Google Uses Computer Vision to Cut 1% of Global Warming from Contrails
John Platt explains how Google predicts atmospheric supersaturation with computer vision and agentic AI to eliminate airplane contrails.
🔬 Google's AI Scientist Started as an Attempt to Automate Kaggle — John Platt, Google Fellow
With John Platt · Wednesday, September 23, 2026
Google Fellow John Platt discusses Google's Empirical Research Assistance (ERA) framework, an agentic AI system combining LLMs with Monte Carlo tree search to automate scientific experimentation. Platt explains how ERA grew from an attempt to automate Kaggle, the critical distinction between predictive and descriptive scientific models, and how AI is unlocking real-world climate interventions from contrail mitigation to wildfire detection.
Key takeaways
- Aviation contrails contribute roughly 1% of all human-driven global warming by trapping outgoing blackbody infrared radiation. Read more →
- AI coding tools make output fast, but skipping manual implementation prevents researchers from developing scientific taste and intuition. Read more →
- Commercial fusion is roughly three years away from demonstration, likely arriving before 2030 rather than in the proverbial thirty years. Read more →
- Google's Empirical Research Assistance (ERA) framework pairs LLMs with Monte Carlo tree search to automate experimentation, evolving out of attempts to automate Kaggle competitions. Read more →
- Wildfire smoke causes roughly 300,000 excess deaths annually worldwide, according to World Health Organization estimates. Read more →
5 articles from this episode
John Platt on Why AI Tools Cannot Replace Scientific Taste
Google Fellow John Platt explains why relying entirely on AI coding tools robs engineers of domain taste, and why he protects 20% time for deep exploration.
Why Commercial Fusion Is 3 Years Away, Not 30
Google Fellow John Platt explains how the Lawson criterion and simpler control systems bring commercial fusion within reach this decade.
Why AI Science Models Fail Outside Their Training Data
Google Fellow John Platt explains why predictive AI fails at extrapolation and how agentic optimization triggers Goodhart's law in science.
Spotting 5-Meter Wildfires From Orbit With 50 Satellites
John Platt explains how Google and Earth Fire Alliance use 50 LEO satellites and AI super-resolution to spot 5-meter wildfires in 20 minutes.
More Latent Space episodes
- Designing for Agents — Zeno Rocha, ResendNov 15 · 6 articles
- What Frontier Labs Buy: A Data Deep Dive with Handshake AI — Garrett Lord & TeamNov 15 · 5 articles
- Mini Brains, Beating Heart Organoids, and the AI That Predicts Them — Fabian Theis, Helmholtz MunichNov 8 · 4 articles
- Reducto Final YouTube 1080p Sandwich TitleNov 8 · 4 articles