IPHS 300: Artificial Intelligence for the Humanities: Text, Image, and Sound
Document Type
Poster
Publication Date
Fall 2019
Abstract
Someday, in the not-so-distant future, will the writers’ room be replaced by artificial intelligence? This poster was created by a group of actors and writers who were curious about the role artificial intelligence could play in the future of television writing. In the first stage of this project, we prompted GPT-2 with the opening lines of a scene from the television show Sex and the City, and had GPT-2 finish the scene. We filmed the three best GPT-2 generated scenes, and screened these scenes for our “AI for the Humanities” class, along with a filmed version of the original Sex and the City scene. We hypothesized that the GPT-2 generated script would be able to convincingly mimic the simple back-and-forth dialogue of Sex and the City. Unfortunately, the students in this class easily distinguished between the GPT-2 generated scenes and the human written scene. After reaching the conclusion that GPT-2 is not able to write a believable Sex and the City script, we generated another series of scenes, this time using a GPT-2 model that had been trained on every episode of Sex and the City. We found that the trained GPT-2 was able to mimic the style of Sex and the City better than the untrained model, but even the trained model struggled with creating cohesive scenes. Based on the results of this experiment, it can be concluded that GPT-2 is capable of mimicking the style of Sex and the City, but is not yet capable of writing a scene that an audience would believe to be written by a human.
Recommended Citation
Destine-DeFreece, Alasia; Handelsman, Samara; Light Rake, Talia; Merkel, Ally; and Moses, Gracie, "Can GPT-2 Replace a Sex and the City Writers’ Room?" (2019). IPHS 300: Artificial Intelligence for the Humanities: Text, Image, and Sound. Paper 15.
https://digital.kenyon.edu/dh_iphs_ai/15
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.