Study: AI Writing Strips Mystery and Complexity from Stories (neurosciencenews.com)

Researchers at the University of North Carolina at Chapel Hill have unveiled a study revealing that AI-generated stories often lack the complex, mysterious qualities that make human-authored fiction memorable. By developing an automated evaluation framework called CASPER, the team analyzed thousands of narratives across eight axes of literary theory. They found that AI models tend to "play it safe" by relying on predictable character archetypes and providing tidy resolutions, stripping away the narrative ambiguity that keeps readers engaged. Even scaling up AI models did not yield richer character development, emphasizing that the root issue lies in the underlying storytelling mechanics rather than just model size. This significant finding highlights the limitations of current AI in creative writing, pointing out that while tools like Sudowrite and Squibler can assist authors, they risk homogenizing narratives if relied on too heavily. The CASPER benchmark offers a crucial framework for assessing the depth and complexity of characters in AI-generated content, guiding future advancements in AI storytelling. Ultimately, the research underscores the importance of the human touch in writing, advocating that the most compelling stories require a willingness to embrace ambiguity, contradiction, and unresolved character arcs—qualities that AI struggles to replicate.