Study: Human Diversity Boosts Creativity, AI Homogenizes It

A study registered in advance on arXiv (2607.26899) examined whether generative AI leads to uniformity in creative work or diminishes human diversity. Both native (L1) and non-native (L2) English authors undertook a creative metaphor exercise across three scenarios: without AI, with AI-generated concepts (AI ideation), and with AI enhancing their own ideas. Findings revealed that L2 writers provided greater overall diversity compared to L1 writers, with native-language ideation yielding the richest variety. AI ideation reduced collective diversity for all participants and negated the L2 advantage, whereas AI refinement maintained diversity. Researchers then modeled the entire writer group using personas derived from actual participant backgrounds, three model families, native-language prompts, and higher sampling temperatures. None of the simulations managed to replicate the collective diversity found in human groups, demonstrating that large language models cannot mimic or uphold the diversity essential for collective creativity.