Answers are grounded in this paper and cite their sources.
Scholars have proposed ethical principles and frameworks to support industry practitioners developing responsible AI. To enable these from theory to practice, a subset of research explores how practitioners actually enact abstract ethical concepts and guidance, surfacing factors that limit their efforts—e.g., organizational support—and informing ways to strengthen efforts towards safe and ethical AI in practice—e.g., practical checklists, educational workbooks. Curiously, this has been underexplored in relation to practitioners curating ethical datasets, the very foundation needed for fair ML technologies, despite the existence of various data ethics principles. It is unsurprising, then, that reports continue to emerge of ML datasets curated through questionable practices, including non-consensual web-scraping and exploitative labor practices. To address this gap, we describe the experiences of a project team curating FHIBE: a fair, human-centric benchmark dataset for computer vision technologies, grounded in recommendations by data ethics and fair ML scholars. Using an ethnographic lens, we share the team’s motivation and the set of principles FHIBE was based on, the challenges team members faced trying to execute multiple principles at once, and the ways they adapted in lieu of resources or insights from preceding scholarship. By tracing one team’s end-to-end data curation efforts, we offer new insights into practitioners’ realities translating data ethics, to inform valuable avenues for HCI and policy researchers to address in the future.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: