counterfeit_daily_header_v2.jpg

Artificial Intelligence has changed the way images are being created. The Generative Adversarial Neural Network(GAN) is an algorithm that breaks down images in to their pixel values and then mathematically learns the relationships between the pixels. By processing large amounts of images and updating its understanding of underlying relationships withing the dataset of images the GAN then attempts to create counterfeit images based on the original images that it have been provided.

 

For architects and designers images play a significant role in our creative process. Websites such as archdaily.com have become data banks of precedent that we continually turn to. These precedents are applied in differing ways but often the process of precedent has become like shopping in a supermarket. The speed at which the design world functions has fully embraced technology and where it is taking us and in return these technologies our affecting the design process of all designers.

The ability of the designer to process images pales in comparison, at least in numbers, to the ability of algorithms. This project, Counterfeiting Daily, attempts to look at what the role of images could be in the design process if they were embraced as the main source of production. By downloading the image media from the Archdaily page "Best Houses of 2018" and running the images through Taehoon Kim's implementation of DCGAN an attempt was made to embark on a process strictly guided by images and the results produced by the algorithm, with hopes of creating a counterfeit "Best House". 

Through creating datasets based on site plans, exterior perspectives and interior perspectives and allowing the algorithm to produce outcomes, the results were analyzed to gain an understanding of image dominated processes, understand the technology of the neural network, and understand the role of the architect in creating with algorithms. By gaining more knowledge in these areas as a designer it is then possible to engage in thoughtful critique, instead of blindly accepting the results that algorithms provide back to us.

The results from the algorithm change the way that we as designers must look at images. The images are a product of what the algorithm predicts the next pixels to be and therefore the process of making is not subject to the structural, spacial, social, economical, and other constraints that a designer brings to the design process. In doing so the algorithm outputs reflect findings within the dataset but ones that take time for the designer to discover. This changes the relationship to images by forcing the designer to engage in a thoughtful discourse with the images in an attempt to understand the language that has produced them.

Site Plans

Selection of DCGAN outputs of Site Plans

1.jpg
4.jpg
7.jpg
2.jpg
5.jpg
8.jpg
3.jpg
6.jpg
9.jpg

Selection from site plan dataset

Site_plan_dataset.jpg

Exterior Perspectives

Selection of DCGAN outputs of Exterior Perspectives

6.jpg
1.jpg
5.jpg
2.jpg
3.jpg
4.jpg
iteration_view.jpg

Selection from exterior perspective dataset

Exterior_perspective_dataset.jpg

Interior Perspectives

Selection of DCGAN outputs of Interior Perspectives

9.jpg
3.jpg
5.jpg
2.jpg
4.jpg
1.jpg
iteration_view_int.jpg

Selection from Interior Perspectives Dataset

Interior_perspective_dataset.jpg