Predicting urban futures has long been central to urban science. This article traces the evolution of urban prediction from structural-equilibrium and process-based models to today’s emerging algorithmic regime. We argue that this shift marks a fundamental epistemic transition from deductive explanation to inductive inference. Examining tensions between explanation, prediction, and policy control, we show that gains in predictive accuracy often come at the cost of causal intelligibility and democratic accountability. The article critiques the governance implications of algorithmic foresight and proposes a framework for democratic urban foresight that reconnects modelling with epistemic democracy, spatial justice, and plural urban futures.