Fall of the House of Usher I, 2017

Single channel video installation (made with a series of interconnecting GANs) 

Duration: 12:00

Fall of the House of Usher I (2017) is a 12 minute animation, exploring Watson and Webber’s 1929 silent film adaption of the Edgar Allan Poe short story from 1839; a tale about decay and destruction. The work fuses two hundred ink drawings made with the technical process of machine learning to accentuate the horror story in the original film and notions of fear around artificial intelligence itself. Each still has been generated by a general adversarial network trained on original ink drawings. 

A hand made training set of two hundred ink drawings inspired by the 1929 silent film version of The Fall of the House of Usher, was processed using a series of interconnected general adversarial networks. The end result was an animation created from a series of machine made stills. By manipulating the reciprocity and feedback between the original film, drawing and this form of technology, the film’s original motifs are Ridler heightened and intensified by liberating fugitive aspects of memory to create a sense of the uncanny that is partly machine-made.

This work is the predecessor to Fall of the House of Usher II (2017), a display of the artist’s ink drawings which serve as the dataset for the animation. Exhibited at Ars Electronica (2017), the V&A Museum (2018), Nature Morte (2018) and the Centraal Museum (2018).

Please email for a link to the film.

Research & Process

Fall of the House of Usher I  explores concepts of repetition, remembrance and re-creation that are central to the original story by Edgar Allan Poe and the 1920s silent film version.  By manipulating the reciprocity and feedback between the original film, drawing and this form of technology, the film’s original motifs become heightened and intensified by liberating fugitive aspects of memory to create a sense of the uncanny that is partly machine-made.It is a copy of a copy (film) of the original (book); accordingly, things appear and disappear, are remembered or misremembered or mis-imagined and it calls into question our ability to recall one perfect version, ideas that are integral to the central story.  

The images that are created by this type of early GAN tend to all have a very distinct quality. They are watery and painterly and have the quality of being touched, by a non-natural intervention where materials can “develop uncanny lives of their own and display their power to metamorphose”.  This style (particularly in a “poorly-trained" model, or a model that is only given a small amount of data or trained for only a few epochs) is ruined, decaying and decomposed and studded with artifacts, the errors that expose the artificial nature of the image produced, the result of the uneven overlap when an image is being made by the GAN. These imperfections - the traces of the process -  that I was interested in working with in this piece. 

Using Watson and Webber’s 1929 silent film version as the base, each still was regenerated by a general adversarial network trained on original ink drawings; the output of which was then used to form the basis of the next training set. This project became  a loop now between copying and repetition, both manually and by the machine, between the labour intensive nature of creating the drawings for the training set and the speed of the algorithm producing the stills. There is an interplay between the digital and the physical: the original film was on celluloid film that was transferred to digital that was the printed that was then drawn and scanned and then reconstructed by the GAN. Celluloid is prone to accident and deletion and I kept this element, drawing in the light leaks and scratches that I saw on the film. I also drew in the errors and artefacts of the GAN  when I made my second set of drawings, all of which are unique to the particular model that I made. Although mechanically produced, the piece is unique to that particular cut of the film and output of the model.

It is a piece that could have been hand animated but by choosing machine learning I was able to heighten and increase these themes around the role of the creator, the reciprocity between art and technology, and aspects of memory in a way that would not be available to me otherwise. The errors and choices that are made when drawing are amplified. Eyes it finds difficult and faces. It  holds a mirror up to my own drawing and makes me realise things that I was not aware of: what do I find the most important, what are the things that I always edit out (eyes it keeps turning into eyebrows as I draw them so similarly, a chair that appears and disappears because sometimes I remembered to draw it in and sometimes I did not).  Machine learning models learn the human-ness and mistakes that are fed into it. 

There are moments when the model produces images that are incredibly good - at the start, in the sections where I have given it lots of moments of reference. As the animation progresses, it has less and less of a frame of reference to draw on, leading to uncanny, eerie moments and repeating and remembering  leads to misremembering and as the film progresses, the information starts to break down. There are no training images so the programme is having to construct every frame from what it already knows leading to uncanny, eerie moments. Sometimes this works, sometimes it does not. Claude Shannon, the inventor of information theory, talks about the minimal amounts of information needed for understanding how “familiarity with words, idioms, cliches and grammar enables us to fill in missing letters in proofreading and unfinished phrase in conversation”.  Familiarity with cinema and classic Hollywood, which have their own cliches and idioms and grammar allows the viewer to grasp hold of and understand meaning. Memory becomes part of the material that is needed to understand the work.