Why Students Don't Use Urban AI: A Study of Machine Learning Tool Adoption in the Design Studio

Machine learning (ML) tools are increasingly promoted for urban design, but little is known about their uptake in architecture studios. This paper examines conditions that enable or constrain students' meaningful use of a web-based ML urban analytics tool in a third-year urban design studio. The tool predicted pedestrian volumes as a proxy for walkability and displayed results on a map. Data included a usage survey and 12 semi-structured interviews. The findings suggest that although students encountered the tool and occasionally tried it, they did not integrate it into their iterative design practice. This reflected a sociotechnical misfit involving model validity, explainability, interface design, pedagogical support, and studio time pressures. Four barriers shaped non-adoption: mistrust of implausible, opaque outputs; difficulty translating predictive maps into design decisions; unmet expectations for dialogic AI; and unfavorable time-benefit tradeoffs shaped by studio culture and assessment structures themselves.