Title: How to Trick Your Algorithm
Category: Research in progress, conducted as part of the Ventulett NEXT Fellowship at Georgia Tech
Year: 2023-present
This project asks what potential AI poses for the development of new material and craft practices in architecture, and how these practices can better integrate ecological goals than existing techniques. In exploring this question, I have developed a process of critical interrogation of AI algorithms used by popular programs like Midjourney and the visual data with which it is trained. As AI is continually marketed as a super-human, magical technology, it is easy forget that it is built upon an accumulation of immense amounts of data created by humans; an interrogation of this data can reveal larger political priorities and the AI algorithm is a method to extract, simplify, and sort this data. Furthermore, AI is ecological in that it draws things together in unexpected ways, and this process attempts to redirect AI’s objectives toward emergence, as opposed to the “rational calculative methods” it is currently engineered for.
Camouflage ModelA “turquoise and green plastic bag”, produced in Canva AI Hybrid image of the plastic bag and camouflage model, produced in Midjourney Architectural pavilion, produced in Midjourney
Rendered isometric drawing of architectural pavilion model, constructed and produced by authorThis process began with the fabrication of a “camouflage model” designed using a combination of computational drawings created using “noise” algorithms and spatial manipulations in Rhino. Drawings produced with “noise” are capable of confusing AI image recognition and machine learning algorithms, causing incorrectly identified images. This technique was used in order to push the AI algorithms beyond the boundaries they are engineered to exist within. Additionally, the process of artifact making relied on spatial manipulations intended to introduce another unexpected visual element to confuse the algorithm.
Once the “camouflage model” was completed, I asked an AI program to describe it. In this first exploration, the AI program presented several options but consistently labeled this image as a “turquoise and green plastic bag”. While the camouflage model is an abstract construction, the AI algorithm consistently attempted to identify it as an object, and continuously defaulted to one of the most ubiquitous items littered across the earth: the plastic shopping bag. This result was particularly revealing, in that the default to the plastic bag provided a peak into the data upon which the algorithm was trained.
Following this step, this description was subsequently fed back into the AI program as a prompt to create an image. This AI generated image was then combined with the “camouflage model” to produce a hybridized object. This was followed by the use of this image as part of a prompt used by the AI program to produce an architectural pavilion. This image was used as the basis for the development of an architectural model rationalized for construction. As the AI explorations utilized the imagery of plastic bags, the pavilion model used this imagery as the basis to explore architectural construction techniques using plastic. I built and rendered this model with the intention of creating a series of scaled models and detail models exploring how the reuse of plastic can to inform architectural construction.
Camouflage ModelA “colorful computer monitor”, produced in MidjourneyHybrid image of the colorful computer monitor and the camouflage model, produced in MidjourneyArchitectural pavilion, producedin Midjourney