UN turns to Google to make its global data ready for AI agents

The United Nations announced a collaboration with Google to develop the UN System Data Commons, a new platform aimed at making its extensive global statistics readily available for AI systems. This initiative, unveiled on Thursday, replaces the previous UNData portal, which relied on a more conventional database interface for users to browse and search for information.

Built upon Google's open-source Data Commons platform, the new system facilitates natural-language queries for statistics across various UN agencies. It also incorporates the Model Context Protocol (MCP), a standard that enables AI systems to directly connect with external data sources for information retrieval.

Google.org provided $2 million in capacity-building funding and technical support to establish the core infrastructure for the UN System Data Commons. The platform launched with data from nearly 20 UN entities, with a total of 26 committing to the system, and aims to integrate 80% of the UN system's statistical datasets by 2027.

A UNICEF benchmark study, which tested six large language models against over 133,000 questions on global development indicators, revealed an average accuracy score of just 21.2%. Approximately three out of five responses failed to provide a usable numerical answer. Furthermore, models that provided a number on two separate runs returned the identical figure only about half the time.

UNICEF has observed a significant increase in traffic from generative AI assistants to its data website this year, which receives over 6 million monthly visits. Referrals from ChatGPT answers alone saw a 67% year-over-year rise between January 1 and September 14, accounting for 6.4% of all sessions, with AI assistants estimated to contribute to about one in ten visits overall.

The United Nations, specifically its Statistics Division and UNICEF, is leading the initiative, with significant partnership from Google. Shantanu Mukherjee, the acting director of the UN Statistics Division, highlighted the platform's advanced scale, scope, and flexibility in connecting numerous UN agencies and preparing data for AI.

Prem Ramaswami, who leads Google's Data Commons team, stated that the system is hosted on a UN-governed instance and is intended for independent maintenance, operation, and scaling by the UN. João Pedro Azevedo, UNICEF's chief statistician, provided insights into the challenges AI systems face in accurately retrieving authoritative data.

This collaboration addresses the critical need for AI systems to reliably access and utilise authoritative data, a challenge highlighted by the low accuracy rates observed in current large language models. By providing a structured, AI-ready platform, the UN System Data Commons aims to enhance the trustworthiness and utility of AI-driven information retrieval.

The platform's ability to trace statistics back to their original UN sources is crucial for ensuring transparency and accountability as more individuals rely on AI tools for information. It also enables AI systems to directly query data, generate dashboards, charts, and written analyses from multiple indicators, streamlining data interpretation without manual compilation.

Attention will be on the UN's progress in independently maintaining, operating, and scaling the Data Commons platform, as well as its efforts to integrate 80% of the UN system's statistical datasets by 2027. The ongoing development of the Model Context Protocol and its adoption by more AI systems will also be a key area.

Despite the provision of authoritative data, the importance of human review for AI-generated outputs remains paramount. As Google's Prem Ramaswami noted, models can misinterpret nuance, necessitating human oversight before citing or publishing any AI-derived conclusions.