Crafting a planetary-scale system that watches & forecasts the entire world

Beyond Keywords: AI’s Vision for a Real-Time Global Understanding and Forecasting System

Kalev Leetaru, Founder of GDELT, outlined his vision at Web Summit Vancouver 2026 for a planetary-scale system that uses AI to understand, visualize, and forecast global events. His work focuses on transforming the vast daily tapestry of global information into actionable insights. This involves making the Internet Archive’s extensive television news archive searchable, which comprises 300 channels from 50 countries in over 150 languages, a dataset previously inaccessible.

The breakthrough came with large speech models, enabling the transcription of 3 million hours of uncaptioned television, totaling 14 billion words, even handling multiple languages within a single broadcast. Additionally, 294 billion words were extracted from on-screen text across 19 billion seconds of video. These massive datasets, generating two petabytes of JSON annotations, are then processed by AI models like Gemini to catalog material second by second, moving beyond keyword searches to semantic indexing for complex, non-linear narratives.

For human users, 6 billion seconds of international coverage are translated into English for a remarkably low cost of $74,000, demonstrating AI’s efficiency and accuracy. A key innovation is using AI to discover unknown questions through infographics. These visuals compel models to “reason” and distill vast information into fixed spaces, fundamentally restructuring text into contextual imagery. This capability has been applied to diverse materials, from academic papers and resumes to government laws and multi-hour hearings, enhancing transparency and understanding.

The system scales to summarize entire days of global television coverage or even the US government budget into single infographics, revealing shared planetary stories and unique regional narratives often missed by local media. GDELT now produces daily reports on US legislative activity, available at blog.gproject.org, which summarize trends and forecast future scenarios. This serves as a benchmark for AI reasoning and its potential for comparative analysis, instantly highlighting divergent perspectives from different media sources.

The concept extends to an “agentic think tank in a box,” where AI generates diplomatic reports and backgrounders. In an experiment, AI-generated rebuttals, adopting the persona of the US State Department, were indistinguishable from human-written ones, even identifying connections human analysts missed. While powerful, current models face limitations like token limits, which can lead to incomplete reports without explicit warnings, highlighting the need for continued development.

To overcome these limitations for true planetary-scale reasoning, a “bottom-up” approach is being developed. This involves analyzing data at an “atomic level” using traditional AI techniques like graph, geospatial, and temporal reasoning, before aggregating them. The ultimate goal is a live, real-time model of the entire planet, capable of summarizing everything said globally each day and forecasting future trajectories, uncovering patterns invisible to human analysis and addressing the world’s grandest challenges.