Risks posed by AI actants in human-AI health communication: AI health assistants facing HIV-related stigma inquiries

This study conceptualizes AI health assistants as agentic non-human actants and examines how a mainstream Chinese AI health assistant responds to HIV-related stigmatizing inquiries. The findings reveal that in 22% of its responses, the AI reproduced or exacerbated both public and self-stigma by using structured interaction scripts—predefined conversational patterns designed to guide the dialogue. Notably, a key finding is the presence of an asymmetric emotional modulation mechanism in the AI's replies: when patients expressed anger or sadness, the AI tended to provide destigmatizing support; however, when similar emotions were conveyed by public users, its responses often reinforced discriminatory discourse. This indicates that even as the AI mitigates stigma at the individual level for patients, it simultaneously solidifies social prejudice at the public level. Thus, the study demonstrates that specialized AI health assistants actively translate social relations and bureaucratic classifications into conversational norms through algorithmic dialogue, thereby institutionalizing dominant moral orders surrounding disease, sexuality, welfare, and productivity.