Paper short abstract:

With climate disasters portrayed as a global concern, and AI as a potent solution, international organizations turn to Big Tech for answers. I trace how such responses evolve, how they make their way into new localities, and how communities develop their own ways of knowing climate disasters.

Paper long abstract:

With climate disasters portrayed as an issue of global concern, and artificial intelligence as a potent solution, international organizations turn to “Big Tech” for answers to the climate crisis. In this project, I inquire into the development, deployment, and appropriation of machine learning models in flood forecasting. First, I trace the evolution of the Early Warnings for All initiative and how it is co-produced (Jasanoff 2004) with histories of hydrology, (sustainable) development, and machine learning. I build on Edwards (2010) to identify how these initiatives embed asymmetrical visions of the global with climate disasters, and how these visions encode which nations that are capable of bringing about progressive transformation, and which ones that are subject to change (Decker and McMahon 2020; Bandopadhyay 2022). Second, I juxtapose these visions with the utility envisioned among developers of machine learning models for flood forecasting. I follow how these models are developed, by whom, and with what knowledge and data. Further, I trace how these models make their way from transnational technology corporations headquartered in the United States into new localities in the Global South through what I call “piloting.” Finally, I attempt to challenge the international epistemic-political power asymmetry and what Decker and McMahon (2020) refer to as the development episteme. I do so by drawing attention to vernacular sites (Greiner 2022) and processes for flood forecasting with non-scalable (Tsing 2012), situated epistemologies (Haraway 1988) and collaborative knowledge production.