Description
This paper examines the development of AI-based migration prediction tools within the European Union (EU). While these sophisticated, albeit experimental tools are attracting significant funding from state, international, and humanitarian actors, assessing their future role in migration management is difficult: On the one hand, the prospect of harnessing increasing volumes of online, Google Search, and social media data promises increasingly accurate predictions of population movements and raises the prospect of prediction-based pushbacks. On the other, present shortcomings of these models, including biased training datasets and flawed underlying theoretical assumptions, cast doubt on the accuracy even of future models. Against this background, the presentation considers these tools as means to take epistemic control of an essentially unknowable phenomenon: what makes migration complex, namely the unpredictable dynamics of violent displacement and the highly subjective decision to flee, must remain out of reach of statistical models. Drawing on document analysis and interviews with model developers, the paper considers prediction models developed by the EU’s Asylum Agency (EUAA) and ITFLOWS, an EU-funded research consortium. Grounded in a decolonial and feminist data justice approach, the paper interrogates how these technology-mediated bordering practices serve to advance and entrench a future-oriented racialized EUropean border regime.
