The Weather and Climate Science AI Revolution Isn't Revolutionary
Original: The weather and climate science AI revolution isn’t revolutionary
AI speeds up weather forecasting but faces limits with extreme events and physical laws, making it an evolutionary tool rather than a revolution.
While AI models like Google's GraphCast have dramatically accelerated weather forecasting, experts argue the "AI revolution" in climate science is overstated. Machine learning models struggle with unprecedented extreme events due to their reliance on historical training data, and they often violate fundamental physical laws. Consequently, AI is currently acting as an emulator to speed up traditional physics-based models rather than replacing them, pointing toward a hybrid future.
In recent years, artificial intelligence (AI) and machine learning (ML) have sparked a craze in the field of weather forecasting, with models such as Google DeepMind's GraphCast and Huawei's Pangu-Weather repeatedly breaking the records of the traditional European Centre for Medium-Range Weather Forecasts (ECMWF) in both speed and accuracy, prompting the outside world to exclaim that the "AI revolution" has arrived. However, this in-depth analysis from Ars Technica points out that this so-called revolution is actually not as disruptive as imagined, and machine learning has limitations in climate and meteorological science that are difficult to overcome.
Free shows the 3-line summary; Pro unlocks the full deep summary (~300 words) so you never have to click through.
See Pro plans →Want the original English / full article?
Read on Ars Technica AI →Related
Summaries are AI-generated; the original article is authoritative.