Inscribing Knowledge: (Re)Enactment of Expertise and the Credibility of Medical Image Annotators in China's AI Industry

 Data annotation is a critical step in developing machine learning-based AI systems, particularly in supervised learning. In medical image analysis, annotation is widely viewed as requiring substantial clinical expertise and is ideally performed by credentialed professionals, such as radiologists. However, sourcing such experts is costly and challenging, especially given the scale of annotations required. In response, the Chinese medical AI industry has increasingly turned to full-time annotators without formal medical credentials, who are more accessible and cost-effective. How is the credibility of these annotators established, and what constitutes “medical expertise” in this context? Situating data annotation in the broader historical context of AI research, this paper theorizes annotation work as a practice of “knowledge inscription,” where domain expertise is translated into immutable digital labels through ostensive enactments under specific socio-technical networks. Drawing on five months of ethnographic fieldwork at a Chinese medical AI startup and interviews with annotators and radiologists, I further analyze the day-to-day work of medical image annotation as knowledge inscription practices. I argue that annotation involves the (re)enactment of clinical expertise within a distinct socio-technical network that enables uncredentialed professionals to work as credible annotators. This paper concludes by exploring the ethical and normative implications of these practices for deploying AI in high-stakes domains that rely on expert judgment.