The concept of Interpretable box models
As part of the overall interactive experience, I wanted to also represent the concept of the “Black box” in the context of Machine Learning. As a result of my recent readings on Artificial Intelligence and the ethical considerations behind it, I stumbled upon the concept of the Black Box models. These models are created from data, with minimum input from humans and which often produce results that are complex to explain and are often opaque. (Rudin & Radin, 2019)
However, I found it more relevant to focus on the counterpart of these obscure models, that is, Interpretable models. As a more ethical model, it offers a better understanding of how predictions are made. The way I chose to represent this, is by literally building a box from which all the training of my model is happening. All the elements that go inside it (the camera and the cards with patterns) are visible to everyone.
This documentation shows the different considerations I went through when building the box, and adapting it to my concept and the other elements of the interaction.
First Iteration
In this phase I experimented with two different box designs. The size of the box would depend on the size of the camera that would be placed inside. This first testing was done in cardboard, but the material I want to use to communicate (quite literally) the concept of transparency is clear acrylic. The nature of the material won’t allow easy flexibility, and therefore the Design One presented a better choice.
Second Iteration
I also decided to add a pattern to the box itself. This design choice was initially aimed to help light come through the box, and therefore support the detection of the patterns by the camera. Later, it also became an analogy about the artisan patterns I was researching.
Third Iteration (Acrylic ?)
References
Rudin, C. and Radin, J. (2019) Why are we using black box models in AI when we don’t need to? A lesson from an explainable AI competition, Harvard Data Science Review. Available at: https://hdsr.mitpress.mit.edu/pub/f9kuryi8/release/8 (Accessed: 01 January 2024).