average results in haiku composition, short story writing and plot summaries.
But this is where the study gets really interesting.
The floor gave way. The ceiling is not.
When the researchers looked not at the average but at the top performers—the most creative people in the sample—the picture changed completely. The most creative half of human participants outperformed all AI models tested. The gap between the top 10% continued to grow. At the highest level of human creativity, AI is not competitive.
What sets these people apart from the average attendee? The researchers do not say this explicitly, but the structure of the task suggests it. The Divergent Association Task rewards people who persist—after giving a clear answer, they refuse to settle there. Those who, in addition to the “galaxy” and the “forks” and the “storm”, move to something alien, something further away, in the conceptual space, something they had to work for.
Artificial intelligence optimizes for the likely answer. People with the highest scores optimize for the surprising—and are apparently willing to put up with a lot of discomfort to get there.
That’s not a small difference. The average human, like the average artificial intelligence, stops when the answer seems right. The exceptional person keeps moving past the right, good, smart things of the past, into something that costs more to produce and feels less secure upon arrival. AI doesn’t seem to be able to replicate that willingness to stay in the discomfort of not feeling good enough.
Why this result is not reassuring in an obvious way
The temptation is to read this study as a consolation: the best people are still ahead of us. But there are at least two reasons to take comfort lightly.
First, the study confirmed that AI creativity is not constant—it can be modified by changing temperature settings and instantaneous structure. Higher temperatures result in more unexpected outputs. Calls for etymological thinking create more distant associations. The ceiling of AI is being deliberately and continuously raised. Human creativity develops in different ways: slowly, through practice, exposure, failure, and a certain stubborn refusal to be satisfied.
Second, the researchers found something troubling at the lower end of the human distribution. It’s not that average humans are somehow worse than AI. It’s that a significant number of human participants fared significantly worse—producing traditional, predictable, semantically clustered responses that any basic language model could easily outperform. The question to ask isn’t “are the best people still ahead?” but “what are we doing culturally and educationally to protect and develop the cognitive habits of people who can become the best?”
The finding is not that AI has become creative. Human creativity is now visibly stratified—between those who have retained the ability to get along well enough, and those for whom this friction has been quietly, comfortably, optimized.
Sovereign Mind lens
This is exactly the kind of problem the Sovereign Mind framework It was created to solve this: when a system optimized for efficiency begins to erode cognitive abilities that cannot be optimized – only practiced.
- Unlearning: An assumption worth examining is that creativity is a trait—something you either have or you don’t. Rather, the data suggest that this is a practice that most educational and professional settings are systematically undoing by rewarding a quick, correct response over a slow, odd response. What scenarios have you adopted about what “creative enough” looks like?
- Renovation: Divergent thinking—the kind that creates a wide range of concepts—requires cognitive conditions that are increasingly rare: unstructured time, low pressure to perform, permission to produce bad ideas before good ones. The environment that produces the peak of human creativity is almost the opposite of the environment in which most of us live. Rebuilding this space is not a luxury. The study now provides a rationale supported by data.
- Protection: A particular kind of intellectual manipulation is at play when AI-generated content is optimized to feel creative—to feel surprising, fresh, unexpected—while actually occupying the statistical center of people’s expectations. Recognizing the difference between genuine novelty and sophisticated pattern matching is increasingly a cognitive ability worth consciously developing. The question to ask of any creative output, human or machine, is: did it come from somewhere, or does it just sound like it is?
The habits that protect you from replacement
If we look at what distinguishes the top 10% of humans at this task, it boils down to a specific set of cognitive behaviors: comfort with ambiguity, tolerance for open-endedness, a willingness to generate many ideas before choosing any, and—critically—a resistance to complacency. Herbert Simon’s term for the human tendency to accept the first solution that reaches the minimum threshold is satisfactory. This is effective. It’s also the creative death spiral, and it’s what AI has perfected.
People who score highest on different creativity tasks are inefficient. They don’t stop at the right one. They produce more opportunities than they should, walk on strange paths, endure the friction of not yet knowing. These are not natural tendencies – they are cultivated and need an environment that rewards rather than punishes them for delay.
And because AI can handle the right answers faster and more fluidly than most humans, this willingness to operate inefficiently—to keep going, to sit in uncertainty, to reach for the farther word—may be the most enduring human creative domain.
