ITU AI for Good: Early Warning for All — Leveraging AI to Identify and Reach Unconnected Communities

The ITU AI for Good platform featured a session on April 22 addressing one of the most urgent gaps in AI-powered disaster preparedness: how to extend effective early warning coverage to the billions of people in climate-vulnerable communities who remain unreachable through standard digital alert systems. The UN 'Early Warnings for All' initiative — jointly led by WMO, UNDRR, ITU, and the International Federation of Red Cross — has a target of universal early warning coverage by 2027, with AI playing a central role in expanding coverage and improving forecast accuracy. Despite AI flood, cyclone, and drought prediction systems now covering 80+ countries and delivering alerts to 460M+ people, an estimated 2.5 billion people in the most climate-exposed regions — including rural sub-Saharan Africa, the Pacific islands, and remote South Asian communities — remain effectively unreachable through standard digital alert delivery. The session reviewed how AI models can identify populations likely to be offline during extreme weather events (based on connectivity data, socioeconomic indicators, and geographic isolation), enabling targeted investment in non-digital last-mile alert systems including community radio, loudspeaker networks, and trained community disaster responders. A key challenge discussed was data scarcity in the most vulnerable regions: AI forecasting models trained on data-rich environments often underperform in data-sparse regions, precisely where early warning is most critical. The session referenced the WMO Multi-Hazard Early Warning System guidelines and ITU 'Connecting the Unconnected' connectivity assessments as foundational frameworks. Pre-summit programming for the July 2026 AI for Good Global Summit in Geneva is using these sessions to build global consensus on AI early warning equity standards.