ACM Conference on Fairness, Accountability, and Transparency (FAccT)/2024
Visual Attribution of Model Uncertainty in Autonomous Copilot Systems
Designing visual affordances that communicate algorithmic confidence without degrading workflow velocity or inducing decision paralysis.
Autonomous AI systems frequently fail by projecting unwarranted confidence when generating probabilistic responses. This paper details visual affordances that communicate attribution weights and confidence margins directly inside active user editing states.
Field studies with software engineers and legal researchers demonstrated that interactive confidence underlines enabled operators to spot hallucinated citations 3.4x faster than flat unweighted text presentations.
KEY INSIGHT:
True AI ergonomics does not hide uncertainty; it frames it as an invitation for human judgment.
VENUE / JOURNALACM Conference on Fairness, Accountability, and Transparency (FAccT)
YEAR2024
DOI IDENTIFIER10.1145/3531146.3533190
RESEARCH TOPICS
Explainable AICognitive TrustModel TransparencyHuman-in-the-Loop
