Irony machines: Artificial Intelligence, literary language, and the opacities of trust

This essay begins with an exploration of recent efforts among computational scientists to create Large Language Models (LLMs) capable of detecting irony. It notes the difficulty, if not impossibility, of this task, given that LLMs generate natural language responses to prompts based on statistical analysis of colossal caches of explicit statements, while irony involves saying one thing but meaning something else. It proceeds to develop a richer conception of irony through an exploration of two approaches: the neo-Romantic deconstruction approach represented by Paul de Man and a pragmatic speech act theory inspired approach represented by John Searle. It seeks to show that irony is not one trope among many, but a condition of all language and all social relations. Drawing on the work of the classical sociologist Georg Simmel and the contemporary cultural critic Byung-Chul Han, it develops a theory of asocial sociability, or the enigmatic sense in which humans are held together by being held apart.