Nowadays, many writers rely on AI to help produce text—sometimes for grammar and fluency, and sometimes to generate entire passages. Recent research suggests that approximately 9% of newly published American newspaper articles are now produced partially or entirely by large language models (LLMs). Yet most of this AI involvement goes undisclosed, raising concerns around transparency, public trust, authorship, and accountability.
This talk will examine the broad spectrum of AI-assisted writing, ranging from light editing to the generation of large amounts of content. We will consider when AI assistance may be acceptable to a reader—and when it may not—especially in cases where the role of AI is hidden.
Dr. Iyyer will then discuss methods for post-hoc detection:
Given a piece of text, can we tell how much AI contributed to its writing?
The talk will introduce “Frankentexts”, a new form of text generation in which an LLM generates the ideas for a piece but then repurposes fragments of human-written text to construct the final output. This method complicates detection and raises higher-level questions for copyright, creativity, and the boundaries of authorship.
The talk will conclude with ongoing work that connects AI detection research to broader questions about memorization, originality, and creative contribution in human–AI writing partnerships.
About the Speaker
Dr. Mohit Iyyer is an Associate Professor in Computer Science at the University of Maryland, College Park, specializing in natural language generation. He has received multiple Best Paper awards at major NLP conferences and was named the 2022 Samsung AI Researcher of the Year. He earned his PhD in Computer Science from UMD, conducted postdoctoral research at the Allen Institute for AI, and previously served as faculty at UMass Amherst.
Learn more: