Event Details

Dr Vittorio Tantucci shares his latest research on Chinese AI chatbots

Large language models (LLMs) are part of our everyday communication and will become increasingly so (Hassabis, 2025). Human–AI interaction raises a fundamental pragmatic question: do we interact with AI as if it were human? I approach this question through the Principle of (Im)politeness Reciprocity (Culpeper & Tantucci, 2021), according to which interlocutors tend to reciprocate the perceived politeness or impoliteness of others. I will first examine reciprocity in human-AI interaction, showing that LLMs can ’seek revenge’. They can learn to reciprocate human aggression, sometimes escalating their responses and overriding embedded moral safeguards (Tantucci & Culpeper, 2026). More centrally for this talk, I consider what happens when the direction is reversed: how do humans reciprocate AI behaviour? I will focus on human interactions with CommentR (Yuze et al. forthcoming), a peculiar LLM-based conversational agent that intervenes in users’ posts on Chinese social media platform Weibo. Preliminary findings show that humans reciprocate CommentR’s perceived politeness and impoliteness as if they were actually confronting a conscious human interlocutor. Some cultural idiosyncrasies may also be at work, as Chinese netizens tend to respond quite enthusiastically to creatively poetic outputs by the LLM, a pattern that may be less prominent on Western social media. Based on this pilot data, I speculate that, as human-AI interactions and dialogues will become increasingly entrenched and conventionalised in our everyday lives, our awareness of AI being an agent without consciousness may progressively decrease (Yuze et al. forthcoming). This hypothesis partly draws on the so-called fundamental attribution error (Ross, 1977; Culpeper, 2005): whatever the context, with marked (im)politeness, communicators often ignore the contextual conditions leading to markedly (im)polite behaviours.

Dr Tantucci's study focuses on interactions with CommentR, a conversational agent on Weibo, and examines how Chinese netizens respond to its unsolicited comments in remarkably similar ways to how they would respond to a human interlocutor. CommentR’s language can also be strikingly creative, at times echoing classical Chinese poetry and prompting strongly appreciative reactions from users. These and other patterns in the management of politeness and reciprocity in human–AI interaction appear to be quite context-specific.