Scaffolding End-User Auditing of Generative AI Systems
A workflow and interactive system (WeAudit) that structures how everyday users collectively find, deliberate on, and report perceived harms and biases in generative AI, and translates those findings into a form industry practitioners can act on. Students learn a set of design goals for user-engaged AI auditing drawn from CSCW crowdsourcing and sensemaking research, and how each auditing stage - comparing generated outputs, reflecting on one's own and others' findings, and structured reporting/verification - can be supported so lay users produce actionable, credible audit reports of systems like text-to-image models.
WeAudit: Scaffolding User Auditors and AI Practitioners in Auditing Generative AI
End users often surface harmful or biased behaviors in AI systems that practitioners miss, but there is little guidance on how to scaffold that user auditing so it yields insights practitioners can a…