The Representation of Moral Intuitions in Stories Created by ChatGPT

Abstract: Generative AI, such as ChatGPT, has become commonplace. One way AI is being used is to create stories; however, the moral values they highlight are unclear. The current study examines the types of moral intuition exemplars most commonly found in AI-generated narratives. Using randomized prompts that specified different types of narratives (quality-level, genre, medium, target audience, and intended response), ChatGPT-4o was instructed to create ten thousand stories, which were coded for the presence of the five moral intuitions from Moral Foundations Theory. Results show that higher-quality stories were more likely to include moral intuitions that were upheld rather than vioated. Likewise, the upholdings/violations of each moral intuition varied when prompted to create stories for different genres, mediums, and target audiences. The degree to which AI-generated stories reflect the intuition exemplars prominent in the media landscape in general and for audiences across moral subcultures is discussed.