Abstract
This paper proposes a shift in how we understand artificial intelligence models—not as simulations of human intelligence, but as representations of cultural patterns. Building on the concept of “Models of Culture,” this research investigates how contemporary multimodal models, specifically CLIP-based embedding models, interpret and reproduce subcultural aesthetics. Using the SWAG era as a case study—a fashion movement rooted in the late 2000s and early 2010s—we examine whether style-based aesthetic values can be computationally measured. We explore two central research questions: (1) Can embedding spaces be used to measure personal style (SWAG)? (2) What cultural biases are reflected in such measurements? By designing a method to quantify SWAG through similarity scoring in latent spaces, we reveal how embeddings encode subjective visual cues. A Digital Probing Device enables interactive navigation through these spaces, making visible the structural relationships and latent assumptions embedded in the models. Our findings are presented through an interactive exhibition that visualizes these latent spaces and invites reflection on the cultural dimensions of machine learning systems. Rather than evaluating AI by its proximity to general intelligence, we advocate for viewing these systems as mirrors of the data cultures they emerge from. This approach highlights how computational models participate in aesthetic judgments and cultural memory, offering new ways to critically engage with machine learning and its role in shaping subcultural narratives.
Presenters
Suzan Ela HanowStudent, MA Design and Computation, Technical University of Berlin and University of Arts Berlin, Berlin, Germany
Details
Presentation Type
Paper Presentation in a Themed Session
Theme
KEYWORDS
Models of Culture, Embeddings, Subculture, Aesthetics, Cultural Memory, Latent Space
