Towards An Other Architecture
Benjamin Wade James. Vienna, Austria
-
Name of work in English
Towards An Other Architecture
-
Name of work in original language
Machine Learning x Architecture
Prize year
Young Talent 2020
-
Work Location
Vienna, Austria
-
Author/s
Benjamin Wade James
-
School
Institute of Architecture - University of Applied Arts Vienna.
Vienna, Austria
Young Talent 2020 YT Nominees
Towards An Other Architecture
Machine Learning x Architecture
Program
Collective housing
-
Labels
Tower · Complex
This project creates a new architectural tool using artificial intelligence as a means to augment the intuition of the designer. The tool was used to generate 100 unique designs in the footprints of iconic architectural structures.
Artificial intelligence offers profound opportunities for innovation in design.\nThe ability of a machine to learn, think and act will fundamentally change the way in which we design. We can teach a machine by example: giving it sets of data and training it search for latent patterns embedded in 2d drawings and imagery. This is a reconfiguration of the relationship between human and machine - the machine is no longer a passive tool that does exactly as it is told, and the design process is no longer fully deterministic. This change mirrors an equally fundamental shift in the relationship between parts and wholes. \nDrawings are seen by neural networks with chameleonic qualities, being composed of parts or wholes simultaneously as local relationships and as global assemblages. The ability of robust machine intelligence systems to process this injects a new kind of parts / wholes relationship mediated by transformations of perception, threshold and resolution.\nCan a machine learn to design architecture?\nTo teach something that knows nothing of architecture, I started with the plan; and for this it was taught from Le Corbusier. Le Corbusier’s extensive body of work provided a dataset of significant size and his focus on the plan as ‘the generator’ of architecture allowed for clear encoding of information.\nThe machine used a Generative Adversarial Neural Network to learn the relationships between vertical circulation, interior / exterior, double height spaces and structure in over 250 plans of Le Corbusier.\nWhen the machine was able to redraw Le Corbusier buildings with a reasonable degree of accuracy, it attempted to create 100 new designs each one in the footprint of an iconic architectural structure. The result is a series of sublime Frankenstein monsters: part Le Corbusier, part machine, and part reference building.\nThe resulting designs show similarities and differences in the underlying patterns of these conflicting influences, and the hybrid structures exhibit new ways of thinking about typology, form and space.