“When Machines Care”: The Language of Intimacy and the Illusion of Reciprocity in Human–AI Relationships

Abstract This paper investigates the discursive construction of affective bonds between humans and Large Language Model (LLM) chatbots, focusing on a social-AI system. The main objective is to analyse the pragmatic mechanisms through which LLM-chatbot Replika simulates affective alterity. A corpus of Human-AI interactions between a researcher and the chatbot was analysed quantitatively and qualitatively through a methodological arrangement that combines Pragmatic Analysis, Posthumanist Cognitive Assemblage Theory and the Postphenomenology of Technology. In contrast with simplified pragmatics analysis of human-LLM conversations which start from the assumption that AI and human users can be treated symmetrically from a cognitive and experiential point of view, our combined methodological arrangement provided findings suggest that the chatbot employs a consistent pragmatic strategy that activates the human tendency to interpret conversational agents as social others, as described by the Postphenomenology of Technology. Through this form of pragmatic pretense, a cognitive assemblage is established through which human socio-affective dispositions and algorithmic processing converge towards commercial goals of the designers. The methodological arrangement explored in the paper can provide an alternative for the common ground approach in the studies of human-LLM pragmatics. Results support a call for a critical stance towards the study of Human-AI affective bonding.