top of page
a tool to decipher and augment designerly intuition
This project was part of my scholarship for AI at the Fraunhofer Institute in Stuttgart. The key goal was thinking conceptually with and about AI, and acquire new AI prototyping skills.
Dr. Truong Le
100 AI Talents - scholarship
at Fraunhofer IAO Stuttgart
Design knowledge is often tacit. When sketching ideas or making decisions, we rather intuitively feel then rationally elaborate. Intuition is thereby not as subjective as we often think*. Try it yourself an intuitively assign the words 'Takete' and 'Maluma' to the shapes on the left. You can find the common answer here.
That means intuition and emotions have underlying patterns, we as humans are just not always able to recognize. Designing for certain emotions, associations or semantics makes it necessary to uncover these principles behind the our tacit designerly knowledge.
In order to augment our abilities to make decisions, humans long included machines and computers in the process of making a choice. Seeing value in the pattern recognition ability of AI, I created a concept that can help designers to better analyze their user‘s abstract emotions or experiences. My main intention was to create a collaborative tool, that combines the designer‘s ability to creatively search for solutions and AI‘s possibilities in analyzing data.
The concept uses a generative adversarial network (GAN). In the illustrated example, chairs with certain aesthetic properties “light” and “elegant” are aimed to be designed.
The designer is curating the data and thus baking in his tacit knowledge and intuitive perception of existing chairs by classifying them in “heavy”-appearing, or “light”-appearing.
When the discriminator network finally comes to the decision, that the generated image can no longer be distinguished from real images, an unlimited amount of 'strong' chairs can be generated. Combining the underlying visual patterns of the training data, the designer can use the artificial stimuli to expand his understanding of his intuition and/or use the image as a starting point for the design phase.
Working with AI is a learning process. For the algorithm, for the user and for me as a designer. The first tests were run with a self coded GAN. Due to the large computational effort of training a GAN from scratch, I soon realized a cloud based, more powerful model will be necessary. Final results were then achieved with RunwayML.
This project is of exploratory nature. It remains questionable whether the computational costs are worth the design benefit.
the algorithm studies
the training data.
In this case simplified
shapes as the
first step towards
intuitively labeled chairs
left 'light', right 'strong'
bottom of page