The deception of an infinite view – exploring machine vision in digital art

Abstract

This paper examines prediction product called Queryable Earth a project to “make Earth searchable for all”. A project pitched by the company Planet, owner of the largest fleet of Earth-imaging satellites in orbit and an archive of satellite images growing with terabytes of fresh data every day. The aim of Queryable Earth is to combine geospatial intelligence with machine learning. By training artificial neural networks to classify objects, identify geographic features, and monitor change over time, the implied intention is to create a predictive, omniscient oracle. In this paper Queryable Earth functions as an example of a ‘nonconscious cognitive assemblage’ combining aerial image with machine learning techniques such as artificial neural networks. To examine the predictive potential and the assumed objectivity of machine vision systems such as Queryable Earth I turn to histories of aerial photography and examples of contemporary digital art to illustrate how human and technical cognition entwine revealing how seemingly automated processes such as rendering of satellite images and pattern recognition still inherit human biases and are prone to emphasize them. Furthermore, I use digital artworks to illustrate how Queryable Earth as an “all seeing machine” is limited to a singular aerial perspective which cannot penetrate the surface and how predictions produced by such systems are constrained the quality and selection of data they are trained on.