summary: By developing a new automated assessment framework called CASPER, researchers analyzed thousands of human-written and machine-generated stories across eight different axes of literary theory. The findings conclusively demonstrate that AI models systematically remove one of the hallmarks of memorable fiction: mystery.
While human authors routinely embrace ambiguity in their stories, leave serious questions unanswered, and allow their characters to remain beautifully contradictory, AI models uniformly “play it safe.” They rely heavily on flat, predictable archetypes, forcing storylines into artificial, perfectly tidy resolutions.
important facts
- Safe resolution bias: Lead author Anneliese Bligh points out that AI systems have inherent mathematical biases to neatly organize the story. They actively resolve internal conflicts, answer every mystery, and ensure that by the final page the characters fit perfectly into their designated story arcs.
- Illusion of scale: An important finding from this study is that increasing the size of the parameters does not solve the problem. The large, state-of-the-art flagship LLM produced characters as flat and typical as those produced by significantly smaller and less complex models. This shortfall is not caused by processing power, but by how the model understands storytelling.
- Accept unresolved: When analyzing human writers, the CASPER framework revealed a high comfort level with chaos. In fiction written by humans, characters are often unresolved, morally gray, or fundamentally open to interpretation, and it is precisely this structural ambiguity that makes a story stick with readers.
- Character evolution rating: The study methodically mapped characters’ actions against eight core dimensions of literary theory, analyzed their precise transitions from hyper-exaggerated caricatures to realistic figures, and tracked whether characters truly evolve or simply follow a script.
- CASPER Benchmark: In addition to exposing creative limits, CASPER serves as an important standardized benchmarking framework. This will enable AI developers and creative studios to assess whether upcoming next-generation models truly advance narrative depth and character complexity, rather than just grammatical fluency.
- Lessons for the writing community: The UNC study provides a critical warning for authors who rely on AI as assistants and co-authors in interactive brainstorming. In other words, letting machines dictate character development risks homogenizing the story, making the human touch essential to reintroducing contradictions, subverting expectations, and intentionally injecting uncertainty.
sauce: UNC Chapel Hill
Researchers at the University of North Carolina at Chapel Hill have found that while artificial intelligence is able to weave increasingly compelling stories, its characters may still lack one of the qualities that makes human novels memorable: mystery.
As AI writing tools become more common in publishing and entertainment, researchers at Carolina wanted to understand whether the characters created by these systems are as diverse and nuanced as those created by human authors. Their findings suggest that despite advances in technology, AI still tends to rely on familiar patterns.
This study investigated how characters in AI-generated stories compare to stories written by humans. Using ideas from literary theory, the researchers analyzed eight different aspects of character depiction, including whether characters seem realistic or exaggerated, whether they evolve over time, and whether they remain mysterious or fully understood by the end of the story.
To do this, the team developed CASPER. It is an automated framework that evaluates thousands of stories and measures character traits in a way that has never been systematically applied to AI-generated fiction before.
“We found that AI models tend to play it safe with characters in the sense that they keep the storylines together neatly,” said Anneliese Bly, a computer science graduate student at UNC-Chapel Hill and lead author of the study.
“Human writers, on the other hand, may prefer to leave questions unanswered and characters a mystery. This difference is important because ambiguity often creates stories that stick with readers.”
The study comes at a time when AI tools designed specifically for creative writing are gaining traction. Platforms like Sudowrite and Squibler are useful for drafting novels, but film and TV are increasingly using AI to generate script outlines and dialogue. Research also shows that many novelists are now incorporating AI as part of their creative process.
Their analysis revealed that AI-generated characters rely heavily on recognizable archetypes and tend to arrive at a neat resolution by the end of the story. In contrast, human writers seemed comfortable allowing their characters to be unresolved, contradictory, and open to interpretation.
“One of our most surprising findings is that larger, more powerful AI models do not necessarily produce more diverse characters than smaller AI models,” said Nicolas Sanaye, a University of Carolina computer science undergraduate student and co-author of the study.
“This tells us that the challenge is not just about scale; it’s about how these models understand storytelling itself.”
CASPER gives researchers, developers, and creative professionals a way to benchmark whether new AI systems are actually improving the depiction of complex characters, rather than just becoming more fluent writers. It also has the potential to guide the development of future storytelling tools that better support creativity and narrative depth.
“As more and more people collaborate with AI to write novels, screenplays, and other creative works, we need ways to understand what these systems do well and where they fall short,” said Snigdha Chaturvedi, associate professor of computer science at UNC-Chapel Hill and senior author of the study.
“CASPER provides a lens through which to assess the depth and diversity of characters, ultimately helping developers build storytelling systems that better reflect the complexity of the human experience.”
For writers experimenting with AI, the findings offer practical lessons. AI may become an increasingly capable creative partner, but the most compelling stories may still require a distinctly human willingness to embrace uncertainty, contradiction, and characters that don’t fit the familiar mold.
Answers to key questions:
answer: This “neatness” bias is built directly into how large-scale language models are trained. AI models are trained based on mathematical probabilities to predict the most satisfying and logical next word based on vast amounts of internet data. When building a story, the model naturally optimizes for the most probable path. This means that the model leans toward predictable structure and neat resolution. Rather than sitting in discomfort, you are programmed to provide answers. But human life is full of impasses and contradictions, and while human writers intentionally capture these qualities, AI sees them as statistical anomalies that should be smoothed out.
answer: CASPER is an automated computational linguistics framework designed by the UNC team to turn abstract literary theory into measurable data. It analyzes thousands of text blocks and tracks how characters are portrayed and how their actions unfold in eight specific dimensions from beginning to end of the story. To measure “mystery” or ambiguity, Casper scans whether a character’s inner motivations are fully explained by the narrator by the climax, or whether a character’s actions remain beautifully unmapped, contradictory, and open to multiple interpretations, quantifying the precise intangible characteristics that distinguish flat caricatures from haunting literary icons.
answer: Not at all. Researchers see AI as an incredibly capable creative partner, but it requires a steady human hand. AI is great for helping writers brainstorm plot outlines, cure blank page syndrome, and quickly flesh out simple background explanations. The real lesson here is that you can’t delegate. soul Machine character development. If a novelist lets an AI write the main characters without permission, those characters will inevitably turn into flat, safe clichés. True storytelling magic requires human writers who are willing to intervene and deliberately mess with the code by adding unresolvable flaws, nasty contradictions, and a healthy dose of genuine mystery.
Editorial note:
- This article was edited by the editors of Neuroscience News.
- Journal articles were reviewed in full text.
- Additional context added by staff.
About this AI and creativity research news
author: gabriella neyman
sauce: University of North Carolina at Chapel Hill
contact: Gabriella Neiman – University of North Carolina at Chapel Hill
image: Image credited to Neuroscience News
