Object 3

In these prints, I explore the idea of the ‘object’ that is intrinsic to image recognition and annotation, to printing and its associated abstractions and the many small black holes of uncertainty that lie within computerized vision. Different levels of objectification exist here, each one abstracting the spaces further as further: from opening up intimate sites of work (for the voyeuristic human eye and the impersonal algorithmic eye simultaneously), to object detection algorithms that split, cut, re-write, and re-cast these spaces as legible and illegible.

Digital images are sent into Amazon’s Mechanical Turk in the crores to be broken down into their elements (a tree, a car, a person, a road) known as objects. For computer vision to take place, this is crucial. The results are often fed into machine learning neural networks so that algorithms can identify these objects when it ‘sees’ one. Sometimes, however, the vision goes terribly wrong. No amount of training can ensure computer vision can determine that a cat will be identified as a cat every time it sees one; it can however get better and better and better at it.

The following prints play with this error. A series of images sourced from Mturk workers of their place of work* were fed into image recognition algorithms. The resulting images, meant to be digital replicas of the original, were hardly recognizable. These unrecognizable images were then broken up into colour channels – further abstracting them, distilling them, breaking them down further – only to be printed as a constructed ‘whole’ again. How we understand and view these images are a part of how we understand techno-determinism, rooms-for-error, and the potential for outlandish storytelling.

*The site of work is an important feature of online gig work. It can be argued that this kind of home working space is the most updated form of the IT cubicle.

edition of 15 risograph prints.