Azernews.Az

The United Nations and Google have launched the UN System Data Commons, an open-source platform designed to bring together global statistics from across the UN system in one place. Available through the UN’s data portal, the project aims to replace the outdated UNData platform and make verified statistical information easier to find, analyze and use.

One of the main reasons behind the initiative is the growing difficulty of obtaining reliable statistics from artificial intelligence systems. A UNICEF study that analyzed more than 133,000 responses from six AI models found that their average accuracy when answering statistical questions was only 21.2%.

The new platform brings indicators, time series and geographic data into a unified system. Users can search for information using natural language and receive interactive charts and visualizations. The datasets are reviewed by technical experts and UN statisticians, providing an additional layer of quality control.

Google has provided the UN Foundation with a $2 million grant, along with technical assistance and infrastructure support. The platform also supports the Model Context Protocol (MCP), allowing AI agents to connect directly to external data sources and use official statistics when generating analyses, reports and visual materials.

At launch, 26 UN agencies had joined the initiative, with around 20 already contributing data. The UN plans to integrate approximately 80% of its statistical datasets into the platform by 2027.

Interestingly, the project could become particularly important as AI-generated content becomes more common. Instead of relying solely on information learned during training, AI systems will increasingly be able to retrieve statistics directly from authoritative sources. This could help reduce one of the most common problems with AI-generated reports: confident-looking answers based on incorrect or outdated numbers.

At the same time, the developers stress that AI is not a substitute for human verification. For important publications, users are still encouraged to check the original datasets and sources before relying on or publishing the information.