Title: AI and the Writer: How Language Models Support Creative Writers

Abstract:  Writing underlies a vast landscape of cultural artifacts, from poetry to journalism to scientific papers. In this talk, I consider how natural language generation might assist us with these complicated kinds of writing. In particular I look at writing tasks that are constrained by some external expectation, such as the logic of a metaphor or the details of a technical topic, but also require creativity to write a sentence or paragraph that is novel, surprising, and engaging to read. I introduce a design space, based on the cognitive process model of writing, that articulates the current paradigm of writing support tools. I then present systems to support two constrained, creative writing tasks: metaphor writing and science writing. Finally, based on a qualitative study with a range of creative writers, I outline the social dynamics that modulate how writers respond to such generative writing support tools.

Bio: Katy Ilonka Gero completed her PhD in human-computer interaction at Columbia University in 2022. Her work investigates how creative writers make use of computer-generated text, as well as how to understand bias and variability in computer-generated text. Her research has been supported by a NSF Graduate Research Fellowship and the Brown Institute for Media Innovation. She has also published both traditional and computational poetry, as well as essays about computation, in venues such as Catapult, The HTML Review, and Wired. She is currently a postdoctoral fellow at Harvard University. https://www.katygero.com/