Designing from Some-Where: Generative AI and Foreclosed Futures in Design

Abstract

The integration of generative artificial intelligence into design is widely celebrated for accelerating workflows and expanding ideation. This paper argues that this instrumental perspective obscures a critical epistemic shift: generative AI does not just accelerate design processes but structurally truncates already exclusionary practices. By externalizing exploratory, ambiguous, and material form-seeking into probabilistic operations, AI undermines design as a situated knowledge culture. Drawing on theories of epistemic infrastructure, hauntology, and decolonial critique, we analyze how AI systems function as engines of convergence, projecting a dominant view from a specific some-where while presenting it as a neutral standard for everywhere and everyone. Through a double foreclosure, hauntological and probabilistic, these systems favor forms of knowing already present in the dominant digital archive while rendering alternative design futures structurally inaccessible. To counter this homogenization, the paper proposes a conceptual shift. Reimagining AI models not as universalizing infrastructures but as localized carrier bags containing situated epistemologies, we critically examine how such alternatives may reproduce the asymmetries they seek to address. By centering care as a necessary condition for keeping epistemic processes open, designers and educators may protect the processual conditions under which diverse knowledges emerge.