Noise Performance as Knowledge Production: A Speculative Database Aesthetics of ImageNet

ABSTRACT. SOMA is the performance project of the speculative audio-visual-bio-mimetic platform agency, GOVERNANCE INC, that uses custom electronic tools, signals, digital DNA, speculative philosophy, and genome sequencing to create hybrid soma-cyborganisms. Anabolic performance machines sequence genomes in order to propose more robust possible cell structures for digital organic life moving forward. SOMA mines ImageNet, a large-scale image database used for machine learning processes, to find its genomic origin and biosynthetic emergent forms. These cyborg cells all present different windows onto an unfolding future. SOMA is an engagement with performative knowledge production. It seeks to “shift the frame” away from the norms of scientific discourse to question the conditions of the production of knowledge itself. How does performing a discourse— engaging with it visually, experimentally, or narratively— allow new readings of the taxonomies contained within? How can performance, as a form of knowledge making, decode or change these taxonomies? We suggest that noise performance can stand in as a “proxy” for the performativity of knowledge production within computational machine learning and image recognition databases. Through rearranging, selecting and manipulating the material that machine learning systems metabolize to produce “truths”, we seek to produce different affective relationships with the archive. We suggest that the archive might be reproduced differently through experimental engagement. SOMA considers performance as proxy, or stand-in, for the computational algorithms that typically automate the processing of machine learning databases. It is suggested that engaging the body and activating the framework of experimental noise performance might allow a different expression of the database logic that increasingly undergirds and orders our aesthetic regime, visual culture, and social norms. What bioforms might emerge from the flesh of ImageNet when it becomes a performed object?