AI Fiction in the Wild

Co-sponsored by the Berkeley Institute for Data Science and the School of Information

We know that some professional authors are beginning to use AI tools to help produce their fiction writing. Are readers using AI to generate fiction, too? Thanks to a unique public dataset called WildChat — containing more than one million “real world” conversations with ChatGPT, all shared with user consent — we can say, definitively, yes. 

This talk will share insights from a collaborative analysis of WildChat, where we found that more than a third of the English-language conversations involved fiction generation: original stories, role-play, world-building, fan fiction, erotica, and more. I will argue that this activity, while seemingly novel, extends trajectories of contemporary literature charted by critics like Mark McGurl, in which readers increasingly prize generic forms, repetition, and instantaneity, and where they threaten the traditional authority of the author as a customer and consumer. Lastly, I will speculate about what the WildChat fiction suggests about the future of publishing and entertainment in the age of AI.

The Cultural Analytics Series is a series of lunchtime talks and workshops highlighting research that focuses on the data-driven analysis of cultural phenomena.