You can find a more detailed process of my final project prototyping here.
The final project is a non-interactive mechanical arm that will pick small cubes up and “feed” itself. It will drop the cubes into a larger structure, and the structure will expand with the addition of new cubes. I wanted the piece to showcase a movement indifferent to human presence, so nothing is triggered by human control.
The piece is based largely on thinking about predictive machine learning models, and some conceptual ramifications of predictive model generation. While machine learning can be used to generate eerily accurate pattern predictions, it more interestingly obfuscates any semblance to the users from which the data was pulled, resulting in a hyper-specific set of characteristics about multiple people that will never be made visible, human, or intelligible to those receiving its predictions. I am interested in demonstrating this plurality through the accumulation of cubes, which can resemble different user data.
A lot of my research came from the 3d Additivist Manifesto, and thinking about 3d printing as a way to explain machine learning. I use 3-d printing’s process of concatenating accumulation to help understand machine learning’s paradigm of production. If we think of each chain of plastic in a 3-d printed form as a representation of a unique set of user data, predictive machine learning models give us a sort of cubist portrait of users; a plurality of a few hyper-specific qualities of multiple persons. This “transparent” singular piece constructed of fragmented bits of plastic obfuscates not only each individual blob of glue itself, but also any visible conception that the fragmented pieces could at one time, not be a part of this whole. Moreover, each blob of molten plastic used in 3-d printing functions both as adhesive and as part in a unified whole. If we think about each blob of glue/part as an isolated set of user data, each individual user is not only indistinguishable from the next, it moreover becomes completely invisible which parts of the user were taken and then added to the whole. 3-d printing production stops once the model is made, but predictive machine learning-generative models continually create plural portraits of consumer behavior, and the more data absorbed by the model, the more what went into the model becomes closer to being completely abstracted.
My project removes the liquid “adhesive” nature of 3d printing but simply taking cubes from one place and "ingesting” them in a non-human way. The collection or final form is in the machine’s own body, rather than producing a separate object or prediction like 3-d printing and machine learning generate. Moreover, my project will not “learn” or generate anything outside of itself.