AI trust frameworks: explainability, human oversight, contestability; were largely designed with professional and enterprise users as the implicit default. When those same frameworks get applied to public-facing systems serving marginalized communities, the UX patterns often break down or actively harm the people they claim to serve.
This talk draws from healthcare AI design, civic data work, and housing justice technology to examine where standard human-in-the-loop patterns fail in practice: when "explainability" produces jargon that a caseworker can't parse and a claimant definitely can't contest; when "human oversight" means a social worker with 200 open cases reviewing an automated flag in under 30 seconds; when "contestability" requires a form, a printer, a notary, and two weeks you don't have.
The problem isn't that these systems lack human-in-the-loop design. It's that the human they designed the loop for was never the person most affected by the decision.
I'll propose an alternative framework, one that treats survival-stakes decision-making as its own UX category, with distinct principles around trust, legibility, and community auditability. This includes: designing for low-trust entry points (assuming the user has been burned before), building contestability that works without institutional access, and shifting the accountability surface from the individual user to the community level.
This talk is for designers, AI practitioners, and technologists building or advocating for publicly accountable systems and for organizers who want language to name what's going wrong in the tools they're already fighting.