New research has shed light on how the perceived origin of text impacts human evaluation. The study reveals that readers tend to rate AI-generated short stories more highly than human-authored ones—as long as they believe the content was written by a person.
This study, which analyzes human psychological biases toward text generated by Large Language Models (LLMs), presented readers with a series of short stories. By comparing evaluations of AI-generated content against human-written pieces, the researchers identified a significant disconnect between quality perception and authorship attribution.
The results show that before being informed of the origin, readers consistently preferred AI-generated works over those written by humans. However, once the "AI-generated" label was revealed, these evaluations dropped sharply. This suggests that when assessing the value of content, readers are influenced more by the perceived source than by the actual quality of the writing itself.
As AI technology continues to advance, our ability to generate high-quality content is reaching new heights. However, this study underscores the significant psychological barriers and biases that shape how we perceive authorship. It raises critical questions for our future AI-integrated society regarding how we establish value benchmarks for creative content and how we socially perceive the intersection of machine and human creativity.