D1 67 J3 25 D6 32 n20 o07 r1 33

Photo: Gene Kogan

Mirror, Mirror, on the Wall …

Style Transfer

Take a look into Gene Kogan’s Cubist Mirror and you might be surprised by your own reflection. Captured by a webcam, your image becomes a living, breathing, and moving painting. Smile, yawn, or shake your head, and watch the canvas mirror your moves in real-time. Kogan’s fascinating ‘style transfer wouldn’t be possible without Artificial Intelligence (AI). To turn us into a living work of art, AI determines the key image characteristics and converts them into the style of Picasso, van Gogh, or other master painters. At two other stations, the artist and developer lets us try first-hand how Artificial Intelligence could help us support and unleash our own creativity.

D1 67 J3 25 D6 32 n20 o07 r1 33

Photo: Gene Kogan

Painting by numbers

“Here, visitors can sketch a landscape and a neural network will turn their sketch into a real looking landscape painting. Or they could piece together a simple map with buildings, roads, parks, and water and turn it into a satellite photograph,” explains Kogan. “Generally speaking, I use machine learning to ‘translate’ one image into another.”

All three stations follow the same principle: Kogan feeds his neural networks very large data sets to convert one piece of information to another. His programs keep scanning this data for patterns and correlations. According to the artist, “learning how to map one kind of information into another kind of information based on previous examples is essentially the idea of learning.” Machine learning is the art of letting a computer do useful things without actually programming it to do so.

We build this city

Kogan’s urban planning exercise lets visitors create their own city of the future from a few coloured discs, representing streets, houses, parks, and water. The markers are quickly assembled into rough, imaginary maps. Captured by a camera, Kogan’s AI transforms our improvised cityscape into a satellite image. To achieve deceptively realistic results, Kogan trained the neural network with a large number of bird’s-eye views of Berlin. This translates into a colourful way to get involved in city planning.

A matter of training

At the same time, there’s plenty going on behind the scenes of this seemingly effortless process. Real-time transformations take plenty of brute programming power and a surprisingly large amount of training. “In addition to a lot of open source software and large data sets, you need some very powerful computers and a lot of time.

Although machine learning has been around for about 60 years, most of the advances have been made over the last five years, especially in so-called ‘deep learning’. Now, artists like me with no scientific background can use machine-learning software for our own work. I like to call it ‘augmented creativity’. Why? Because now we can do some very, very complicated things (masterful paintings) with very, very simple inputs (a few strokes of the pen).”

A brave new world

The actual idea behind it – recognising, transforming and restaging people, objects and emotions – has more than mere artistic value. It also promises to play an increasingly important role in everyday life, from medical technology to stage performances and to autonomous driving. “There are already applications on your phone that turn you into a painting or personal digital assistants that answer questions for you. These use multiple different kinds of machine learning in order to work out what you say, then to get an answer for you and to turn that into a language you understand.”

While excited by the technology’s potential, Kogan has clear reservations about the way AI may filter into most aspects of our life. “I think the future is going to be really weird. We're going to have all sorts of new questions to answer. What will happen when anyone can generate a realistic image of anything or anyone at any time? In times of fake news and deep fakes, how will we be able to tell ‘real’ content apart from computer-generated content? Our greatest challenge will be to create AI agents that interact with their environments in a way that is safe and benefiting humanity.”