Agenda

Large language models and generative AI increasingly shape how knowledge is used and shared, but their reliability depends on the quality and provenance of training data. As AI expands into governance, public services and markets, identifying trustworthy, well-documented sources has become a global priority.

Official statistics provide a key public good: they are produced under professional standards, transparent methodologies and public oversight, drawing on administrative records, surveys and privately held data. They offer essential ground truth for validating, calibrating and benchmarking AI outputs. In a context of growing data volumes and uneven quality, authoritative statistical datasets help verify results and ensure consistency with established evidence.

Combining diverse data for AI raises challenges around quality assurance, metadata, representativeness, bias, intellectual property and privacy. The WSIS+20 and GDC processes emphasize shared approaches for safe, inclusive and interoperable digital ecosystems.

This session will examine how statistical and geospatial communities, private companies, NGOs and initiatives such as the Financing for Development work on the Future of Data and the Trusted Data Observatory can develop joint action on “better data for AI.” It will explore principles for identifying, curating and sharing high-quality datasets and metadata. Experts from national statistical offices, international organizations, academia and the private sector will discuss roles of official and private data, practices for documenting AI-ready datasets, innovations in AI-for-data, and opportunities for cooperation to ensure the next generation of AI is built on trusted, accountable and globally beneficial data foundations.