AI Surpasses Average Human Creativity in Massive Study of 100,000 Participants

AI Outperforms Average Human in Creativity, but Top Minds Still Reign Supreme

A groundbreaking study published this week in Scientific Reports has marked a definitive milestone in the evolution of artificial intelligence: for the first time, generative AI models have demonstrated the ability to surpass the average human in standardized creativity tests. However, the study, which analyzed data from over 100,000 participants, offers a critical caveat—the most creative human minds still significantly outperform even the most advanced AI systems.

Conducted by a multidisciplinary team from the Université de Montréal, Concordia University, and Google DeepMind, the research provides the largest comparative analysis of human versus machine creativity to date. The findings suggest that while AI has democratized "average" creativity, the upper echelons of imaginative capability remain a distinctly human stronghold.

The Scale of the Study: 100,000 Participants

Led by Professor Karim Jerbi of the Université de Montréal and featuring contributions from AI pioneer Yoshua Bengio, the study sought to settle the long-standing debate: Can AI truly be creative? To answer this, researchers moved beyond small-scale anecdotes and rigorous Turing tests, deploying a massive dataset involving 100,000 human subjects.

The core of the assessment relied on the Divergent Association Task (DAT). Unlike subjective art critiques, the DAT is a standardized psychological instrument designed to measure divergent thinking—a key component of creativity that involves generating multiple unique solutions to an open-ended problem. Participants were asked to generate ten nouns that were as semantically distant from one another as possible.

For example, a low-scoring sequence might be "cat, dog, pet, animal," while a high-scoring, highly creative sequence might look like "galaxy, fork, freedom, algae, harmonica." The semantic distance between these words was calculated computationally to derive a creativity score.

Comparison Results: The "Average" Threshold

The results revealed a shifting landscape. Modern Large Language Models (LLMs), including GPT-4, Claude, and Gemini, consistently scored higher than the average human participant on the DAT. The AI models demonstrated a superior ability to break semantic clusters and retrieve disparate concepts rapidly, a hallmark of divergent thinking.

However, the data also highlighted a "creativity ceiling" for AI. While machines easily surpassed the median human performance, they failed to compete with the top percentile of human participants. The most creative humans—specifically the top 10%—consistently generated semantic associations that were more original and varied than the best outputs from GPT-4 or its peers.

The following table summarizes the comparative performance levels observed in the study:

Participant Group	Performance Standing	Key Characteristics
Average Humans	Baseline	Tends to cluster concepts (e.g., listing related household items)
Generative AI (GPT-4)	Above Average	High semantic distance; exceeds median human capability
Top 10% Humans	Superior	Exceptional divergence; highly original, non-linear connections

Beyond Simple Associations

To ensure the findings were not limited to simple word games, the researchers extended the comparison to more complex creative tasks, including writing haikus, summarizing movie plots, and crafting short stories.

In these qualitative assessments, the pattern held firm. AI models produced technically proficient and structurally sound creative text that outperformed the average layperson's attempts. Yet, when compared to skilled human writers or highly creative individuals, the AI output often lacked the subtle novelty and emotional resonance that characterized the top human work.

Professor Jerbi noted in the study that while AI acts as a formidable "remixer" of existing data, allowing it to outperform humans who may rely on predictable associations, it struggles to replicate the intentional, erratic, and deeply novel leaps made by the most talented human creators.

The Role of "Temperature" and Prompting

The study also delved into the technical parameters that influence AI creativity. Researchers found that adjusting the "temperature"—a parameter that controls the randomness of an AI's output—significantly impacted performance. Higher temperatures allowed the models to take greater risks, generating more divergent answers that pushed them closer to high-level human performance, though often at the cost of coherence.

Furthermore, prompting strategies played a crucial role. When AI models were prompted to "think specifically about etymology" or given other structural constraints, their creativity scores improved. This suggests that AI creativity is not a fixed trait but a modulatable capability that depends heavily on human guidance.

A Tool, Not a Replacement

The implications of this study for the creative industries are profound but nuanced. Rather than signaling the obsolescence of human creativity, the findings position Generative AI as a powerful augmentation tool.

For the average person, AI can serve as a "creativity engine," lifting their output to a higher baseline of quality and divergence. For top-tier creatives, AI serves as a competent assistant that can handle the "average" heavy lifting of brainstorming, allowing the human to focus on the high-level conceptual work that machines still cannot touch.

"We need to move beyond this misleading sense of competition," Professor Jerbi stated regarding the findings. "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."

Scientific Significance

This research, published in Scientific Reports, validates the rapid progress of Scientific Research in the field of AI evaluation. By establishing a quantifiable metric for creativity that applies to both biological and synthetic minds, the study provides a foundation for future AI development.

It also reassures the artistic community that while the floor for creativity has been raised by automation, the ceiling remains high and distinctly human. As we move further into 2026, the collaboration between the "average-surpassing" AI and the "superior" human mind appears to be the most promising path forward for innovation.