14 September 2020

    Anthea Kemp

    Art through AI: A possible future?

    Making art is now one of the many things artificial intelligence (AI) can accomplish, and the results are compelling enough that people are paying top dollar.

    When Claude Monet began to create the last group of his famous ‘Water Lilies’ series in 1920, it was the culmination of a lifetime of learning to look at the world differently. Painted right up until his death in 1926, the series transformed an ordinary garden water feature into a magnificent, immersive world of colour and light. Much larger than life, at roughly 2 x 12m each, and displayed in the curved exhibition halls of the Orangerie Museum in Paris, they give you the feeling of being right in the pond with the lilies. They were a monumental artistic achievement by an artist whom many consider to be a genius of the modern era.

    AI Art

    The next Monet, however, could be a machine. Making art is now one of the many things artificial intelligence (AI) can accomplish, and the results are compelling enough that people are paying top dollar. Already in October 2018, the world’s first AI-generated artwork went under the hammer at a Christie’s auction for a whopping $432 500, over 40 times the auction house’s estimate of $7 000-10 000. Attributed to the French artist collective Obvious, though made entirely by AI, the work is a portrait of a fictional 17th century nobleman named Edmond de Belamy.

    The portrait, a print on canvas, depicts a vaguely blurry figure whose face is somewhat distorted. Though it’s obvious that the picture is in the style of 17th century portraiture, its features look slightly melted, as if an oil painting had been subjected to a mysterious alchemical process. The method of generating ‘Edmond de Belamy, from La Famille de Belamy’ is not dissimilar to alchemy, and as far as human brains are concerned – or at least this human brain – it certainly is mysterious.

    Like other AI-generated images, the portrait was ‘bred’ using a Generative Adversarial Network, or GAN, a type of algorithm created by a programmer named Ian Goodfellow in 2014. GANs work by training two artificial neural networks using a massive sample of learning images, reference material the GAN will use to generate its own images. For the ‘Edmond de Belamy’ portrait, for example, the GAN was trained using portraiture from the 17th century. The two neural networks are then pitted against each other. One is in charge of generating new images based on the reference material, and the other is responsible for discriminating among these and the sample images, rooting out those which are obviously machine-made and holding onto those which look natural or man-made. As this system becomes better at filtering out the obvious machine-made images (this is the learning part), it becomes harder to tell the difference between the two classes of images. When the discriminatory neural network is no longer able to discriminate, its job is done and an image output is created.

    See for yourself

    The Obvious collective has arguably opened up the market for what some critics think will become a massive sub-genre of fine art. Since its auction debut, more AI initiatives have emerged in traditional art settings, like this exhibition of AI-generated art at the HG Contemporary Gallery in New York last year. And developers, savvy to the wider appeal of this form of creative expression, have created ways for ordinary people to play around with AI art too. Take a trip through Artbreeder and see for yourself!

    Vodacom future options

    Vodacom has a wealth of different options to choose from when you need data. Vodacom Just 4 You offers incredible, tailor-made offers personalised for your own specific usage. Vodacom's fibre offerings also vary according to your own needs. Check out Vodacom's Fibre deals here.

    For tips on how to save data on your other favourite social media platforms, check our #Datawyze tips here.

    Anthea Kemp