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

The United Nations connected Thursday announced that it is moving with Google to marque its immense postulation of planetary statistic easier for AI systems to entree and use.

Called the UN System Data Commons, the caller strategy is built connected Google’s open-source Data Commons level and lets radical hunt for statistic from crossed UN agencies utilizing natural-language queries. It replaces the existing UNData portal, wherever users mostly had to browse and hunt for statistic done a much accepted database interface. The caller level besides supports the Model Context Protocol (MCP), a modular that allows AI systems to link straight to outer information sources.

Users progressively crook to AI tools for answers, but galore systems inactive conflict to reliably aboveground authoritative data. A UNICEF benchmark of six ample connection models crossed much than 133,000 responses to questions astir planetary improvement indicators produced an mean accuracy people of conscionable 21.2%, João Pedro Azevedo, the agency’s main statistician, told reporters successful a virtual briefing.

The trial covered OpenAI’s GPT-4o and GPT-4o-mini, Anthropic’s Claude Sonnet 4.5 and Haiku 4.5, and Google’s Gemini 2.5 Flash and Gemini 2.0 Flash, Azevedo told TechCrunch.

About 3 successful 5 responses did not supply a usable fig astatine all, often due to the fact that the models hedged their answers, Azevedo said. However, erstwhile the aforesaid questions were tally again connected the aforesaid exemplary versions astir 2 days later, models that provided a fig some times returned the identical fig lone astir fractional the time.

The survey is simply a UNICEF moving insubstantial being prepared for diary submission and has not yet been peer-reviewed. The enactment said it plans to merchandise its methodology, code, and information alongside the paper.

UNICEF has besides seen a crisp emergence this twelvemonth successful postulation from generative AI assistants to its information website, which receives much than six cardinal visits a period and is among the agency’s astir fashionable websites. Visits from users clicking links successful ChatGPT answers to the tract roseate 67% twelvemonth implicit twelvemonth betwixt January 1 and September 14, Azevedo told TechCrunch. Such referrals accounted for 6.4% of each sessions this year, portion UNICEF estimates that AI assistants wide present relationship for astir 1 successful 10 visits.

The UN said 26 of its entities person committed to the Data Commons, with information from astir 20 disposable astatine launch. Moreover, it aims to bring 80% of the UN system’s statistical datasets onto the level by 2027.

UN System Data CommonsImage Credits:Google

“We are orders of magnitude much precocious successful scale, scope, and flexibility, connecting for the archetypal clip crossed truthful galore agencies crossed the UN system,” said Shantanu Mukherjee, acting manager of the UN Statistics Division. “And [we are] taking this infinitesimal to besides marque our information AI-ready.”

Google.org provided $2 cardinal successful capacity-building backing and method enactment to found the platform’s halfway infrastructure. Prem Ramaswamy, who leads Google’s Data Commons team, told TechCrunch that the strategy is hosted connected a UN-governed lawsuit and is intended to yet beryllium maintained, operated, and scaled independently by the UN.

“We person taken a “train-the-trainer” attack passim the rollout, and we person already seen the UN strategy squad ramp up quickly,” Ramaswamy said.

Google launched Data Commons successful 2018 arsenic an effort to signifier nationalist datasets from antithetic sources into a communal framework. Last year, it added enactment for MCP, allowing AI agents to straight query Data Commons for statistic and their sources.

The UN’s level besides keeps way of wherever each statistic comes from, truthful radical tin hint information retrieved by an AI strategy backmost to the archetypal UN source. Azevedo told reporters that it was important arsenic much radical trust connected AI tools to find and construe information.

Alongside enabling AI agents to retrieve idiosyncratic statistics, Google demonstrated however an AI strategy connected to the UN information done MCP could propulsion unneurotic aggregate indicators and usage them to make dashboards, charts, and written investigation without a idiosyncratic having to manually find and harvester the underlying datasets.

In 1 demonstration, Google asked an AI strategy to find the interaction of the U.S. President’s Emergency Plan for AIDS Relief successful Africa. The strategy identified applicable UN statistic connected measures including HIV infections, AIDS mortality and beingness expectancy, and utilized them to nutrient an infographic.

However, giving an AI strategy authoritative information does not needfully marque its conclusions authoritative. “Because models tin misinterpret nuance, a quality should ever reappraisal the outputs earlier citing oregon publishing them,” Ramaswamy said.