Authors Reviewing Authors Reviewing Authors Reviewing Authors Reviewing Authors Reviewing Authors Reviewing Authors Reviewing Authors

Submitter: Scott Rettberg, Center for Digital Narrative, U of Bergen

——————————————————

The experiment:

The Center for Digital Narrative has a weekly event called “Defrag Friday” each week, the purpose of which is to collectively participate in a creative, narrative-based activity. In the Authors Reviewing Authors Reviewing Authors… experiment, we set ourselves a goal to produce a book using ChatGPT 3.5 – the majority of the 129-page book was written (prompted) during a two-hour session on a Friday in September 2023. The editing took place over the weekend, and the book was published the following Tuesday.

The book was guided by a simple constraint. We asked the prompters to put their prompts in the form of “Write a review of Author X in the idiomatic voice of Author X.” For example: “Write a review of William Faulkner’s The Sound and the Fury in the idiomatic voice of Robert Coover’s The Public Burning” or “Write a review of William S. Burroughs’s Naked Lunch in the idiomatic voice of Jane Austen’s Pride and Prejudice.” The e-book can be downloaded via the link below.

Results:

The results of this exercise demonstrated some of the extents and limits of what ChatGPT’s training set “knows” about literature. It also demonstrated the ways in which the context, not only of a given prompt, but of a series of prompts and responses in a given conversation, have effects on the subsequent content. Some biases of the training dataset are also revealed by this type of exercise: the responses to prompts involving Norwegian literature, for example were much more cursory than those involving well-known works of American Literature. The responses are often quite comical, and finally demonstrate the “power of the prompt.” Although the majority of the text was composed by ChatGPT, the responses are constrained by the human author largely determines the style and tone of the texts. This is a dialogic genre of one-liners, probing a responsive system that has at its disposal a vast trove of literature and trivia, and can produce compellingly readable results when properly deployed. It is what I describe a process of cyborg authorship, in which human intelligence is interacting with a nonconscious cognitive system.

Relevant resources: http://retts.net/documents/authors_reviewing_authors.pdf

Contact: scott[DOT]rettberg[AT]uib[DOT]no