Communicative surrogates and the ethics of LLM-mediated communication

Abstract As large language model (LLM) interfaces, such as ChatGPT, move into everyday communicative practice, they increasingly occupy roles once reserved for human interlocutors—drafting, advising, and explaining—without clearly signaling when communicative delegation becomes substitution. This paper conceptualizes such systems as communicative surrogates: nonhuman actors that assume communicative labor, while destabilizing the ontological assumptions that ordinarily ground responsibility, authority, and care. Drawing on semi-structured interviews with academic knowledge workers during the early public uptake of ChatGPT, we adopt an abductive, STS-informed analytic orientation that treats participants’ narratives and affective recalibrations—including moments of hesitation, projection, and boundary-setting—as diagnostic signals of sociotechnical instability. Participants were less concerned with correctness than with difficulty determining what kind of communicative participant ChatGPT had become. We show how frictionless usability enables LLMs to drift across communicative contexts, producing ambivalence and unease that undermine the conditions under which ethical accountability can be clearly assigned. Synthesizing these accounts, we identify four figures of communicative surrogacy and argue that communicative ethics is downstream of ontology: ethical concern emerges when it becomes unclear who or what is participating in communication.