Read the whole essay here (in German)
The essay deals with the development of machine vision, of artificial vision, which today makes it possible to recognize faces and objects supposedly automatically and autonomously through neural networks (AI). The text employs the concept of “operative images” (Farocki 2003) used by Harun Farocki as a point of departure for the examination. Farocki’s trilogy Eye/Machine I-III reflects the evolution of this new type of image: images from the factory, from research laboratories, surveillance cameras and images from the Iraq war. For him, they are operative, i.e. directly effective: images that are actively embedded in a process as mathematical-technical operations. Farocki’s precise observations serve as a starting point from which both techno-determinist and popular dystopian narratives of all-seeing surveillance and control scenarios are contrasted with a differentiated materialist-feminist analysis, one that also takes economic changes into account.
Today, almost two decades later, it means putting both the perceptional labor invested in teaching machines how to recognize and the network of workers who carry it out in the foreground. Artificial intelligence is intrinsically linked to this work. It does not function ‘smart’-ly and automatically, as though by a magician’s slight of hand. Rather, its functionality is in the hands of gig economy workers. The magician is a collective of workers and the pixels of the pictures are alive.
Several examples are used to show that supposedly artificial intelligence is in fact animated by global production networks of click workers. The manifold contributions of these workers remain invisible. Their platform work, from filtering out pornographic or violent content to annotating images for object recognition, is ghost work; precarious yet essential work that makes today’s software systems seem smart (see Gray/Suri 2019).
The automation of perception, too often taken for machine autonomy, is revealed to be the result of a complex social relationship involving a planetary division of labor and the collective intelligence of many workers. Farocki described operative images at the beginning of the millennium as “bringing the work of recognition to the fore” (Farocki 2003).
Today, almost two decades later, it means putting both the perceptional labor invested in teaching machines how to recognize and the network of workers who carry it out in the foreground. Artificial intelligence is intrinsically linked to this work. It does not function ‘smart’-ly and automatically, as though by a magician’s slight of hand. Rather, its functionality is in the hands of gig economy workers. The magician is a collective of workers and the pixels of the pictures are alive.
