Whose future is coded? A decolonial-feminist interrogation of the right to collective dreaming in AI governance

Abstract

Contemporary AI governance frames gender justice as a technical problem of bias, measured, audited, and mitigated at the level of the model. This paper argues that the framing misses a deeper harm: the algorithmic foreclosure of the futures communities can imagine for themselves. I pose a single diagnostic question: Whose future does this system make unimaginable? To answer it, the paper develops Dreaming Otherwise, a dialectical framework moving through epistemic disobedience (refusal of imposed legibility), the epistemic commons (collective knowing-together), and collective dreaming (authoring the future rather than receiving it). The foreclosure is not only metaphor: the paper identifies six mechanisms of machine learning through which learned systems close futures differently from written bureaucracies, from distributional support closure and embedding geometry to model collapse. The framework is translated into a three-layered rights architecture: epistemic vulnerability as its ground, Hohfeldian jural relations as its structure, and relational autonomy as its mode of exercise. The architecture is then operationalized inside Article 27(1)(c)-(d) of Regulation (EU) 2024/1689, the fundamental rights impact assessment whose application to stand-alone high-risk systems the Digital Omnibus has deferred to 2 December 2027, where affected groups are already named and group-level risks already required to be identified. I argue that imaginative foreclosure can be read into this existing mechanism as a specific risk of harm, without any new regulatory provision. The paper contributes a decolonial-feminist diagnosis, a dialectical praxis of refusal, commons, and dream, and a rights architecture claimable within existing law and, in principle, beyond it.