Ganksy’s 00001101: promenade (2020)
With Banksy’s market hotter than ever, hopeful collectors might be keen to discover what appears to be a newly released collection of 265 works by the anonymous street artist. Except they are not.Rather, they are the creation of a new artificial intelligence (AI) software named GANksy, which has been programmed to create works that aim to mimic those of “a certain street artist”. For although its title and the images it generates leave little doubt as to who it is attempting to copy, for legal reasons, GANksy’s creator, the web developer Matt Round, declines to confirm nor deny which artist inspired the project. To create these images, Round has used a type of computerised machine learning framework known as a GAN (generative adversarial network). This specific GAN was trained for five days using a portfolio of hundreds of images of (potentially) Banksy’s work, until it was able to produce an image that bears a superficial likeness to the originals. Described by Round on his website VoleWTF as “a twisted visual genius whose work reflects our unsettled times”, GANksy “learns about the structures and textures in the works, looking for common patterns and themes and then tries to recreate that effect itself”. The results are a combination of uncannily resemblant—yet crucially dissimilar—images to the works of Banksy, which adorn many of the UK’s buildings and public spaces.
GANsky’s 00111111: warrior (2020)
Each of these 256 works is now for sale in the form of the exclusive ownership of a GANksy-signed digital file, with prices starting at £1 and rising by £1 every time one is purchased. When asked if Banksy, who is already struggling to protect trademarks for his most famous images, might take legal action against the project, Round says: “There’s absolutely nothing resembling any of Banksy’s work in the output, it doesn’t even have the same tone/style. Hopefully he would appreciate it as a fun, creative project.” “I found that images of street art are so varied, and so dependent on their context, that while the GAN ended up with a grasp of some of the forms and textures, it had no high-level understanding of specific objects or themes. It’s not mimicking the training material, but rather it’s learning a limited amount and then doing its own thing. People are appreciating the art for its originality.”
“I’m increasingly convinced that neural networks will be able to really mess with our perception”
GANksy is the latest addition to a growing conversation around AI-generated art and the potential implications it will have for both the art market, as well as society’s understanding of artistic authorship and process. In October 2018, Christie’s auctioned an AI-generated portrait by the French collective Obvious, also created by a GAN. It soared past its high estimate of $10,000 to sell for $432,500 (with fees). GANs, developed in 2014, have been heralded as the future of algorithm-based machine learning, due to their ability to more closely mimic human thought processing. Unlike older neural networks, such as Google DeepDream, they are able to create entirely new images due to a two-part process. First a “generative” algorithm creates a new image based on a given data set. Then an “adversarial” algorithm will try to distinguish the “fake” image from the human-made images, until it has been fooled successfully enough times. The potential uses for GANs has sparked widespread concerns in recent years due to their ability to create “deepfake” images. “I’m increasingly convinced that neural networks will be able to really mess with our perception,” Round says. “Imagine one trained on distressing imagery and optical illusions.”