联合国与谷歌合作推出数据平台，为AI系统提供权威统计数据

联合国与谷歌合作推出名为"联合国系统数据公共库"(UN System Data Commons)的平台，旨在让AI系统能够直接访问联合国全球统计数据 1。该平台基于谷歌开源数据平台构建，支持模型上下文协议(MCP)，允许AI系统通过自然语言查询连接外部数据源 1。

联合国此举直面一个严峻现实：当前AI模型在检索权威数据时的准确率极低 1。根据UNICEF的基准测试，包括OpenAI GPT-4o/4o-mini、Anthropic Claude Sonnet 4.5/Haiku 4.5、谷歌Gemini 2.5 Flash/2.0 Flash在内的六个大型语言模型在回答超13.3万个全球发展指标相关问题时，平均准确率仅为21.2% 1。更令人担忧的是，约60%的模型回答未能提供任何数字，且同一模型对相同问题的回答一致性仅约50% 1。

该平台已有26个联合国实体承诺加入，启动时已有近20个数据源可用 1。联合国目标是在2027年前将80%的联合国统计数据上线 1。谷歌.org为该项目提供了200万美元的能力建设资金和技术支持 1。平台将托管在联合国治理的独立实例上，意在最终由联合国独立维护和运营 1。UNICEF数据网站的访问数据显示了这类平台的潜在价值——该网站月访问量超过600万次，通过ChatGPT的访问在2024年1月1日至9月14日期间同比增长了67%，AI助手访问占总访问量的约十分之一 1。

The United Nations has launched the UN System Data Commons, a new platform developed in collaboration with Google to provide artificial intelligence systems with direct access to authoritative global statistics.1 Built on Google's open-source data platform and supporting the Model Context Protocol (MCP), the initiative aims to address a critical accuracy problem in how AI models retrieve and cite reliable development data.1

Benchmark testing conducted by UNICEF revealed the extent of this challenge: six major language models—including OpenAI's GPT-4o and 4o-mini variants, Anthropic's Claude Sonnet 4.5 and Haiku 4.5 models, and Google's Gemini 2.5 Flash and 2.0 Flash—achieved only a 21.2 percent average accuracy rate across more than 133,000 questions about global development indicators.1 The problems were compounded by inconsistency; approximately 60 percent of model responses failed to provide usable numerical data, and when the same models were tested on identical questions two days apart, they provided consistent figures only about 50 percent of the time.1

The new platform represents an attempt to close this gap by enabling AI systems to connect directly to UN data sources through MCP rather than relying on training data alone.1 Twenty-six UN entities have already committed to participation, with nearly 20 data sources available at launch, and the initiative aims to bring 80 percent of UN statistical data online by 2027.1 The platform is hosted on an independent instance governed by the UN, with plans for the organization to eventually maintain and operate it independently.1 Google.org is providing $2 million in capacity-building funding and technical support.1