EvidenceLens: A Declarative Evidence-State Interface for Human-AI Visual Inspection
Chaewon Yoon
ACM UIST Adjunct 2026 — Poster
DOI: https://doi.org/10.1145/3830397.3841901
EvidenceLens is a declarative evidence-state interface for bounded human-AI visual inspection. Instead of treating model confidence as sufficient for action, it represents task-critical evidence as Satisfied, Weak, Missing, Contradictory, or Unavailable and routes the user toward commit, recovery, repair, or escalation.
In the reported 24-participant study, EvidenceLens reduced inappropriate reliance from 64% to 23% while preserving correct AI acceptance. The systems evaluation also tested robustness under camera, lighting, OCR, and asynchronous-state perturbations.
This repository is a publication and project page. It does not claim to provide a complete software release unless implementation materials are added later.
Chaewon Yoon
chaewon.yoon.ds@gmail.com