Editor's Synopsis
A new study published in Science Advances explores whether generative AI actually boosts human creativity, using OpenAI's GPT-4 to help participants write short stories. Researchers recruited 293 people, measured their baseline creativity, then split them into groups: one writing unaided, one receiving a single AI-generated story idea, and a third receiving up to five ideas. The vast majority (88.4%) of those offered AI assistance chose to use it.
The results revealed a "leveling effect": writers who scored lower on inherent creativity saw the biggest improvement in their output quality when using AI, while already-creative writers gained little benefit. However, the study also uncovered a significant downside — stories produced with AI assistance were more similar to each other than those written entirely by humans, both in semantics and content. Researcher Tuhin Chakrabarty of Columbia University noted that AI writing tends to be exposition-heavy and full of stereotypes, observing that "good writing is all about showing, not telling. AI is always telling."
The findings suggest that while AI can help individuals become more creative, it reduces collective creativity by homogenizing output. The researchers warn that if industries like publishing widely adopt generative AI, the resulting works could become increasingly uniform, since they all draw from models trained on the same data. Co-author Oliver Hauser emphasized the importance of studying both AI's capabilities and its limitations as society navigates the technology's rapid evolution.
