Comparative aesthetic assessment of machine learning and human judgment for building wall designs

Abstract

Machine learning models can potentially provide alternative options in the field of architecture as aesthetic judgment tools, owing to their high capacity and data-driven environments. If a machine learning model can produce aesthetic evaluation results similar to those of humans, the process may be highly promising for further applications in architectural decision-making. In this study, we propose a series of interconnected workflows for a rigorous comparison, including data collection, machine learning, parametric designs, robotic fabrication, and human surveys, to test the compatibility between human judgment and machine learning models in the aesthetic assessment of architectural objects on the same design objects. We observed a wide gap between the aesthetic judgments of the two groups. We discuss certain drawbacks and current limitations to improve the vulnerability of the study process and conclude by providing an outlook for the subsequent direction of a similar study.