Photobooth Kiss

Can a machine create an image with human emotion?

Each diptych in Photobooth Kiss contains a photograph of a couple looking at the camera and a second image of the same couple kissing. The second image is made by a custom-built Artificial Intelligence program; it is an attempt to make a new “photograph” of the original couple kissing.

A Machine Learning neural network was trained on a database of couples in photobooths. Photobooth strips of 200 couples were collected and two photos from each strip were used in the training: a non-kissing photo and a kissing photo. The neural network “learned” what a non-kissing couple looks like when they kiss. Once trained, a previously unseen photograph of a non-kissing couple is input into the neural network and the program then predicts and renders a photograph of the couple kissing.

While the AI images in Photobooth Kiss fall short of their goal of being a new “photograph”, the images in this series demonstrate that it is technologically possible. If this technology succeeds, AI will have the ability to create photographs of kisses that may not have actually happened, and by extension, events that may not have happened. For now, the question still remains whether an AI generated “photograph” will ever be able to make a genuine kiss. Will it be able to capture a true intimate moment between two people?

Who are these couples in the photobooths? Some are friends, some are friends who are no longer with us, others are strangers. The subjects were chosen for their diversity in sexuality, ethnicity, and time period — to challenge our assumptions of who may or may not be kissing. We may never know if these couples actually kissed in the photobooth that day. Photobooth Kiss constructs an alternative photographic reality where, indeed, the two did kiss.