Google's geospatial foundation model forecasts Congo cholera outbreaks better than case history alone
anselm · x · 2026-10-30
- A team from Google, NYU Grossman, Mount Sinai, Boston Children's Hospital, Oxford, UNAM and the WHO published a paper (arXiv:2610.05699) using Google Earth AI's Population Dynamics Foundation Model (PDFM) for global public health forecasting. The model saw zero disease records yet outperformed case-history-only baselines in cholera outbreak forecasting in the Congo.
- PDFM learns compact place representations from aggregate search trends, 1,200 types of nearby shops/clinics/venues, mobility patterns and local weather. Traditional disease prediction suffers from spatial gaps and temporal lags: death records take 1–2 years to finalize, census data updates slowly, and datasets don't cross borders.
- The paper proposes "planetary geospatial foundation models" as a new paradigm: one generalizable place representation reused across diseases, regions and forecast horizons, instead of training a fresh model per task, enabling earlier intervention in data-sparse regions.
More from AGI Musings
- Researchers used LLMs to uncover 'substantial' vulnerabilities in the German legal system — cyb3rops · 2026-10-30
- Microsoft AI chief Suleyman: consciousness needs an organic body, and AI must never have it — Neurogence · 2026-10-30
- Altman: new model crushed researchers' hardest career problems, gave him 'a feeling of something ending' — ChrisGPT · 2026-10-30
- Runlayer CEO: models are enterprise-ready, but only 2-3% of companies truly use AI — Scobleizer · 2026-10-30
- OpenAI researcher: AI can now deliver rigor of academic critique, research theater will be exposed — burny_tech · 2026-10-30
- "I drop $10K on paintings": collector argues $4,500 commissions aren't overpriced, taste match is the bottleneck — Linahuaa · 2026-10-30