Element of Architecture
The conceptual framework of Rem Koolhaas's Elements of Architecture was used to understand the functionality of Taehoon Kim's implementation of DCGAN. By breaking architecture in to small recognizable elements the algorithms strengths are played to by finding datasets that contain a similar and recognizable focus, while still having variation between the individual images.
While Koolhaas proposes a view of the elements that is a combination of social, historical, and economic factors, DCGAN removes all of that and breaks the elements down to the dominant pixel relationships that cause an image to represent an element. In this digital setting focused on mathematical relationships, the algorithm is free to iterate and provide possibilities that are not beholden to the biases that designers bring to the process.
This is not to say that the algorithm is free from biases. In the creation of the dataset bias plays an important part. For the datasets created for the Doors and Windows datasets, batch downloading from search engines such as google and flickr became the main source of image data. Largely influencing these datasets were travel photography of windows and doors. Because these played a prominent role in the dataset the images the algorithm outputs, feature forms that remind the interpreter of certain historical and locational typologies . In the populist process of accepting every image a generic product of doors and windows is created that moves the outputs away from an architectural intent. This highlights the importance for designers to understand how their interaction with the algorithm can change the output and in this case places the responsibility of the architect to understand the language of the images that they are curating so the intent can be put to use in the design problems that they face.
Doors / Entry
Selection from DCGAN outputs from Doors/Entry
Selection from Doors/Entry Dataset
Selection from DCGAN outputs from Windows
Selection from Windows Dataset