Co-Designing with Artificial Intelligence. Exploring the use of machine learning in the AEC industry

Construction and design workflows are constantly optimised for efficiency to coincide with the technological revolution to reduce time and cost Venkrbec, V., et al., (2018). There is little real-world adoption of artificial intelligence in architectural design processes due to economic constraints requiring known values influenced by a result-driven model. Since the adoption of Computer Aided Architectural Design (CAAD) followed by Building Information Modelling (BIM), there have been untapped project dataset resources to help optimise the design and construction process Pena et al (2021).

Public and private historical design data, when merged with Artificial Intelligence (AI) tools, i.e. Machine Learning (ML) and Deep Learning (DL), provides valueadded data assessment. The results of ML algorithms can be an informer tool for use in architectural workflows. Using ML within an algorithmic design (AD) can silhouette how the design sector thinks about its application of data already owned and what could theoretically be generated using data from the design process. Building on existing research Stojanovski, T (2021) furthers the use of the industry’s design and construction datasets. Combining the available data with design workflow optimises workflows already in use.

In testing, the research investigates experimental algorithms and generative software modelling to analyse and optimise design workflows showing how AI can influence the human interface with design. The research results were analysed for effectiveness and assessed against current workflows in the absence of this technology to understand the influence researched changes could have on design and construction. The research emphasises the interconnection between current and future applications of data within design and architecture. Implications for design and construction amidst the technological revolution and the need for interdisciplinary conversation are addressed in the findings.