Lifetime Citizen Portal Access — AI Briefings, Alerts & Unlimited Follows
U.N. panel warns concentrated training data risks bias; calls for data justice and interoperable governance
Loading...
Summary
A high-level U.N. panel on data governance said major AI training sets are heavily skewed toward North America and Western Europe, urged capacity building and data justice, and tasked a multi-stakeholder working group to craft governance approaches that preserve rights while enabling interoperable data flows.
A U.N. panel convened at Action Day focused on data governance for artificial intelligence and warned that geographic and cultural concentration of training data risks producing systems that do not serve the majority of the world's population.
"In some major data sets used to train artificial intelligence, over 80% of the data comes from North America and Western Europe," the panel moderator said, summarizing evidence raised during the session. Panelists argued that such concentration can embed bias and exclusion into AI systems and urged action on data justice, capacity building and interoperable frameworks.
Panelists said steps to address the problem include improving public data collection and digital statistics, expanding digital public infrastructure, and strengthening local capacity so countries and communities can both produce and govern data. Bruno Bioni said governance must move from an individual privacy perspective toward collective and social approaches, and he flagged data quality and missing metadata as obstacles to accountability.
Lisonbee Gillwold of Research ICT Africa argued that data should be treated as a public good and regulated so communities see direct benefits. "The primary problems with digital inclusion are often on the demand side," she said, noting that coverage does not always translate into meaningful access or representation in datasets.
Peter Major, vice chair of the U.N. Commission on Science and Technology for Development and chair of the commissionworking group on data governance, said the group must establish shared definitions and a common language as a first step and encouraged an inclusive, bottom-up, multi-stakeholder process. "We have to be extremely careful not to build new divides," he said.
Speakers gave regional examples and policy options. Brazilwas cited for creating sectoral regulatory ecosystems and Saudi Arabia for investing in large-scale training and incentives. Panelists also discussed the difference between technical interoperability (standards and APIs) and policy interoperability (compatible legal regimes) and warned that both are needed to enable cross-border data use without entrenching existing inequalities.
Ending: The panel concluded with calls for a sustained, multi-year effort to combine standards, regional coordination and finance to make data more representative and usable for AI that serves diverse populations. Several panelists urged that data governance be iterative: a framework put forward today must be routinely updated to reflect technological change and new evidence.
