Homogenizing Effect of Large Language Model on Creativity: An Empirical Comparison of Human and ChaGPT Writing
Year: 2024
Type: dissertation
Abstract: Despite the potential of Large Language Models (LLMs) to enhance creative processes, increasing concerns have emerged about the standardization of AI outputs, often referred to as the homogenizing effect. In this thesis, I investigated the homogenizing effect of Large Language Models (LLMs) on creativity by empirically comparing human and ChatGPT-generated writings. For this, I analyzed 600 college admissions essays, including 200 written by college applicants and 400 generated by GPT-4 utilizing various settings. I computationally measured the semantic diversity of essays using Divergent Semantic Integration (DSI) as a proxy of creativity. Additionally, I developed a novel approach to assessing collective creativity by measuring the unique semantic diversity incremented by pooling ideas across multiple essays. My findings across two studies revealed that human essays consistently show higher creativity at individual and collective levels than those generated by GPT-4. Moreover, despite enhancements aimed at maximizing the diversity of AI models’ outputs, human-written essays consistently outperform GPT-4, especially in contributing to collective creativity.
Author: Moon, Kibum
Institution: Georgetown University
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7028938241
