Jake Elwes is presenting the video Machine learning porn, which is entirely synthetically produced by a convolutional neural network. The AI is trained by showing it thousands of images with pornographic content. The training set was taken from Yahoo and usually served to help exclude pornographic material from search results. This work is based on running this algorithm in reverse (backpropagation) so that it generates images, which are more likely to be classified as pornographic by the algorithm. As the video progresses, one sees image sequences that have an ever-increasing probability of being pornographic.
Images trigger individual bodily reactions when observed by humans. Generally perception is dependent on one’s own body. With this particular object of observation, our biology, culture, and individual sexuality play especially important roles. This context, which determines the perception of the images is not available to the machine. What relationship do these algorithms then have to the object of their calculations? In the field of robotics, the belief is widely held that a general artificial intelligence is only possible in connection with an own body (embodiment). By using their body to actively intervene in the world and subsequently becoming aware these actions affect the agent itself, AIs can arrive at a more comprehensive understanding of the world.
London-based media artist Jake Elwes (b. 1993) studied Fine Art at Slade School of Art, University College of London (GB) and the School of the Art Institute Chicago (US). Elwes is currently showing in Bloomberg New Contemporaries 2017, a leading organisation supporting emergent art practice from UK art schools.