Combining satellite images and artificial intelligence to measure poverty in 1982-2020, and use these data toexplaining the effects of World Bank and Chinese development programs in Africa
Short description
About 300 million people in Africa live in extreme poverty. Operating on the assumption that life in impoverishedcommunities is fundamentally so different that it can trap people in cycles of deprivation (‘poverty traps’), majordevelopment actors such as China and the World Bank have deployed a stream of projects to break these cycles(‘poverty targeting’). However, as scholars are held back by a data challenge, they are currently unable toanswer questions such as in what capacity do poverty traps exist, and thus, to evaluate what extent theseinterventions release communities from such traps.
Our aim in this project is to identify to what extent African communities are trapped in poverty and explain how competing development programs alter these communities’ prospects to free themselves from deprivation. To address this aim, we will (i) train image recognition algorithms—a form of AI—to identify poverty from satellite images between 1984 to 2020; (ii) use these data to analyze how development actors affect African communities; (iii) using mixed methods to develop theories of the varieties of poverty traps; (iv), develop an R package, PovertyMachine, that will produce poverty estimates from new satellite images—ensuring that our innovations will benefit poverty research. The research tasks are of such a challenging character that a single project or research team cannot address it. Thus, seven social- and computer scientists have joined forces to tackle this project’s aim
Researchers
Adel Daoud, University of Gothenburg, project leader
Maria Brandén, Linköpings university
Devdatt Dubhashi, Chalmers University of Technology
Peter Hedström, Linköpings university
Fredrik Johansson, Chalmers University of Technology
Ellen Lust, University of Gothenburg
Xiao-Li Meng, Harvard University