The United Nations Teams Up With Google to Transform Public Data.

The United Nations announced on Thursday that it is collaborating with Google to organize its vast collection of global statistical data into a format that AI systems can directly read and invoke.

Image source: Generated by Doubao AI

The new system, called the UN System Data Commons, is essentially built on Google's open-source Data Commons platform.

You can search for data from various UN agencies in plain language, without having to navigate through layers of databases as in the old UNData portal.

The new system also supports the Model Context Protocol (MCP), a standard protocol that allows AI to connect directly to external data sources.

Why Go Through All This Trouble?

Because nowadays, everyone turns to AI for answers first, but whether the data provided by AI is reliable is another matter.

Joo Pedro Azevedo, Chief Statistician at UNICEF, calculated during an online briefing with reporters: They tested six major models with over 133,000 responses to questions related to global development indicators, and the average accuracy rate was only 21.2%.

The models tested may sound familiar: OpenAI's GPT-4o and GPT-4o-mini, Anthropic's Claude Sonnet 4.5 and Haiku 4.5, as well as Google's own Gemini 2.5 Flash and 2.0 Flash.

Azevedo said that about three-fifths of the responses failed to provide a usable number, with the models being vague.

Even more absurdly, when the same set of questions was asked again two days later, only about half of the models that provided numbers both times returned the same figures.

Image source: Generated by Doubao AI

This test is currently a working paper by UNICEF, pending submission to a journal and has not yet undergone peer review.

They said that when the paper is published, the methodology, code, and data will be made publicly available.

UNICEF has also noticed this year that an increasing number of generative AI assistants are directing traffic to its data website.

The website is already quite popular, with over 6 million monthly visits, making it one of UNICEF's most visited sites.

Visits from links in ChatGPT responses alone saw a 67% increase from January 1 to September 14, compared to the same period last year.

Azevedo stated that from January 1 to September 14, the number of visits to the website from links in ChatGPT responses increased by 67% compared to the same period last year.

This year, 6.4% of all visits came through these links.

UNICEF estimates that when including various AI assistants, about 1 in 10 visits now comes from AI sources.

Progress on the platform side is also swift. The United Nations said that 26 of its agencies have already joined this Data Commons, with data from nearly 20 agencies available at launch. The goal is to migrate 80% of the UN system's statistical datasets by 2027.

Shantanu Mukherjee, Acting Director of the UN Statistics Division, put it bluntly: This initiative is on a completely different scale in terms of size, coverage, and flexibility compared to previous efforts, and it is the first time that data from so many UN agencies has been connected. Seizing this opportunity, the data is being reformatted for direct AI use.

Google has contributed $2 million through Google.org, primarily for capacity building and technical support to establish the platform's core infrastructure.

However, Prem Ramaswami, head of Google's Data Commons team, revealed that the system runs on instances managed by the United Nations, which will ultimately maintain, operate, and scale the platform themselves.

They have adopted an approach of 'training a few first, then having them train more,' and the UN team is already picking up the skills quickly.

Google itself launched Data Commons in 2018, aiming to consolidate scattered public datasets into a single framework.

Image source: Generated by Doubao AI

Last year, Google added MCP support, allowing AI agents to directly query data and its sources.

The platform also keeps track of where each piece of data comes from, making it easy for AI to trace information back to the original UN data sources.

Azevedo emphasized to reporters that with more people relying on AI to find and read information, traceability has become crucial.

Beyond querying individual numbers, Google demonstrated that AI connected to UN data via MCP can automatically pull together multiple indicators to generate dashboards, charts, and even textual analyses, eliminating the need for users to manually sift through various datasets.

During the demonstration, an example was given where AI was asked to evaluate the impact of the U.S. President's Emergency Plan for AIDS Relief (PEPFAR) in Africa. The system automatically selected relevant indicators such as HIV infection rates, AIDS-related deaths, and life expectancy, and generated an infographic.

However, providing authoritative data to AI does not necessarily guarantee that the conclusions drawn by AI will be authoritative.

"Because models may misinterpret subtle nuances, it is still advisable to have human oversight before citing or publishing," Ramaswami said.

Source: Leikeji

Images in this article are from: 123RF Licensed Image Library Source: Leikeji