The Biggest Book World News of the Week

To Bot or Not to Bot

In a new study, researchers at the University of Pittsburgh found that “nonexpert readers” cannot tell the difference between poems written by beloved authors and those written by ChatGPT in the voices of those authors. Not only that, the participants actually preferred ChatGPT’s poems and were more likely to guess that they were written by humans. It’s a headline designed to incite pearl-clutching, but the details matter. First, sales of poetry books remind us that very few of even the most avid readers reach for poetry. (Per a 2022 study, less than 10% of Americans read any poetry in the last year.) Second, the writers whose work was used in the study—Shakespeare, Emily Dickinson, and T.S. Eliot, among others—all have recognizable, easily mimicked styles. Third, most people have not read these writers since high school. It’s easy to trick folks then their baseline familiarity is low to nonexistent.

Do these findings indicate a threat to poetry’s existence and/or bottom line? Only if you believe that there’s meaningful overlap between the “nonexpert readers” enlisted in the study (this is the part where I remind you that most participants in university-based research are undergraduate students) and the <10% of American adults who seek out poetry on purpose. Read more about the researchers’ methods here, and see for yourself what it looked like when one Lit Hub writer asked ChatGPT to mimic iconic poems.