Commentary by Heidi Biggs
Is Sooji Kim’s thesis automating the color theory of Joseph Albers?(1) In a way, yes. Sooji’s final project explores machine learning applications within a fundamental element of design: color. Over the course of the past year, she built a machine learning (ML) cluster model that can detect and build color palettes from images and developed it into a working prototype of an app. Starting from the idea that an ML algorithm could help someone discover and cultivate their own personal style of interior design, Sooji pivoted to developing a project where ML helps a user find their own personal color preferences. She dug into ML, conducting hands-on learning and making through researching clustering models, following GitHub tutorials, and learning to use Google Collab to build a python notebook. Eventually, Sooji generated her own color-palette detecting algorithm. In tandem with building an ML model for color palettes, she also built a speculative application for the model to demonstrate how it could be used. Through a long-term process of narrowing and ideation, Sooji finally implemented her palette builder within a prototype for an app. The app helps users build color palettes from a self-selected photograph or image and then uses their individualized color palette to help them select and purchase color-specific items for the interior of their home. She reflected that her final prototype (which she developed into a high fidelity tool that is integrated with her palette building model) was a way to materialize and demonstrate her color model in action. In her words, it translates ‘code language’ to ‘design language’.
Designers are currently working hard to understand how to design for and with ML and how to grapple with its implications. It was impressive and ambitious for Sooji to develop a cluster model from scratch, and I wonder how much interaction design is blending with coding (in light of the many projects from this year’s MDes cohort which were coded from the ground up). Sooji reflected on how learning to use ML was empowering and helped her grasp the implications of designing with ML. One way she sees ML will impact design is by speeding up and automating away busywork from the design process. Thinking about what ML ‘designs away’, one could view her color palette builder as removing the necessity of expert knowledge from the process of selecting colors for the design of an interior, bypassing the more laborious task of generating a color palette oneself and shifting the task to finding the palette already composed in the world. It takes the work of color craft from the space of analytic to intuitive, from constructive to found and curatorial. Along the lines of design implications of ML, Sooji and I mused together about the transparency and opacity of the use of ML models in everyday places like her final prototype. The UI wasn’t explicit about the extent to which it used ML, which stands in contrast to the immense time, energy, and labor Sooji put into understanding and developing the algorithm it is based on. This is an interesting tension! As ML inundates interaction in seemingly harmless ways, when and how, and why should we alert users to its integration?
(1) Albers, Josef. Interaction of Color. Yale University Press, 2013.