Epistemic Power in AI Ethics Labor: Legitimizing Located Complaints

Quantified approaches to AI ethics (like Model Cards) gain legitimacy but may inadvertently silence lived experience: The paper found that AI ethics practices relying on automation and quantification achieve authority and acceptance, while those grounded in "richly embodied and situated lived experience"—particularly from affected communities—are marginalized despite their value.

There's a hierarchy of what "counts" as legitimate AI ethics knowledge: Based on 75 interviews with technologists, the research reveals that certain epistemic bases (ways of knowing) are treated as more authoritative than others, with technical and quantified approaches privileged over experiential knowledge from those directly impacted by AI systems.

Current dominant AI ethics practices may inadvertently reproduce power imbalances: By elevating quantification as the primary legitimate form of AI ethics work, these practices risk creating a false equivalence where technical interventions are seen as equally valid to—or more valid than—the complaints and insights of people experiencing AI harms firsthand.

"Humble technical practices" could help balance epistemic power: The author proposes an alternative approach where technical and quantified AI ethics methods explicitly acknowledge their own limitations, rather than claiming comprehensive authority, which could create space for lived experience to be recognized as legitimate knowledge.

Outside criticism of AI ethics as "progress-blocking" may be driving problematic legitimacy choices: The paper suggests that defensive responses to attacks on AI ethics have pushed the field toward quantifiable, automation-based approaches that appear more "objective" and less obstructive.