This research integrates recent AI technologies of machine learning (ML) within architectural theory and practice. It asks how architects can apply ML as an alternative method of heritage analysis and how this might inform design. It proposes to assemble a novel 3D digital dataset of historical styles from which an intelligent system can learn, analyse and generate its own propositions. Applying the new cognitive agent, the project aims to identify previously overlooked cultural and social biases implicit in architectural forms. This process is used to challenge the traditional reading of precedents and as a method for unlearning in architectural history.

A growing number of industries deploy machine learning to tackle Big Data. Machine vision became a contemporary visual condition where vision is no longer optical. This research sits in the context of historical analysis of the ocular tradition of representation to confront the contemporary developments of a new cognitive system: machine vision. From fine art to architectural projects, this research investigates the new works in the context of a long tradition of the Eurocentric gaze. ML is positioned as an emerging semiotic system and as such it asks to reconsider its role in architectural design. The aim of this practice-led project is to train an artificial learning system to generate a space/ spatial design through a ML algorithm, and in doing so analyse and reflect on the implications for architectural design. The premise of this practice-led research is to deploy ML as a process of unlearning in architectural history and practice in order to gain an alternative understanding of spatial properties of a building through new AI readings of space. Based on this analysis a novel space will then be generated; one that encompasses properties of the buildings grouped into a new category through so-called clustering of patterns. The practice challenges traditional categorisation of space making visible their cultural biases and interrogates spatial complexity as perceived by the eye of an algorithm. Architectural history has relied upon the categorisation of style and use among other attributes informing future proposals through methods of precedent.

The project aims to provide a measurable contribution to the field of architecture and spatial design within a context where contemporary perception now extends beyond the human experience. This research aims to contribute to the contemporary debate on human-machine collaboration in the creative process which recently has sparked questions about authorship, creativity, and labour. Understanding the relationship between a learning algorithm, a data-set and the outputs are key for the future of the construction industry and spatial design.