Deep Learning Chair


We aimed to further develop the “G Chair” experimentation using the following method. For a start, we made 3D models based on the nine most spoken languages (three types of chairs per language) totaling 27 pieces, changed their proportions parametrically and made a set of image data. The next step was to recreate these models through the process of deep learning using a machine language. Technically, image data were converted to voxel data and then recreated into optimum voxel data. Contrary to “Global G Chair” implemented by humans, this experiment was implemented through machine-learning.

I expected that this experiment would help us see our past design processes from an objective viewpoint. From what I observed, humans probably do better in the process of analyzing and extracting typologies. On the other hand, it seems that machines can bring more precise results in the process of recreating typical forms through calculations based on vast amounts of data.

When humans do the job, the results may be influenced by the level of empathy one has toward a person who brought the results. Sometimes empathy helped the process, but also hindered it otherwise. Although “Global G Chair” brought unique results, some people may find it ironic depending on how we present it.

The results generated through the machine-driven process appear more convincing. Machine-generated design based on data and calculation may successfully facilitate communication based on something other than empathy. I also expect that machine-generated design may conceive a new kind of humor –– perhaps the same kind of humor that made us laugh when we spotted unexpected correspondence between the shape of the roof and the undulating mountain.

Project Date: 2018.09.30

Photo: Nacása and Partners