The disagreement worth reporting
Panels at summits like this usually resolve into agreement. This one did not, and the fault line ran between two senior UN officials on the question the summit exists to answer. Amandeep Singh Gill, the Secretary-General’s envoy on digital technology, said plainly that AI will not move the 2030 goals, and gave the reason.
Annalena Baerbock, presiding over the General Assembly, twice flagged that she was choosing to look at the good side, casting AI as a chance to bridge an old divide by leapfrogging. But she conceded the premise, and her sentence is the one that holds the two positions together.
The divide inside the divide
Brad Smith of Microsoft supplied the adoption numbers, and the second half of his point is the one that usually goes missing.
Gill added the ladder that a connectivity statistic tends to hide, and it is a measurement argument as much as a development one. Being counted as connected is not the same as being able to do anything with it.
What AI has actually changed
The concrete gains were in forecasting and disaster response, and they are not small. Celeste Saulo of the WMO, a meteorologist by training, gave the before-and-after.
Smith described AI and satellite imagery cutting the time to direct rescue teams from days to a few hours. Kamal Kishore of UNDRR set the coverage gap against it: 128 countries, about 65%, have multi-hazard early-warning systems, and the goal is all of them. He also noted the limit of the vocabulary, since earthquakes admit early alerts, not early warnings.
The counting angle
The panel converged, from four directions, on the least glamorous thing in the room. Kishore could not obtain sex-disaggregated death tolls from disasters, eleven years into the Sendai Framework. His prescription was cultural rather than technical.
Smith, whose company sells the models, landed on the same floor, and it is the single best sentence for what this publication tracks.
Gill pushed toward building data commons as global public goods, with governance bifurcated between open public data and protected personal data, and warned that humanitarian data is dual use: a record of who collects food for how many people can identify a large family or a group of fighters. He pointed to the Red Cross digital emblem as the analogue precedent.
Why it matters for the SDGs
The session ranges across SDG 13 (early warning and climate), SDG 9 and SDG 10 (compute, connectivity and the widening adoption gap), and SDG 17 (data, capacity and finance). It is also the clearest statement in the July cluster of a claim this site keeps encountering from different rooms: the binding constraint on using AI for the goals is not model quality but the ground-level statistics the models would have to run on, and the institutional will to gather them. The same argument was made at Science Day in New York two days later, where the missing thing was impact data, and at the statistics and AI side event, where it was the people who never enter a dataset.
Watch & read
- GZERO Media, Global Stage, the series this panel belongs to.
- AI for Good on YouTube, the source recording.
- Together for AI for Good, the summit’s opening keynote.
- Independent International Scientific Panel on AI, whose first report Gill cites · Full AI for Good coverage.
The summit publishes no official transcript. Quotations are lightly edited from an automated (Otter.ai) transcript of the recorded video and should be read as close paraphrase. Speaker names and titles were reconciled against public records; the transcript garbled several. The date is inferred from a panellist’s reference to the Global Dialogue being on its second day. Figures are as cited by the speakers and were not independently verified, except where noted elsewhere on this site.