GANs vs cGANs
- GANs or ‘Generative Adversarial Networks’ are a set of instructions that engage as networks do. The most common application of this is the production of images that one can tell aren’t real, despite their authentic aesthetic.
- A cGAN or‘Conditional Generative Adversarial Network’ is an upgraded set of instructions but with a lot more context around how they function.
Thanks to technology, the roles of the artist and the scientist, as well as their creative processes, are melding and the rigour of the resulting works is unbelievable. Our creativity has allowed our species to thrive, and now with machines having their own concepts of aesthetics, perception and creativity, it poses some important questions to artists. Artificial Intelligence (AI) is bridging the gap between technology and fine art by challenging creative intelligence and sensorial experiences, integrating networks known as GANs and cGANs, and changing how artists interact with their so-to-speak canvases.
And with India as an AI capital, there’s endless scope to explore this realm. In fact, ‘Gradient Descent’, the country’s first AI creative showcase in Delhi, will be exhibiting this far-reaching evolution.
Assembled by brothers Raghava KK and Karthik Kalyanaraman, co-founders of art curation and research collective 64/1, the show unites an international group of artists who have melded various nodes of AI into riveting artworks which vary across the scale of immersion; building public understanding of how artists and Artificial Intelligence can come together to create art for the post-human age.
‘The Anatomy Lesson of Dr. Algorithm’ by Harshit Agrawal
Bengaluru-born Harshit Agrawal’s works, titled ‘The Anatomy Lesson of Dr Algorithm’, appears to be a disturbing look at surgical procedures. MIT graduate Harshit explores how something that’s so fundamentally human can be performed by something mechanical, like an algorithm he uses to map out a surgery. By varying the amount of training the algorithm gets, Harshit generates vivid abstract painterly images. At the same time, he refuses to erase the pixellations and digital mediation in the artwork — helpful to those who’d be a bit queasy.
‘79530 Self Portraits’ by Mario Klingemann
Mario Klingemann from Germany will be presenting a jarring video of image distortions by AI’s sometimes-unpredictable tendencies. Using AI that has been trained on portraits of the Old Masters, Mario presents a bizarre take on how AI can build upon images. Because he’s continuously adding noise to this image, the portrait, which is morphed, starts to become more and more bizarre.
‘Perception Engines’ by Tom White
Now what about AI’s perception as artwork? A pioneer of AI and art, Tom White from London will be presenting two provocative series. The first series starts from a simple drawing made by an algorithm. Then, an AI algorithm which was trained on photographs from ImageNet, slowly transforms the initial sketch into an abstract representation of a physical object. This painting will not necessarily be recognisable to humans. The second series, a little more provocative, uses the same systems, but instead produces images that will generally be categorised as ‘inappropriate’ or ‘having explicit content’ by online image analysis AIs.
‘Closed Loop’ by Jake Elwes
Grasping the potential of film and power, Anna Ridler from London uses the first 4 minutes of the 1928 film The Fall of the House of Usher by James Sibley Watson, handpaints every frame as a painting and feeds them into a cGAN, a type of AI algorithm that can link the frame from the film and Ridler’s artwork. Then she can generate a whole film using those paintings.
Jake Elwes from London explores the power of language as an art, stringing together sentences and using that language to create an image. Elwes then puts an AI that interprets images in words in conversation with another AI that interprets words in images. A simple concept, the resulting interaction and interpretations made by the two AIs is celebrated for its impact on the viewer.
‘Deep Meditations’ by Memo Akten
Turkish artist Memo Akten brings over his the investigation, ‘Learning To See’, of how the experience or memory of the machine affects what and how it sees. He describes it as “an artificial neural network making predictions on live webcam input, trying to make sense of what it sees in context of what it has seen before.” He’ll also be unveiling ‘Deep Meditations’, an extension of the aforementioned works, using machine-learning algorithms to create a two-hour meditative channel video and sound installation.
‘Imaginary Landscape’ by Nao Tokui
Taking AI perception to the next level, Nao Tokui from Japan will be exhibiting ‘Imaginary Landscape’. As various objects are manipulated in front of a camera, the machine, trained only on specific sets of images (for instance of beaches, or flames) reinterprets these objects as instances of categories it knows. The eerie effect of seeing a mobile phone power adapter reinterpreted by the algorithm as a cliff by a beach is not one a viewer is likely to forget, perhaps causing her to wonder whether this is all of nature that will be left for the future.
Through 64/1’s eyes
Karthik and Raghava share that finding work that is conceptually rich was key, adding, “This is the first time we are actually focussing on AI art as art — usually AI art has been treated as a step to talk about AI in society, but we do need to speak about it as art too, because that brings up the essence of creativity and the role of the human. Stylistic sensibilities across the spectrum were also key here, as well as the spectrum of algorithms they were using. The show serves as a manifesto saying ‘we really care about this and this is how artists can engage with the future’.”
For those overwhelmed by the tech-talk, the exhibition will also feature some workshops and engaging talks about AI, so even those unfamiliar yet curious people can immerse themselves into a new genre of art. Running till September 15, the show at Nature Morte gallery is open to all.