The nineteenth-century newspaper was eclectic and far-reaching: a miscellany of news, entertainment, art, science, opinion, and practical advice. It was also a medium founded on unoriginal writing, as editors composed their papers substantially through the reprinting of material from other newspapers. In the United States, such textual recycling was neither incidental nor covert, but directly supported by both custom and law. Drawing on our extended study of the "exchange system" in the Viral Texts project (https://viraltexts.org), this lecture will ask how this history of unoriginal writing might inform our understanding of—and responses to—the newest technology of unoriginal writing, large language models (LLMs), such as ChatGPT. While scissors and paste have been replaced by algorithmic recombination, nineteenth-century newspapers share with LLMs qualities of collage and pastiche, as well as concerns about veracity and displaced labor. Building on their shared qualities, the lecture will then suggest ways historians might use LLMs—perhaps best suited to produce the average or mundane—to scale up tasks also based on normative patterns across large-scale historical newspaper collections, such as genre classification, textual segmentation, or topic identification. Finally, the lecture will show how methods developed for computational historical research, such as the reprint detection tools developed for Viral Texts, can contribute to bibliographical investigation and description of LLMs and related AI systems.

Please register by May 14.