Abstract
Artificial intelligence (AI) development in healthcare involves multiple forms of training work, yet data annotation labor faces systematic devaluation despite its structural necessity. Drawing on institutional logics theory as our primary framework, we examine how professional logic, as the dominant logic in AI development, produces gendered hierarchies of expertise through which epistemic positioning determines whose contributions gain recognition. We identify compartmentalization—the allocation of competing logics to distinct organizational domains—as the mechanism producing gendered hierarchies of expertise. Professional logic structures formal evaluation, privileging abstract conception over execution, whereas care logic—whose patient care dimension is invoked to justify AI projects while its data care dimension is systematically excluded from labor valuation—operates asymmetrically within professional logic's evaluative framework. This asymmetric allocation devalues annotation work performed predominantly by junior women as “operational” despite its technical complexity. Through the asymmetric application of authorship criteria and internalization of professional logic's hierarchy of conception over execution, essential data work becomes invisible within formal recognition systems. Our analysis demonstrates how institutional mechanisms reproduce gendered knowledge hierarchies in AI development, echoing historical patterns where data processing was relegated to computing's feminized “machine underclass”—determining which forms of technically complex, structurally necessary labor become invisible.
