NOVEL THEORY
SPEAKERS: Elyse Graham (Stonybrook University);
Benjamin Mangrum (MIT); Tom Comitta
Literature as thinking, or, why AI can’t nail metaphors – Elyse Graham
This paper explores why large language models (LLMs), for all their superficial fluency with language, are exquisitely bad at writing metaphors. Drawing on neuroscientific studies of what constitutes “thinking” for such models, it argues that deficiency on the level of metaphor is baked into their functioning — and will remain so even as they continue to grow in sophistication. This deficiency helps us to better understand what tech calls thinking, to borrow a phrase, as well as what distinguishes the literature of machines from the machinery of literature. The rise of AI invites renewed attention to the study of metaphor as the driving engine of literature itself — not just as a mode of speaking, but as a way of knowing.
AI and Social Contract Theory – Benjamin Mangrum
The history of the novel has been adjacent to, and often entangled with, modern theories of a social contract. This historical entanglement has never been stable, changing along with the rise of new media as well as other social and economic shifts. My talk will consider how generative AI forces critics to grapple again with social contract theories in our accounts of the contemporary novel. The rise of generative AI revises longstanding questions about what forms of literary speech count as authentic, which institutions confer cultural legitimacy, and the structural position of critical judgment more generally. My talk looks to these issues as evidence that the novel’s place in an age of AI must be figured within the broader context of the ongoing legitimation crises of modern social institutions.
Human AI – Tom Comitta
On December 1 and 2, 2021, novelist Tom Comitta and Johns Hopkins-based survey design expert Katherine Cornwall polled the literary tastes of the United States, measuring everything from favorite genre to setting to characters to verb tense. With the data in hand, Comitta composed two very different novels: one with everything respondents desired (a James Patterson-esque technothriller) and another with everything that received few or no votes (an experimental Christmas novel set on Mars 100 years in the future featuring elements of romance, historical fiction, and horror). This past summer, Columbia University Press published these books in one volume entitled People’s Choice Literature: The Most Wanted and Unwanted Novels. Called “ingenious” and “funny” by Slate and featured in a New York Times profile, People’s Choice Literature is a work of both rigorous research and Wodehouse-worthy humor.
In “Human AI,” Tom Comitta shares their public opinion research on readerly taste as well as the LLM-driven analysis of bestselling novels that inspired everything from the narrative to syntax to characterization of People’s Choice Literature. Accompanying their multimedia presentation will be a brief reading from each book.
Learn more here: https://mahindrahumanities.harvard.edu/event/novel-age-ai-roundtable-elyse-graham-benjamin-mangrum-and-tom-comitta