August 24, 2026 · MIT News · Nature Communications
MIT uses AI to imagine disasters that have not yet occurred.
Researchers at the MIT have presented a method of automatic learning to generate plausible extreme scenarios although there are no previous examples equally grave in the data.
- Research
- Climate risk
- Infrastructure
- Modeling
Source reviewed
Primary source: MIT NewsPublished on August 24, 2026. The work is supported by an open article in Nature Communications.
What the method proposes
The system learns statistical patterns from normal data, for example daily rain maps, and generates plausible extreme scenarios: storms, fires, heat waves or even rare financial situations.
Why is news
Many critical decisions are taken by asking what would happen in a one-time event every 100 years. The problem is that such events almost never appear in the data. This investigation tries to model them without relying on having seen them before.
The reading Clarumia
It's not an app for normal users, but it shows where useful AI is headed: less flashy demo and more tool for planning real risks in cities, energy, supply chains, and markets.