ID: 1489
Presenting Author: Maxime Charles Pierre
Session: 714 - Assessing Information: Conflicting Data Interpretation and Eroding Public Trust
Status: pending
Analyzing the interpretative vulnerabilities in tritium risk models that fuel public distrust. Proposing explicable AI for transparent and defensible assessments.
Tritium risk assessment is plagued by interpretative conflicts between experts and public distrust, exacerbated post-Fukushima. This study deconstructs the evaluation chain, using the ERICA tool to expose critical limitations: oversimplified ratios, omission of organically bound tritium (OBT), and unquantified uncertainties. Results confirm doses remain below thresholds but reveal that model simplifications create false consensus and amplify interpretive vulnerabilities. The conclusion calls for a paradigm shift toward explainable AI (e.g., SHAP) to quantify the influence of each assumption, transforming debate from "who is right?" to "with what confidence?"—enabling transparent dialogue and restoring trust in contested science.
PhD student in radioprotection at UFRJ and expert in risk modeling. His work focuses on scientific communication and the challenges of public trust surrounding tritium.
Coauthor 1: Loudbery Plancher