A new commission from artist Trevor Paglan takes as its starting point the way in which Artificial Intelligence networks are taught how to ‘see’, ‘hear’ and ‘perceive’.
Artificial Intelligence networks are taught how to perceive the world by engineers who feed them vast training sets. These consist of images, video and sound libraries that depict objects, faces, facial expressions, gestures, actions, speech commands, eye movements and more. Paglen highlights how the advent of autonomous computer vision and AI has developed alongside this new kind of media, not designed for humans, but for machines, which are rife with hidden politics, biases, stereotypes and epistemological assumptions.
For the exhibition, Paglen has installed approximately 30,000 individually printed photographs pinned in a complex mosaic of images along the length of the curved wall. Using ImageNet, one of the most widely shared, publicly available collection of images, which is also used to train artificial intelligence networks, Paglen queries the content of images chosen for machine learning.
ImageNet contains more than 14 million images organised into more than 21,000 categories or 'classes'. In most cases, the connotations of image categories and names are uncontroversial i.e. a 'strawberry' or 'orange'. Others are classified under 'debtors', 'alcoholics' and 'bad persons'. These definitions, if used in AI, suggest a world in which machines will be able to elicit different forms of judgement against humankind.