From Thin to Thick

Abstract

This article argues that contemporary research practices in artificial intelligence will produce AI technologies incompatible with human flourishing, social complexity, or vibrant politics. Adopting James Scott's anarchist squint, it instead proposes a vision of AI that embraces local vernaculars, the multiplicity of objectives, the responsibilities that come with producing persons, and the potential of instigating rather than circumventing political contestation. To clarify the stakes, it draws on the key evolutionary idea of niche construction, contrasting the thin, universalizing, top-down construction characteristic of contemporary AI with thick, local, and contested co-construction. It also introduces the idea of semantic power, showing how AI's ability to impose objective functions on the world shapes our very categories of meaning. Finally, it calls for a new social science of possible worlds, outlining strategies for rigorous exploration of the imaginal world where unrealized but possible social arrangements reside.