Classification of artificial intelligence techniques for early architectural design stages

Abstract

This paper provides a strategic classification of artificial intelligence (AI) techniques based on a systematic literature review and four levels of potential: the levels of input, output, collaboration and creativity. The classification demonstrates the potential and challenges of the AI techniques when used in early stages of architectural design. We aspire to help architects, researchers and developers to choose which AI techniques might be worth pursuing for specific tasks, optimising the use of today’s computational power in architectural design workflows. The results of the classification strongly indicate that Evolutionary Computing, Transformer Models and Graph Machine Learning hold the greatest potential for impact in early architectural design, and thus merit the attention to achieve that potential. Moreover, the classification assists with building multi-technique applications and helps to identify the most suitable AI technique for different circumstances such as the architect’s programming skills, the availability of training data or the nature of the design problem.