Gautam Mukunda warns that while artificial intelligence (AI) is revolutionizing weather forecasting by outperforming traditional physics-based models—especially demonstrated during the 2025 Atlantic hurricane season when DeepMind’s model excelled—there are significant limitations. These advanced AI models, now preferred by meteorological services due to their efficiency and cost-effectiveness, may struggle in unpredictable “gray swan” scenarios where they rely heavily on past data. This shift towards AI forecasting, particularly beneficial for developing countries lacking expensive infrastructure, raises concerns about the resilience of weather prediction in a rapidly changing climate.

Gautam Mukunda: Gautam Mukunda is a leadership professor at Yale School of Management who researches innovation, ethical leadership, and technology’s societal impacts. He authors books and opinion pieces on management challenges. Here, he argues that AI’s reliance on past data poses risks for unprecedented future weather events, urging caution in abandoning physics-based models.
Google DeepMind: Google DeepMind is an AI research lab under Alphabet that develops machine learning models for scientific challenges, particularly in weather forecasting with systems like WeatherNext 2 and GraphCast. These models generate probabilistic predictions by learning from historical data patterns more efficiently than traditional methods. In the news, DeepMind’s model outperformed physics-based forecasts during the 2025 Atlantic hurricane season, highlighting the shift to cheaper AI alternatives.
European Centre for Medium-Range Weather Forecasts: The European Centre for Medium-Range Weather Forecasts is an intergovernmental body producing global numerical weather predictions and climate data using both physics-based and AI-enhanced systems like AIFS. It fosters AI innovation through competitions such as the AI Weather Quest. The article cites its machine learning models as matching or exceeding top physics-based systems, aiding the retirement of resource-intensive forecasting infrastructure.

`json
{
“AI Advancements”: “AI models developed by DeepMind and ECMWF now match or outperform traditional physics-based models in medium-range weather forecasting.”,
“Model Limitations”: “AI models may face challenges in predicting extreme weather events that fall outside the training data, where traditional physics-based models have historically been relied upon.”,
“Hurricane Performance”: “During the 2025 Atlantic hurricane season, DeepMind’s AI model outperformed many traditional physical models, according to National Hurricane Center evaluations.”
}
`