New AI Models Developed by MIT Researchers Could Accelerate Trustworthy Climate Prediction

MIT researchers have developed a new technique to streamline the production of fast, trustworthy, and adaptable artificial intelligence models, known as climate emulators, for climate prediction. These specialized AI models can rapidly simulate the effects of human activities on future climate, such as predicting impacts on temperatures or sea level, in a matter of seconds. The new training technique boosts the performance and adaptability of these emulators while requiring only a fraction of the data typically needed, significantly cutting production time. The research demonstrates that an emulator trained with a smaller amount of carefully selected synthetic data can accurately predict outcomes of scenarios it did not encounter during training, unlike traditional emulators that failed to adapt to new situations.