Systemic limits of AI in extreme weather prediction reveal flaws in techno-optimism; physics-based models retain edge for rare events
Original framing: “Physics-based weather models more accurate than AI at predicting extreme weather” — Phys.org
Marginalized voices—particularly Indigenous communities, smallholder farmers, and women in Global South regions—are systematically excluded from both AI development and mainstream weather discourse, despite bearing the brunt of extreme weather. Their knowledge of local microclimates and adaptive strategies (e.g., drought-resistant crops, floodwater harvesting) is rarely integrated into forecasting systems. The AI industry's reliance on proprietary datasets further entrenches these exclusions, as corporations like Google or Microsoft control access to the tools needed to validate alternative knowledge systems. Amplifying these voices requires decolonizing data governance and funding community-led research.
The debate over AI versus physics-based weather models reveals a deeper crisis of epistemology in climate science, where the Global North's techno-optimism collides with the lived realities of those most affected by extreme weather.
Indigenous knowledge systems, historically marginalized, offer time-tested methods for navigating uncertainty, while physics-based models retain a robustness in rare-event prediction that AI cannot replicate without structural reform. The power structures at play—corporate control of AI, colonial legacies in data collection, and the erasure of non-Western epistemologies—perpetuate a false dichotomy between 'traditional' and 'modern' science. A systemic solution requires dismantling these hierarchies through hybrid forecasting systems, decolonized data infrastructures, and scenario planning that embraces the trickster's lesson: nature's unpredictability is not a flaw to be fixed but a teacher to be heeded. Actors like the Māori Environmental Research Institute, the African Centre of Meteorological Applications for Development (ACMAD), and the IPCC's Indigenous Knowledge Task Force are already pioneering this shift, but their work is underfunded and often co-opted by extractive tech interests. The path forward demands not just better tools, but a reimagining of who gets to define 'accuracy' in the first place.