CAAD Forum | Compositional Intelligence: Architecture for Higher Dimensional Environments

Compositional Intelligence: Architecture for Higher Dimensional Environments
Neural networks exploit a counterintuitive geometric property: in high dimensions, nearly-orthogonal directions are abundant, enabling superposition—more features than dimensions, overlapping without interference. Based on the book Compositional Intelligence (Routledge, 2025), this lecture asks what such geometry means for architecture. Generative AI encodes vernacular knowledge statistically, correlations across millions of images where climate, materials, and spatial decisions co-occur. Without design, this intelligence defaults to hyper-materialization: spaces optimized for single variables, hostile to the complexity of life. The lecture proposes synthetic ground as alternative—spatial configurations preserving compositional richness by superimposing multiple logics in geometries requiring their co-presence. Through reciprocal navigation between computational inference, human judgment, and material response, architecture becomes the projection of high-dimensional abundance into inhabitable three-dimensional form.

The CAAD Forum series is hosted by the Center for American Architecture and Design to bring faculty, students, and staff together for informal and inquisitive discussions about ideas relating to architecture and its history, theory, practice, and future. Presentations introduce and off insight into new and ongoing research, and are followed by time for Q&A. Lunch provided; first-come first-served. View past CAAD Forums on the Texas Architecture YouTube channel.