AI for Good Global Summit 2026

I was back at the AI for Good Global Summit in Geneva for the second year running — five days at the ITU's annual gathering where governments, researchers, civil society, and industry come together around one question: how do we make AI actually useful for the planet and the people most left behind? It's one of the few spaces where that question is asked seriously, and where the SDGs feel like a real design constraint rather than a slide deck talking point.

GeoAI finally got its moment

What struck me most was how much space geospatial AI had claimed compared to previous editions. The GeoAI for our shared future session on Tuesday was packed — standing room — and the conversation had clearly matured: we weren't debating whether geospatial foundation models are coming, but what it'll take to make them useful beyond the well-resourced institutions that can already afford frontier tooling. Friday's half-day on GeoGPT and OneEarth felt like a glimpse of where things are heading — multimodal models pulling together satellite imagery, remote sensing, and geological surveys into something like a shared planetary knowledge layer. As someone who spends a lot of time thinking about open geospatial data, that resonated deeply.

Sessions that stayed with me

• Fail and grow: why failure stories matter in AI development — a refreshingly honest session. The argument was simple: if we only share success stories, we slow everyone down. Development contexts especially need the permission to fail fast and learn openly.
• Advancing responsible AI: open source for inclusive development and collaborative governance — the governance session I'd been looking for. Who gets to set the rules for AI systems deployed in humanitarian contexts? Open source isn't a silver bullet, but it's the best lever we have for keeping those decisions accountable.
• The Early Warning Connectivity Map in action (ITU / Microsoft AI for Good Lab / Planet Labs / IHME) — this one hit close to home. Missing data isn't a footnote in early warning systems — it's the central problem. The data point that stuck with me: roughly 7 countries and 39 years of losses-and-damage records, covering a fraction of what we'd need. The crowd-sourced connectivity angle reminded me of exactly why OSM matters in these conversations.
• GeoGPT and OneEarth: empowering global geoscientists — a strong close to the week, and further evidence that geo is no longer peripheral to the AI conversation.