Send message to Author

Paper short abstract

Interviews with 25 Chinese AI mental-health developers show how “care” is translated into scalable data practices. Distress becomes measurable “pain points,” and success shifts to retention. Platform infrastructures embed care while redistributing exclusion and responsibility.

Paper long abstract

Across China’s rapidly expanding AI mental-health sector, “care” is increasingly mediated through data infrastructures rather than professional institutions. Drawing on 25 in-depth interviews with developers (founders, product managers, engineers, and researchers) and analysis of product documentation, this paper examines how developers translate care into forms compatible with platform scaling. I show, first, how care is redefined as a user’s subjective “feeling cared for,” which then demands an objective trigger that can be designed, logged, and optimised. In practice, developers name this trigger “pain points”: user distress is rendered into stable categories that can enter roadmaps and A/B tests, allowing care to be embedded materially in interaction scripts and metric dashboards rather than in situated tinkering. Second, I trace how success is re-measured when care must scale. A clinical “arch” trajectory—users improving and leaving—becomes commercially non-viable, while a “wave” model reframes care as cyclical return, positioning AI as an always-available infrastructure users repeatedly draw on. This shift makes relational attention measurable (retention, engagement) and thus governable, but it also reorganises exclusion and responsibility: forms of suffering that do not map neatly onto shared pain points, or that require moment-by-moment adjustment, are structurally sidelined; users are expected to self-manage within preconfigured pathways. By foregrounding translation as a communicative care practice—turning lived vulnerability into actionable data—I argue that algorithmic care infrastructures perform a politics of care and data justice: they distribute who can be cared for, what counts as care, and where obligations land unevenly, across individual, organisational, regulatory, and societal scales