Collection of fakes: Identifying and generating character in Ringön by exploring the use of generative adversarial networks

Abstract

In recent times, artificial intelligence has become a valuable and creative tool in the design processes to amplify our cognitive abilities and to design things that are out of our reach. However, it is essential to acknowledge that artificial intelligence is not inherently a positivistic tool, it does not come without negative consequences. Bias is a challenge with AI currently being discussed and debated, and machine learning models are only as good as the data they are trained on. If the data that is fed into machine learning models are biased, that, in turn, shapes the machine learning models and data we consume to be biased. This thesis explores how generative adversarial networks (GANs) can support and influence a design process and lead to an architectural proposal. This project uses these tools to explore how one can identify and generate character in relation to a site. The chosen site for these explorations has been Ringön, an island in Gothenburg, with the notable feature that it is a central industrial district that has managed to keep its character. Using GANs, images of specific characteristics from Ringön are generated based on limited data sets collected in the form of photographs from the site. Data sets are collected by identifying and photographing idiosyncratic and a bit quirky elements that give the buildings’ character on Ringön. The data sets are fed into a machine learning model that generates images that resemble the collected data set. Consequently, this project is a subjective perception of the character in Ringön and an exercise in translating images into a design proposal. What is biased data, and why is it important to be critical when generating images? This thesis highlights the consequences of bias when collecting data sets and subjective selections through experimental methods, with the aim to explore new approaches to design to find a unique expression of a building. Through research through design method, this thesis can be viewed as a contribution to exploring how contemporary digital tools can be part of a design process and what consequences and results this can have.