Arch. Malgorzata Starzynska, School of Architecture, Royal College of Art (UK)
Abstract: Architectural application of machine learning (ML) has primarily been explored in building delivery and construction manufacturing; however, the application to architectural criticism, specifically pedagogical methodologies, is yet to be proposed. This paper argues that, by understanding the cognitive mechanisms of algorithmic analysis, it is possible to identify opportunities for new methodologies of archival studies of architectural heritage. This paper demonstrates that the ML framework offers an opportunity for unlearning in architectural theory and inform a critical framework of architectural practice.
ML mechanisms replicate and amplify data distribution tendencies, making visible some of the implicit truths otherwise overlooked by social conditioning or the current theoretical paradigm. Building on the works of data philosophers such as Matteo Pasquinelli, Luciana Paris and Katherine Hayles, this paper demonstrates that, unlike traditional methods of enquiry, an algorithm’s cognitive ability introduces an independent critical agency that tends to reveal truths beyond the scope of the research question.
The proposed paper demonstrates how the practical application of DeepDream, a computational method for visualising non-optical
cognition of an image classification algorithm, results in new methodologies for heritage studies. Based on the practical application
of DeepDream to architectural styles analysis, the system was able to highlight tendencies that extended beyond the original hypothesis of this paper. Investigating visual representation of gothic, renaissance, baroque and modernism in the user-generated repositories, this paper demonstrates a strong relationship between the styles’ ornament and unconsciously adopted photography techniques. This paper aims to demonstrate that the opportunities offered by the application of ML to cultural
studies can challenge existing datasets supporting methodologies of traditional archival studies. Building on the methods of precedent analysis, the algorithmic agent is positioned as an opportunity to challenge and further develop our understanding of past and present architectural designs.
Short CV: Malgorzata is an architect, an educator and a current PhD candidate and research associate at the Royal College of Art in London. She is also a current Machine Learning Teaching Lead at an Architect Apprenticeship course. In her research, Malgorzata explores the transdisciplinary advancement in AI and applications of machine learning to image-based and 3D space recognition. Her research work has been awarded, published and presented at several international conferences and lectures.