THE MEANINGFUL
NOISE COLLECTIVE

(Joint work: Giacomo Aldegheri, Alina Rogalska, Ahmed Youssef, Eugenia Iofinova. Published at NeurIPS for Creativity Workshop, 2023) We propose a method to ’hack’ generative models, pushing their outputs away from the original training distribution towards a new objective. We inject a small-scale trainable module between the intermediate layers of the model and train it for a low number of iterations, keeping the rest of the network frozen. The resulting output images display an uncanny quality, given by the tension between the original and new objectives that can be exploited for artistic purposes.

(Published at NeurIPS for Creativity Workshop, 2024) Recent advances across the field of machine learning have created a world in which the existent, publicly available models, training tools, and compute enable unprecedented access to model building and deployment. This development creates a number of well-documented novel dangers, it also presents an opportunity for relatively under-resourced artistic cooperatives and even individual artists, to create interactive and performance art outside of the scope we conventionally see in the “AI Art” community today. Inspired by this development, this work argues for the artistic merit of expanding the scope of what we think of AI Art far past what we have so far seen exhibited in venues such as NeurIPS and in museums. We propose a movement toward art that is more directly provocative, that centers humans rather than the “AI”, and that engages with model training rather than simply inference. As a proof of concept, we describe a fictional interactive exhibit, the Penametron, which invites the users to interact with, and contribute to the training of, a model that estimates the length of a (fully clothed) visitor’s penis.

(Exhibited at Vienna Art Week 2025 and Austrian Culture Forum in Warsaw 2026) The Penametron v1 is an interactive art exhibit whose primary purpose is to estimate the total length of penis present in the camera’s image using state-of-the-art machine learning methods. In obtaining and publicly showcasing this estimate, the Penametron asks the viewer to engage with the breakdown of the divide between public and private information in the age of AI, the provenance and veracity of data on which the models are built.