26 October 2018 Issue 25: Oct/Nov 2018

Portrait of the artist as an algorithm

Armed with three laptops and good wi-fi, a Paris-based collective has started using artificial intelligence to make art – although not in everyone's eyes.

In a traditional gold frame, the picture appears unfinished though eerily familiar – like some Old Master or a Cézanne. It’s of a man, his facial features blurred, rising from a dark frock coat and white collar.

A label beneath a matching portrait of his wife says: ‘Have you met my husband? Rumours say he was created by an AI. I’ve known him for my whole life. It can’t be, can it?’

The answer is revealed in the bottom right corner of the portrait via the signature of, according to the Financial Times, the ‘hot new artist’ of 2018: the algorithm min G max D Ex [log D (x))] + Ez [log(1—D(G(z)))].

The painting, if it can be called that, is titled Edmond De Belamy and is one of 13 controversial works created by Obvious, a Paris-based collective of three best friends: Pierre Fautrel, Hugo Caselles-Dupré and Gauthier Vernier. In late October, it will go under the hammer at Christie’s in New York – the first time a work of art made by artificial intelligence has been auctioned.

The sale marks a seminal moment, but trying to convince the art world of the team’s motto – ‘Creativity isn’t just for humans’ – is proving tricky. While there are creative coders using similar technologies out there, few of them are considered contemporary artists. And many prominent art-world figures insist on calling Obvious a startup. Either way, says Fautrel, AI art is here to stay: ‘Over the next five to 10 years, it will completely disrupt the art world. This is just the beginning.’

Method

The trio live and work together out of an apartment in the 10th arrondissement of Paris, a short walk from the cafés and bars that line the banks of Canal St Martin. Before starting Obvious, Fautrel had founded his own record label, Vernier was finishing up at business school and Caselles-Dupré was completing a masters in AI.

‘I’m the most entrepreneurial,’ explains Fautrel. ‘Gauthier is the best with clients, media and the art world, and Hugo is the really brainy one.’

Around the kitchen table one night in April 2017, they came across a paper by Ian Goodfellow, an American AI researcher, now based in San Francisco, about ‘generative adversarial networks’, or GANs. (Obvious titled their first works Belamy, a rough French translation of Goodfellow, in tribute to him.)

‘Straight away we could see the potential for how to combine the algorithm with art,’ says Fautrel. ‘We wanted to teach a computer about art history and show it how to make its own work.’

Every day over the next five months, they worked from 7pm into the early hours collecting a set of 15,000 portraits from an online art encyclopaedia, spanning the 14th to the 19th century, and feeding them into the GAN algorithm.

Put simply, GAN algorithms have two parts: the generator and the discriminator. The generator learns the ‘rules’ of the portraits – for example, noting that all the faces in the pictures fed into it have two ears, a nose and a mouth, or that no one ever has face tattoos. Then the programme starts to create new images based on those rules.

Meanwhile, the discriminator’s job is to review the images and guess which are ‘real’ ones from the dataset and which are ‘fake ones’ from the generator. As the generator attempts to fool the discriminator, it learns with each failure. When it succeeds, the process is finished and a new image is produced.

Fautrel compares how the algorithm works to the relationship between a student and teacher. ‘The student makes a picture and wants to fool his teacher that it’s a real Picasso. So they make thousands and thousands of pictures until, one day, their teacher eventually thinks they are looking at an original Picasso. Once the teacher is fooled, the image is ready.’

The pictures produced are not, Obvious insist, copies or a mixture of the existing 15,000 images fed into the programme. ‘When we are happy with a work, we think of it as completely original, as the 15,001st image,’ says Fautrel. ‘That’s why it’s so cool.’

Startup or artists?

The apartment Obvious lives in and works from looks more like student accommodation than it does an artist’s studio. ‘All we need is three laptops and good wi-fi,’ says Fautrel. Moving forwards, Obvious will start producing works more quickly and in contemporary or impressionist styles. ‘It was important that we started by making works of old-looking portraits in golden frames, so when people look at it them they instantly think of the kind of “proper” and “serious” art they see in museums,’ says Fautrel.

‘Of course, we are a business,’ he continues, ‘we need to make enough money to make the next work, and we require a lot of computer power, which isn’t cheap. But above all, we view ourselves as artists. At the moment, we are on the edge of being thought of as a startup, which is bad for our credibility.’

In fact, the history of art and the history of technology have always overlapped. Artists are often defined by the tools available to them: flint knives were considered to be cutting edge technology in the Stone Age, enabling humans to sculpt the first pieces of figurative art out of mammoth ivory.

When cameras were invented in the mid-1800s, they were used by highly qualified engineers while artists shunned them. More recently, Spotify caused controversy among musicians when it hired Francois Pachet, a French professor and AI researcher, who teaches computers to create their own music.

With digital technology now enabling machines to recognise, learn from and respond to humans and the world – from driverless cars to sex robots – an inevitable question follows: can machines ever truly be creative? And if they can, where does that leave humans?

Hammer time

For the prominent art collector Nicolas Laugero Lasserre – known for his collection of urban art by the likes Banksy and Invader – the work produced by Obvious is ‘grotesque and amazing at the same time’.

In February, he bought Le Comte de Belamy for close to £9,000. The portrait will join the other 150 human-made works in his collection. ‘I just find it amazing that some young people built a program allowing the creation of an original artwork,’ he said.

It was his purchase that led Richard Lloyd, of Christie’s auction house, to discover Obvious’s work. It remains to be seen how bidders will react to the AI work, but Lloyd remains optimistic.

‘The top of the art world are having a difficult time taking it on board, while the generation coming up are more relaxed,’ he says. ‘AI-generated art could be the next medium for the art market to embrace, and we wanted to be a part of that conversation. In some ways, what Obvious is doing is just as profound a challenge to the concept of art as the introduction of photography or the questions Duchamp was asking.’

For Lloyd, ‘What’s amazing is you have these three young guys with enormous analytical and mathematical skills. Instead of working for Goldman Sachs or somewhere similar, they’ve channeled their intelligence creatively. The art world isn’t a natural habitat for them, but maybe that’s changing.’

Despite the controversy, the collective remain relaxed. ‘We’re the new generation of creatives, it will take time for people to understand our work. Hopefully, after the auction, we will be taken more seriously,’ says Fautrel. ‘But for that to happen, the portrait had better sell.’