Do chatbots dream of AI sheep? A semantic–pragmatic investigation of "naturalness" in human–AI interaction

Abstract This paper investigates what constitutes “natural” interaction between humans and AI chatbots, challenging the prevailing assumption that naturalness must equate to human-likeness. Drawing on key concepts of semantics and pragmatics, we analyze interactions with two state-of-the-art systems, namely, ChatGPT and Claude. Through qualitative analysis of prompt–response interactions, we assess how these systems manage meaning and social nuance. Our findings highlight significant limitations in AI’s pragmatic competence, especially in grounding communication and interpreting implicit cues. We argue that naturalness in human–AI interaction is not a fixed standard based on human norms, but a socially negotiated and context-dependent construct. Instead of designing AI to imitate human behavior perfectly, we advocate for redefining naturalness in terms of communicative functionality, user expectations, and adaptive interactional strategies. This reconceptualization foregrounds the importance of designing AI systems that foster trust, clarity, and usability within their specific social roles.