Life drawing and machine learning: An interview with artist Anna Ridler
Machine learning already plays a big part in your everyday life, and its role is only going to grow. Google searches and muttered requests to Amazon’s Alexa may tap into a veiled world of clever algorithms, but these techniques teeter on something much larger: a world of self-developing artificial intelligence.
Deep learning, and the neural networks that do the thinking, is becoming an integral seam to digital technology. By extension, artificial intelligence is having a growing effect on our experience of the world and, as an artist, it is a material that can’t be ignored. That at least is the thinking of Anna Ridler, who is building a name for herself with works that hoist machine-learning techniques and bring them into the gallery.
Ridler’s Fall of the House of Usher (2017), for example, involved training a neural network on the artist’s own ink drawings, which were themselves based on stills from a 1928 film version of the Edgar Allan Poe short story. The AI was then able to make its own version of the film; based on the type of images it thought should follow each frame, resulting in an impressionistic stream of inky interiors and half-glimpsed faces.
More than a tech demo, the piece pokes at questions about narrative and memory – how these form and mis-form in our minds, as well as those of machines.
For Drawing with Sound (2017), Ridler collaborated with composer Ben Heim, training a neural network to recognise shapes from the artist’s life-drawing sketches, then to interpret these shapes as sounds developed by Heim using a sampled soprano.
Framed as a performance, the piece encompasses Ridler using charcoal to make lines on a white surface. She wears a webcam embedded into a pair of glasses, which picks up the shapes she is making and translates them into a cacophony of processed human voices.
The foregrounding of Ridler’s body, coupled with the splinters of charcoal, the roots in life drawing and the complex computational processes hidden beneath it all, makes Drawing with Sound a particularly impressive investigation into the limits of what can be classed as human performance. I got in contact with the artist to ask her more about her work, and her general thoughts around machine learning and art.
How important do you think it is for artists to be interrogating machine learning and artificial intelligence?
Extremely important. Art can do multiple things. It can provide a way of engaging with machine learning for people who are outside academia or technology, particularly engaging with some of the moral and ethical debates around it. Without people outside the computer science community engaging with these ideas and interrogating them, they will be developed inside a bubble.
Art can also suggest new ways of working with this technology. I was recently at a conference where there was a lot of discussion about how science fiction has influenced technology and I think there is the potential for art to do the same thing with machine learning. Finally, for me, in my work, machine learning is a tool, just like ink or charcoal. It does different things and offers a different history when you choose to work with it. Its importance will only increase. To ignore it just seems to be wilful.
What was the initial idea behind Drawing with Sound?
I took all of the life drawings that I did over two years and scanned them. Then I broke them into grids and used something called a t-SNE to group the marks (so all of the diagonal lines together, all of the straight lines together, etc) to work out what were the marks that I most commonly made in my work. I then worked with a composer Ben Heim to think about what these marks would sound like.
“A line coming from one direction will generate a different sound to a line coming from the other”
It’s not only about what the lines would sound like when drawn: the act of drawing was very important to the piece. Drawing is of course both a verb and a noun. It reacts to the process – a line coming from one direction will generate a different sound to a line coming from the other. A semi-circle will start a sound that will move smoothly into a full circle. Making sure that all of the sounds that make up the composite parts of shapes that I commonly draw worked together sonically was actually very difficult.
Once we had all of the marks assigned to sounds, we then worked together to create a score that takes about half an hour to draw, which is both a performance and a notation at the same time. Because I know the relationship between the marks and the sounds, I can use the piece almost to play jazz.
(Above: Drawing with Sound, 2017)
We worked quite hard to make sure that the sound – a soprano singing scales – also matched the medium, charcoal. The female vocal going up and down and becoming louder and quieter, with the ability to make quite strong heavy marks and erase them, felt appropriate. The ephemeral quality of the voice matches the ephemeral nature of charcoal.
Also, both life drawing and scales are fundamental practices to both art and music. It felt appropriate to take them, cut them and use this avant-garde technology to make something new.
There’s a strong performance element to the work. Why did you design the piece around you drawing, and not offering up the tool for participants to draw the work themselves?
I was very concerned about this merely becoming a “cool tech demo” where people just mess about. I wanted this to be a piece of art, not design. The physicality of drawing for me was also important – when you draw, you move; it is very gestural. When doing this on a large scale, you almost end up dancing and I wanted to capture that in this piece. This is why it is glasses that you wear so that you are forced to look and relook and move your face and body, rather than merely a camera that zooms over a piece of paper on a table.
Although I don’t allow people to draw, I do allow them to wear the glasses to “re-play” the performance that I have given. While it’s the same drawing, each time someone tries to play it, they generate a new track. They stand in a different place, move more quickly or slowly, etc, which ends up being a nice metaphor for how people can never look at the same thing in the same way as someone else.
The setup of the piece is incredibly minimal – white walls, speakers, charcoals – so that the focus is very much of the moment of experience, on the sound.
Why did you want the neural net to interpret the drawings as sound, as opposed to some other output?
It seemed to me almost magical to be able to draw and create sound and I was inspired by the concept of synesthesia. I also wanted to push the idea of notation into new directions, inspired by people like Daphne Oram. There is a long history of alternative notation systems and trying to match sound with mark. I wanted to take this, but collapse the idea of performance and notation into one.
For a piece that uses a lot of technology, Drawing with Sound really foregrounds the human body. To what degree would you say this foregrounding is a reaction against the technology in your work?
I don’t think it’s a reaction to technology, it’s more a way of trying to use technology in new or perhaps more seamless ways. Things like machine learning now allow to do things that 15, 20 years ago would have been in the realm of fiction.
(Above: Fall of the House of Usher, 2017)
“At the end of the performances there is so much mess”
I’m not using AI to critique AI, but rather using AI to push what drawing can be into new realms. Part of doing this is by trying to use it with the body, by making it real and tangible and not just hidden away on a screen.
At the end of the performances there is so much mess that is generated – charcoal shards on the floor, smudged charcoal all over the wall, smears on the glasses – that I quite like. These traces of humanness that seem somehow out of place in a digital work, but I think are so important to have.
Image credit: Anna Ridler