Responsible AI in Statistical Production: From Methods to Trusted Practice
Conference
Proposal Description
This session explores the integration of Artificial Intelligence in official statistical production, focusing on the principles and practices of Responsible AI. Contributions examine how to translate key concepts—such as transparency, reliability, fairness, and accountability—into concrete methodological choices across the AI lifecycle. Emphasis is placed on structured frameworks, including the Total Machine Learning Error, to ensure quality and robustness. The session also addresses institutional and organizational challenges for National Statistical Offices, alongside innovative approaches like Symbolic Data Analysis for constructing interpretable indicators. Practical experiences highlight solutions for validation, data governance, confidentiality, and trust-building in AI-driven statistics.