Sean Scanlan

This project involved using digital trash as the raw material to construct recombinant images, and then using machine learning to describe those images. Over the course of a few months I saved the bytes from all of the screen shots I threw away on my computer. I then created new images from these bytes through a process similar to digital deletion, where I overwrote bytes of one image with random bytes from all the other deleted images. The resulting images were extremely broken: I would open them five different times and be shown five different images. Blocks of color would be shifted, moved around, changed. However, using a machine learning algorithm to provide descriptions of the images would return a consistent output every time, regardless of the display of the image. The images above are examples of the results of this process, in this case they are all a file that was labelled as "a television screen with a picture of a cat on it."

This work is about the aesthetic and informational potential of digital waste within the context of the shifting nature of visual culture. In A Sea of Data: Apophenia and Pattern (Mis-)Recognition Hito Steyerl writes “Not seeing anything intelligible is the new normal... Seeing is superseded by calculating probabilities. Vision loses importance and is replaced by filtering, decrypting, and pattern recognition.” Machines are excellent at this: machine learning algorithms are trained on hundreds of thousands to millions of images, and then when they encounter new images they are able to filter, calculate the probabilities, and use pattern recognition to decipher them. But, when presented with “images” that are actually disjointed records and not a visual representation of something the process becomes disconnected from what we think of as vision. We cannot ascribe meaning to glitched images containing bands of color, but a machine easily can. Instinctively, it feels like the machine is wrong. However, this is only the case if we prioritize human vision, which increasingly makes less sense given that most images now are created and interpreted by machines. When looking at these broken, recombinant images constructed from fractured records of myself, what does the machine understand that we do not?