New research could lead to faster, trustworthy AI models for climate prediction

What happened: MIT researchers developed a training technique for climate emulators, AI models that rapidly simulate climate impacts, using a fraction of the data typically required. The research appears in the Proceedings of the National Academy of Sciences. Why it matters: The researchers report that emulators trained with their method accurately predicted scenarios not seen during training, while traditional emulators failed to adapt. They say this could make emulator outputs more trustworthy for policy decisions.

MIT researchers developed a training technique for climate emulators, AI models that rapidly simulate climate impacts, using a fraction of the data typically required. The research appears in the Proceedings of the National Academy of Sciences.

The researchers report that emulators trained with their method accurately predicted scenarios not seen during training, while traditional emulators failed to adapt. They say this could make emulator outputs more trustworthy for policy decisions.