(How) Do Machines Make Sense? Respecifying Meaning in LLM-based Conversational AI

In this paper I revisit the question of whether machines "make sense" by respecifying meaning through an ethnomethodological perspective. Rather than treating meaning as a cognitive state or semantic property, ethnomethodology conceptualizes it as an emergent accomplishment of social interaction. Drawing on this approach, the paper argues that large language model (LLM)–based conversational AI should be evaluated not in terms of whether it “truly” understands or produces “genuinely” meaningful utterances, but in terms of how it participates in producing, maintaining, and repairing interactional order. While machines still struggle with tasks such as common sense reasoning or grasping irony and humor, they excel at one thing: maintaining the flow of conversation even at the cost of "bullshitting". In this paper, I argue that this emergent behavior, often dismissed as a failure, should instead be understood as a distinctive mode of engagement that signals contemporary artificial agents’ capacity to participate in meaning-making. A “mechnomethodology” shifts attention from how humans make sense of AI to how artificial agents make sense of, to, and with humans and other machines, thereby encouraging empirical investigation of machines’ meaning-making rather than metaphysical speculation about internal cognition or the status of their utterances.