Here, i teach an AI what the human insides look like by showing it several videos of surgical operations and dissections. The algorithm is then allowed to produce its own images of imagined dissections, which are
animated to mimic the flow within. By experimenting with the amount of training the algorithm gets, i try to generate vivid abstract painterly images which recall Shiraga and sometimes de Kooning. There is an ironic reference to Rembrandt’s early masterpiece in the title- that painting (The Anatomy Lesson of Dr. Nicolaes Tulp) was produced by Rembrandt in an era of troubled fascination with medical technology- this one is produced by AI, in an era of troubled excitement about its rise.

This work was part of the iconic ‘Gradient Descent’ AI Art show in 2018 at Nature Morte Gallery in Delhi, India. It is touted as the first AI Art show in a contemporary gallery globally. i was the only Indian artist among 7 global AI Art pioneers. The other artists were- Mario Klingemann, Anna Riddler, Tom White, Memo Akten, Jake Elwes and Nao Takui.

One of my early works made with GANs. i’ve often stated- generative AI models should feel more like pencils and not stencils and GANs feel like that.

Do Universals Exist? Both western and Indian philosophies have been very concerned with how we intuitively form universal concepts (like blackness or even numbers for instance) when all we have access to are objects that are similar or dissimilar in several ways. Plato, for instance, famously concluded the only reason we think something is beautiful is because ‘Beauty’ actually exists in the world of ideal forms! But this is a debate that is still ongoing. A question i try to pose with this artwork is- how would a machine understand a universal quality, given all it sees are examples. The images it produces tell us something about our own selves, refracted through the alien eye of the machine looking at the way we attach words to the world!

Here, i use Attn GAN to create the work, an AI model that uses a massive database of images with captions (like “facade of an old shop”) to learn verbal-visual connections. It is then trained to produce images for sentences it has never seen before. In particular, it is asked to produce images that correspond to sentences like “This is white” or “This is black” which elicit from it its visual representation of universal qualities like ‘old’, ‘young’, ‘black’, ‘white’ etc. Each column pair represents one such universal notion and and its opposite. The overall collection of images is reminiscent of
Beeple’s famous ‘Everydays: the first 5000 Days,’ bringing to the fore the aesthetic of multiplicity (often inherent to digital art, especially AI and generative art) here created through human-machine partnership.

This work was inspired by conversations with 64/1 (Karthik Kalyanaraman & Raghava KK).

Part human, part human-machine (hand)writing.

We are increasingly offloading a lot of our mental and subjective tasks to machines. In every such interaction, we effectively authorise the machine to momentarily substitute for our mind with its own intelligence, taking decisions and executing (its) plans on our behalf. We achieve an output and feel ownership of it, without explicitly knowing how we achieved it. We have become so accustomed to internalizing this substituted mind, that we do not even acknowledge its authorship in our tasks, let alone reflect on its influence in our everyday thoughts and actions. In this scope, we investigate how this relationship evolves, when the substitution leaks out of just the cognitive domain, and finds its way onto our physical body. For this, we create a handwriting system, where a machine moves our hand on a paper surface to write out its thoughts. How do we feel when our hand ‘mindlessly’ moves on the paper, but eventually writes something meaningful. What happens to our relationship with this other author of our everyday lives as it rises out from behind the surface onto the tips of our hand. How does our sense of ownership of the handwritten outcome vary, if at all? What’s the effect if our own thoughts and actions write all the content, half of it, quarter, none? Offloading how much control to the machine is enough? Also, a lot of entities and their intelligences are involved in writing this piece of paper- the user, the machine, the creator of the program, the various people whose text and thoughts for the part of the dataset- all that eventually makes your hand write something. This already happens in our everyday lives to some extent, often without us explicitly realising. However, once carried through a physical medium, does it make us more aware of the multitude of intelligences we inherit in our everyday, simple tasks?

How do we ultimately extend this experience to rethinking the balance of authorship and authorisation in our lives, especially as machines continue to grow in their intelligence?

Made in 2017 during a residency at X-Lab in Japan with Prof. Yasuaki Kakehi and Junichi Yamaoka.

For this series of works, i amass a large collection of landscape paintings from the WikiArt archive to train my AI on. The outputs created capture a dream like imagination of landscapes – there’s an immediate sense of recognition, but the physics and compositions of the landscapes deviate from the obvious, drawing one’s attention in. To highlight this uncanniness, i take these static outputs from the machine and animate the landscapes, adding subtle, realistic motion to these non-standard compositions, to push this blurring between digitally imagined and natural realities. These AI imagined and animated works of the otherwise traditional art theme of landscapes, speak of our ongoing transition into digitally native beings, where we imagine digital alternate realities of everyday existence and then bring them to life.

In this work, the artist creates a 3D world of a poppy field. Upon closer inspection, it’s revealed this world is entirely made up of tiny videos playing. These are advertisement videos most money has been spent on in India over the last 2 years, sourced from the Facebook API. We are losing our natural habitat to one of media today, constantly surrounded and increasingly immersed in it. Advertisements have been the mechanism of luring us into the addiction of free media, as technology giants continue to carry out a ‘data-grab’ akin to the land-grab of 19th century colonizers, as if it were simply there for the taking and profiteering of, creating a level of influence, addiction and displacement of the natural, that overtakes our entire sense of ‘reality’. Technology giants are using individuals’ data to fund ‘digital colonialism’, while addicting us to the idea of free media and mindless scrolling (much like poppy addiction).

This work presents a study of still-life aesthetics through the lens of artificial intelligence computer vision. Positing the question, can machines be taught aesthetics, here the artist trains a machine to look at thousands of still-life paintings, some in their entirety, and some in their details, to try and guide a machine to learn both composition and the painterly nuances of aesthetic. This work tries to start to teach AI the conceptual distinction between the compositional and the painterly. In any painting what is the relation of the part (as fetish) to the whole (as icon)? How can one teach a computer compositional structure and painterly texture?

Through this training, the machine is able to abstract out a relational sense of form, color, composition and produce outputs that resemble an uncanny likeness, yet the obvious departure to (real) life, very similar to this moment of digital transition that we are living in. For the video work, the artist uses this learning of the machine, across the icon and fetish, to interpolate from the fetish to the icon to the fetish, in a seamless morphing of varying detail levels of still life paintings. The work highlights the increasing digitization of natural ecology too- and the need for rapidly changing media. Still Life (the artistic subject of centuries) has become (Un)Still in today’s times.

A grid artwork is created, where for the central work, the AI develops a sense of form from studying examples of whole paintings in its dataset of European still lives of floral arrangements. In the surrounding works, the AI develops its aesthetics by only studying random details in the still life paintings, where i randomly zoom into the paintings and limit the region the AI trains on. I further create a distinction for these, by altering the training process for both, using a slower learning rate for the central piece (allowing larger time for the AI to learn details) and faster learning rate for the outer pieces (not giving the AI enough scope to pick up on details).

An experiment in understanding computer vision (and an attempt towards advancing the field of AI art), this work tries to start to teach AI the conceptual distinction between the compositional and the painterly. In any painting what is the relation of the part (as fetish) to the whole (as icon)? How can one teach a computer compositional structure and painterly texture? This work tries to start poking into these set of questions.