Latent prism

Illuminated sculpture of a translucent aerial scene on a marble pedestal surrounded by long scroll of credits containing training data authorship information for all 17 thousand images used to train the model that generated the sculpture frames.
An AI-infused lucite prism unveils an ethereal realm of undulating landscapes while challenging the concepts of credit and artistic ownership amidst data-driven transformation.
Abstract
Latent Prism is a visually captivating and thought-provoking piece that incorporates artificial intelligence (AI) and data visualization. It presents a projection of an imagined environment, created through a generative adversarial network (GAN) trained on thousands of aerial photographs from royalty-free stock photo websites. The sculpture takes the form of a polished transparent lucite prism, within which layers of translucent mylar film are suspended. These films display frames extracted from an AI-generated video known as a “latent walk,” showcasing undulating ocean and forest landscapes captured from an aerial perspective. Surrounding the prism is a haphazardly piled 120ft. (36.5m)-long roll of credits, listing the names and photographers for every image used to train the model on a 2.5in (6.35cm) wide strip of drafting paper. The sculpture echoes this year’s VISAP theme of Perpetual Presence, which emphasizes the transformative power of data and its influence on cultural, economic, and political landscapes. By representing an imagined environment derived from AI-generated imagery, the artwork visualizes a vector of a possble environmental future, one which also subtly suggests at the high carbon costs of training large generative models, through the illustration of rising tides. In piling the credits for the training data around the base of the sculpture, the work calls into question the nature of 'credit' and the burden of acknowledgement in modern fine arts, blurring the line between medium and authorship, artwork and gallery tag.
Materials
PDF | Preprint | News | BibTeX
Authors
Citation

Khoury Vis Lab — Northeastern University
West Village H, Room 302
440 Huntington Ave, Boston, MA 02115, USA