Here's a puzzle that should worry anyone who cares about human creativity: what if AI makes each of us more creative as individuals, while simultaneously making all of us more similar? A 2024 study published in Science Advances provides compelling evidence that this is exactly what's happening — and the implications extend far beyond creative writing.
The Experiment
Researchers Anil Doshi (UCL) and Oliver Hauser (University of Exeter) designed an elegant experiment. They asked hundreds of participants to write short stories — some with access to AI-generated ideas (from GPT-4), some without. The stories were then evaluated by independent judges on multiple dimensions of creativity, and analyzed computationally for thematic diversity.
The setup was carefully controlled: participants in the AI-assisted group received story ideas generated by GPT-4, which they could use, modify, or ignore. This mirrors the real-world scenario of a writer using AI for brainstorming rather than having AI write the whole piece.
Key Findings
- Individual creativity increased significantly — stories written with AI assistance were rated as more creative, more novel, and better written than stories written without AI. The effect was especially strong for participants who self-reported as less creative.
- Collective diversity plummeted — when you looked at the full set of AI-assisted stories, they were significantly more similar to each other than the stories written without AI. The AI had funneled diverse thinkers toward similar ideas.
- The "creativity gap" narrowed but homogenized — less creative writers improved the most, but they improved by converging toward the same AI-suggested patterns. The result was a smaller variance in quality but also a smaller variance in content.
- Writers were largely unaware of this effect — when asked, most AI-assisted participants believed their stories were highly original. They didn't recognize that they had been nudged toward the same themes, tropes, and narrative structures as everyone else.
The Bitter & The Sweet
The sweet side is real and significant. AI-assisted creativity tools can help people who feel stuck, provide inspiration when ideas aren't flowing, and raise the baseline quality of creative output. For anyone who has stared at a blank page, the value of a good prompt or suggestion is undeniable. If AI can make more people feel capable of creative expression, that's a genuine democratization of a fundamental human activity.
The bitter side is more insidious because it's invisible at the individual level. No single writer notices that their AI-inspired story about "a lonely astronaut discovering meaning in isolation" is similar to hundreds of other AI-inspired stories about lonely astronauts. The homogenization happens at the population level, not the individual level — which means it can't be detected by the people participating in it.
This has profound implications beyond creative writing:
- Innovation — if researchers, entrepreneurs, and designers all use AI for brainstorming, the diversity of approaches to problems may shrink. Innovation depends on cognitive diversity; AI may reduce it.
- Culture — if musicians, writers, and artists all draw on the same AI systems for inspiration, cultural production could become more polished but less varied. A world of competent sameness.
- Decision-making — if teams use AI to generate options and strategies, they may converge on similar solutions, reducing the resilience that comes from diverse approaches.
"Our findings suggest a potential 'creativity trap': AI enhances the quality of individual creative output while reducing the collective diversity of ideas. This trade-off has significant implications for innovation ecosystems that depend on a wide exploration of the idea space." — Doshi & Hauser, 2024
Methodology & Limitations
The study used a between-subjects experimental design with random assignment, strong controls, and both human and computational evaluation of creativity. Its main limitation is scope: the task was short-story writing, and it's unclear whether the same pattern holds for scientific creativity, engineering design, or other domains. The researchers also used a specific version of GPT-4; different models or prompting strategies might produce different diversity outcomes.
What This Means Going Forward
This study introduces a concept that will likely become central to discussions about AI and human cognition: the individual-collective creativity trade-off. It suggests that the widespread adoption of AI creativity tools requires deliberate countermeasures — not to prevent AI use, but to preserve the diversity of thinking that AI tends to erode.
Practical strategies might include: using AI for refinement rather than ideation, deliberately generating ideas before consulting AI, and building teams where not everyone uses the same AI tools. At a societal level, the study argues for investing in the conditions that produce cognitive diversity — different educational traditions, cultural perspectives, and ways of thinking — as a counterbalance to AI's homogenizing tendencies.
References
- Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28). doi:10.1126/sciadv.adn5290
- Dell'Acqua, F., et al. (2023). Navigating the Jagged Technological Frontier. Harvard Business School Working Paper.
- Page, S. E. (2007). The Difference: How the Power of Diversity Creates Better Groups, Firms, Schools, and Societies. Princeton University Press.
- Amabile, T. M. (1996). Creativity in Context. Westview Press.