Künstliche Intelligenz: ChatGPT schreibt für viele die fesselnderen Kurzgeschichten

A recent study published in the journal "Judgment and Decision Making" suggests that readers may perceive artificial intelligence-generated short stories as more compelling than those written by human authors. In the first of three experiments involving around 1,700 participants, they could not reliably determine the author of the story they read, with the researchers misleadingly attributing AI-chosen ChatGPT stories to human authors and vice versa.

Deena Weisberg, a psychologist at Villanova University in the United States, found that when participants believed a story was written by a human, their positive judgments were more pronounced. This is likely because people prefer to think that creative writing requires unique human abilities such as life experience or emotional understanding, which leads them to underestimate AI's capabilities.

In the subsequent two experiments, Weisberg and her colleague Sydney Sears confirmed the same finding. Less than one in two hundred participants correctly identified which story was generated by AI and which was written by a human in two additional story readings, resulting in a statistically insignificant outcome. Weisberg suspects that this can be explained by a sort of feedback loop: AI texts are typically clearer, direct, and easier to process, while human-written stories tend to be more subtle and complex.

AI-generated stories often explicitly named themes instead of challenging readers to infer meanings from the characters' actions and words. Consequently, participants preferred the AI-created stories. However, they attributed the story they liked more to human authors due to their negative attitude towards AI. The study notes that AI experience helped participants identify the authors better.

Participants who reported using ChatGPT and similar tools regularly were more likely to correctly attribute the text's authorship. Weisberg suggests that improving AI competence might help people navigate the new AI-dominated world we live in. In Germany, many teachers are concerned about detecting students who secretly use AI for their homework, and one possible clue could be the lack of variety, as chatbots often repeat phrases and seldom use idiomatic expressions, slang, or dialect.

However, the internet also contains many guides on how to avoid these simple detection features and make the language more human-like.