Large Language Model Integration Risks Homogenizing Human Creativity and Stifling Originality Across Diverse Industries

PNAS Nexus research finds AI models produce repetitive ideas compared to humans, warning that relying on LLMs for art and brainstorming may harm human thought.

By: AXL Media

Published: Mar 24, 2026, 8:56 AM EDT

Source: Information for this report was sourced from PNAS Nexus

The Paradox of Artificial Originality

New research conducted by Emily Wenger and Yoed N. Kenett has identified a fundamental tension in the creative capabilities of large language models. While a single response from an AI might be rated as more creative than the average human effort, a broader analysis reveals a lack of true variety. When evaluating tasks such as generating diverse uses for common objects, the researchers found that AI outputs cluster together in feature space. This suggests that while the machine can mimic a high level of creative flair in isolation, it lacks the expansive, unpredictable range that characterizes human imagination across a population.

Measuring Cognitive Diversity in Machine Outputs

To quantify this phenomenon, the study utilized participants from the Prolific platform alongside a suite of major models, including Gemini, GPT, and Llama. Participants were tasked with divergent thinking exercises, such as listing ten nouns that are as conceptually distant from one another as possible. The data consistently showed that LLM responses were significantly more similar to one another than those produced by humans. This collective conformity indicates that the underlying architecture of these models leads them toward a shared set of "likely" creative answers, rather than the idiosyncratic outliers often produced by the human mind.

The Failure of Randomness as a Substitute for Inspiration

In an attempt to bypass this homogeneity, researchers experimented with increasing the "temperature" of the models, a setting that controls the randomness of the generated text. While higher temperatures did succeed in making the responses more variable, the strategy was ultimately self-defeating. Beyond a certain threshold, the increased randomness resulted in incoherent gibberish that failed to meet the basic requirements of the assigned tasks. This suggests that the "creativity" found in AI is a fragile balance between predictable patterns and total chaos, lacking the structured originality inherent in human cognition.