When the Environment Becomes the Body: Architectural Body Intelligence and Developmental World-Model Formation

Generative artificial intelligence can plausibly complete objects, relations, and internal structures that are not directly observed, drawing on regularities in text and images. Both text and video, however, are primarily descriptions or recordings of a world experienced by humans. Physical properties such as mass, friction, resistance, support, and collision can be estimated statistically from such data, but they are never given as sensory consequences of the artificial agent's own actions. This paper proposes Architectural Body Intelligence (ABI) as a third pathway of world-model formation that addresses this gap. In ABI, the built environment itself is organized as the primary body and developmental environment of an artificial agent. Floor pressure distributions, structural vibrations, acoustics, effector states, and wide-area visual input from distributed cameras are integrated into a single learning loop, and indoor robots and manipulators serve as effectors. Standard ABI does not occlude vision. Visual and structural sensing are learned jointly as consequences of active intervention, and what ABI extracts from human development is not an ordering that delays vision but the principle that multisensory correspondences between perception and action precede linguistic symbolization. The paper positions sensorimotor learning as a third foundation alongside linguistic and visual learning, presents the architectural body as a developmental niche in which machines learn the world bodily, and translates simultaneous multisensory, physics-first, vision-first, and language-first configurations into a comparable experimental program. ABI is offered as a constructive conceptual model for examining tacit assumptions of artificial embodiment rather than as a report of an implemented system.