Dreaming Architecture

Collaborating with machines has the potential to broaden our horizons for finding unconventional solutions. Dreaming Architecture is an exploratory research project. AI shows promise as a source of creative inspiration within the early phase of a design process. It also gives us immediate access to endless solutions for any given design task.

The interest in Neural Networks in form of Generative Adversarial Networks is constantly on the rise in the arts and design. Since Neural Networks are inspired by Neuroscience and cognitive behaviour, Machine Learning is often referred to in terms of Vision, Dreaming and Hallucination. What is there to discover if we let machines dream about architectural elements, images and representation? What if one could use Artificial Intelligence as a ‘magic mirror’? Feeding it with everything one can find related to architectural style and letting it dream about it, in order to get access to any thinkable design solution, widening the field of possibilities, combinations or thinking. The project is divided into three parts of a process. Firstly, it introduces Generative Adversarial Networks and explained how they were technically used for further operations with pixel-based data. The second part documents an experimental process, which consists of creating data sets, training machine learning models with them and evaluating the generated data. Due to their visual indeterminacy and the estrangement of familiar objects, the artificially generated images are uncanny, charming and tend to be fascinating. When discussing creatively using Artificial Intelligence in Architecture, one of the first reactions can be fear of losing control over the design process. Including tools like GANs in a design process does not verify this concern. The process documented in the project involved many decisions taken by the author: The machine was chosen, the data sets were selected manually and another exciting moment was examining the generated images and picking some of them for further testing. Finally, the third part of the project is dedicated to exploring and establishing methods of transforming two-dimensional images into three-dimensional forms as an example of further application in design processes. The result is a collaborative design process, which does not compete with human intuition, but augments it. It opens new pathways and interesting, unconventional discussions.