AI has entered a stage where scale and capability alone are no longer sufficient. Despite remarkable advances in perception, language, and generation, modern AI systems continue to struggle with a fundamental challenge: understanding, representing, and acting within the structured physical world. As intelligent systems increasingly move beyond static inputs and isolated tasks, the question shifts from how well a model predicts tokens or labels to how it forms internal models of space, time, action, and causality. This shift marks the emergence of spatial intelligence as a central organizing paradigm in artificial intelligence.
Spatial intelligence concerns the ability of an intelligent system to build and manipulate internal representations of the world that support perception, reasoning, prediction, and embodied interaction. It is inherently world-centered, grounded in geometry, motion, and physical constraints, and inseparable from action and experience. Rather than treating space as an auxiliary signal or downstream output, spatial intelligence positions it as a core substrate through which intelligence is structured. This perspective cuts across traditional boundaries, unifying problems in computer vision, robotics, autonomous systems, simulation, cognitive science, spatial computing, AR/VR, and geospatial analysis under a shared representational question: how should an intelligent system model the world it inhabits?
This chapter establishes spatial intelligence as a rising AI paradigm by addressing two foundational questions. Section 1.1 examines why spatial intelligence has become critical now, tracing its emergence to converging shifts in AI methodology, data modalities, system scale, and problem scope. Section 1.2 then situates this paradigm historically, organizing decades of fragmented progress into a coherent sequence of representational eras. Together, these perspectives provide the conceptual grounding for the rest of the handbook, framing spatial intelligence not as a collection of tasks or techniques, but as a unifying foundation for world-centered artificial intelligence.