Models based on physical principles continue to be superior to artificial intelligence (AI) models for predicting extreme weather events. The main limitation of AI lies in its dependence on historical training data; if an event is unprecedented in this data, AI models struggle to forecast it accurately. This finding underscores the fundamental importance of understanding the underlying physics in climate modeling, especially in the face of a changing climate where unprecedented phenomena are increasingly likely.

Physical models, in contrast, build their predictions from equations describing atmospheric and oceanic processes, such as fluid dynamics, thermodynamics, and radiation transfer. This approach allows them to simulate conditions that have never been directly observed, extrapolating from the fundamental laws of nature. Although AI has proven very effective in identifying patterns and optimizing processes within known ranges, its ability to generalize to completely new scenarios is limited, making it less robust for predicting climatic "black swans."

This analysis suggests that, while AI can complement and improve certain aspects of climate modeling (e.g., in data assimilation or bias correction), it cannot replace the physical basis for predicting extreme events. To address the challenges of climate change and its unpredictable consequences, it is crucial to continue investing in the development and improvement of detailed physical models, which are the only ones capable of offering reliable prospective insights in historically unprecedented situations.