July 31, 2020

What does an AI map show us?

Martin Disley

Multi-Channel VideoMapping Mechanical Imaginary, 2020

The ‘deep fake’, data intrusions and the ubiquity of algorithmic intervention have moved public consciousness to consider technology as an instrument of power. Through pairing a contemporary technology (neural networks) with a much older one (cartography), this work seeks to extend this locus of criticality beyond contemporary technologies (envisioned or operational), to those which are older (invisible or naturalised). Much like AI, maps have the potential to advance any combination of insight and lies, to construct reality as much as represent it. They seek to obscure contested histories through a form of representation which facilitates domination and control. This multichannel video installation details the mechanic hallucinations of a triptych of neural networks trained on collections of maps from the National Library of Scotland’s archives. Having been trained on thousands of images, the neural networks are able to generate compelling “fakes” of those images. The generative videos are renderings of what the neural network thinks each collection looks like. Their morphing textures expose the generative grammar that makes cartography legible. No longer tethered to a nominally objective rendering, they reveal how noise can be rendered coherent, and thus how falsehoods can be coded in institutional and computational legitimacy.