This project was a submission for the eVolo Skyscraper competition where we collected a large dataset of texts about and images of architecture with which we trained different machine learning algorithms to inform a submission for the competition.
we collected all the images (posters) of the winning projects and honorable mentions submitted to the competition prior to the year of our submission (2006-2021) and trained Style-GAN ADA2 on these images
we collected the texts (abstracts) of all the winning projects and honorable mentions from the competition prior to the year of our submission (2006-2021). we also collected the texts of all the articles published in the Architectural Design magazine (2005-2021). We used this dataset of texts about architecture to train > TensorFlow to generate new texts, and VQGAN+clip to generate new images
We used this data to inform a design for a submission to the competition. We reflect on the usefulness of AI generators trained on architectural data for conceptual design in architecture.
Publications based on the project:
AI for Conceptual Architecture: reflections on designing with text-to-text, text-to-image, and image-to-image generators. 2024. [AS. Horvath, P. Pouliou]. Frontiers of Architectural Research. https://doi.org/10.1016/j.foar.2024.02.006
Results from training Style-GAN ADA2 on images of all previous submissions to the eVolo skyscraper competition prior to our submission. (images of all winning projects and honorable mentions between 2006 and 2021)
Results from training Style-GAN ADA2 on images of all previous submissions to the eVolo skyscraper competition prior to our submission. (images of all winning projects and honorable mentions between 2006 and 2021)
This project was funded by the Human-Centered AI cluster at the Department of Communication and Psychology, Faculty of Social Sciences and Humanities, Aalborg University.
Project description:
Using machine learning to generate works of art, design and architecture has been an emerging research field in the last decades, and experiments have been made as early as the 1960, inspired by Alan Turing's question ‘can computers think?’.
In architecture, machine learning allows the exploration of large design spaces and optimization of certain aspects which can be expressed in a numerical format, such as areas, volumes, material use or energy consumption. In the last years, much effort has gone in automating floor plan generation and in generative design tools which optimize material use in concert with structural requirements. Additionally, initial work has shown results from training a generative adversarial neural network to ‘hallucinate’, creating images about architecture (del Campo et al., 2021).
In continuation of previous work, Reconsidering otherness is a design-based research project aiming to create conceptual architecture using machine learning tools for generating texts, images and annotated point clouds. This might help to uncover implicit biases in training models and will contribute to current discussions on (creative) authorship in a (post)-digital age (Carpo, 2017). The work builds on Kyle Steinfeld’s (Steinfeld, 2021) categorization of machine learning in art and architecture as (a) actor, (b) as material and (c) as provocateur.
---
Carpo, M. (2017). The Second Digital Turn Design Beyond Intelligence. MIT Press.
del Campo, M., Carlson, A., & Manninger, S. (2021). Towards Hallucinating Machines - Designing with Computational Vision. International Journal of Architectural Computing, 19(1). https://doi.org/10.1177/1478077120963366
Steinfeld, K. (2021). Significant others: Machine learning as actor, material and provocateur in art and design. In I. As & P. Basu (Eds.), The Routledge Companion to Artificial Intelligence in Architecture. Routledge. https://doi.org/10.4324/9780367824259
In continuation of previous work, Reconsidering otherness is a design-based research project aiming to create conceptual architecture using machine learning tools for generating texts, images and annotated point clouds. This might help to uncover implicit biases in training models and will contribute to current discussions on (creative) authorship in a (post)-digital age (Carpo, 2017). The work builds on Kyle Steinfeld’s (Steinfeld, 2021) categorization of machine learning in art and architecture as (a) actor, (b) as material and (c) as provocateur.
---
Carpo, M. (2017). The Second Digital Turn Design Beyond Intelligence. MIT Press.
del Campo, M., Carlson, A., & Manninger, S. (2021). Towards Hallucinating Machines - Designing with Computational Vision. International Journal of Architectural Computing, 19(1). https://doi.org/10.1177/1478077120963366
Steinfeld, K. (2021). Significant others: Machine learning as actor, material and provocateur in art and design. In I. As & P. Basu (Eds.), The Routledge Companion to Artificial Intelligence in Architecture. Routledge. https://doi.org/10.4324/9780367824259