Closing the Loop: Relocated Expertise and Building as Anthropological Practice

This concluding commentary to the Human Organization special issue on anthropology and AI takes up the practical half of the question the issue opened with: what it would take for anthropologists to design the loop rather than only diagnose it. Building computational tools is now within practical reach of anthropologists, not because the discipline has produced programmers, but because large language models relocate the expertise required to build them. Drawing on Collins and Evans, the commentary argues that building now demands interactional expertise, enough fluency to specify what one wants, recognize when the output is wrong, and revise until it matches, combined with contributory expertise in one’s own domain. It works through a case. After finding that only 59 percent of the Society for Applied Anthropology’s articles had a record in Wikidata, the author built a pipeline that downloaded the SfAA’s record from Crossref and populated Wikidata, and a successor effort now applies the same practice across hundreds of anthropology journals. Two claims follow. Anthropology should build its own tools rather than relying on or critiquing what others build, and making something is itself a way of knowing, not only a way of applying knowledge settled elsewhere.