New media art has been around nearly two centuries, and with every new technology, whether intended for art or not, there is an inevitable burst of creativity. These developments are systematically deconstructed and reimagined by artists in a ceaseless endeavour to find novel ways of expression and experimentation in the modern age.
For this reason, it should be no surprise that in the age of mass information, there has been an increase in data-driven artwork. Every day sees quadrillions of bytes of data are being generated globally. With the internet at our fingertips (and in our pockets), it’s fair to say that data-driven digital artists have a wealth of inspiration at their disposal.
But how is one person expected to digest and interpret all this data alone? This is a question creative coders have been asking for years now, and it seems as though GAN’s may be the answer they were looking for.
Generative adversarial network (GAN) is a deep learning-based generative model designed by Ian Goodfellow and his team in 2014. By harnessing the power of neural networks, machines can process mass amounts of data and begin to recognise patterns within the sets to generate their own data that is so accurate it could’ve been part of the original set.
While these systems have an enormous amount of practical applications (think facial recognition, advanced search engines, and even predicting stock prices), digital artists and creative coders have found unique success in training systems to generate art. These systems, guided by the hand of their programmers, have become tools for exploring the world of possibilities created through collaboration between artist and machine.
Here are a few examples of the ways these emerging technologies have ushered in a new wave of art:
Mario Klingemann: Natural Decay_756
Created by training a system on portraits from the nineteenth century and manipulating the output with networks trained on organic decay processes, this series is presented as an unsettlingly familiar rotting reflection on history.
Nathan Shipley: Generated Figure Drawings and Pop Art
In this video, we see the interpolation of a GAN being trained on the figure drawing and pop art sections of WikiArt. The two pieces are trained with the same random vector interpolations, giving similar motions despite varied styles.
Sicilian Alleyway - Joanne Hastie
By combining a set of photographs taken while on an artist residency in Italy, this painting is created by applying a network trained to emulate her painting style. This creates a work of her memories and creative practice, as performed by a machine.
Uw0GSnfgvN: Momento Mori - Zdzislaw Beksinski
This video shows the process of a GAN learning from a dataset of Zdzislaw Beksinski’s works. Compiled of 500 iterations of output, this piece demonstrates the process of learning and presents these intermediary stages to animate the thought process of the neural network.
This is so cool! I didn't know anything like this existed. Have you had the chance to try it out?