Beyond Confidence Scores: Designing Contestable Uncertainty Interfaces for High-Stakes Professional AI

This repository serves as the official reproducibility package and supplementary material for the manuscript "Beyond Confidence Scores: Designing Contestable Uncertainty Interfaces for High-Stakes Professional AI," submitted to the special issue "Uncertainty in the Era of AI" in RSS: Data Science and Artificial Intelligence. The package addresses the critical gap between quantifiable model uncertainty (epistemic) and interpretive uncertainty (hermeneutic) in professional AI systems. It provides the evidentiary basis, design specifications, and operational frameworks necessary to implement "contestable interfaces" that comply with EU AI Act Article 14 (Human Oversight) and mitigate automation bias. Contents: 1. Manuscript: The full peer-review ready paper arguing for the distinction between epistemic and hermeneutic uncertainty. 2. Technical Design Specification: Detailed UI/UX patterns for implementing dual-uncertainty displays (confidence scores + contestability flags). 3. Operational Framework: A guide for aligning professional workflows with EU AI Act Article 14, focusing on override capacity and audit trails. 4. Case Review: An analysis of the Post Office Horizon IT Inquiry, illustrating the transition from software errors to social injustice when contestability is absent. 5. Practitioner Guide: "Beyond the Score" – A condensed guide for developers and board members on implementing uncertainty communication. 6. Audio Briefing: A narrated deep dive summary of the core arguments for accessible consumption. Context: Current AI systems often compress heterogeneous uncertainty types into single confidence scores, encouraging institutional deference. This package provides the tools to separate statistical uncertainty from contested meaning, preserving human hermeneutic responsibility in high-stakes domains like healthcare, law, and hiring.