by Girinandini SinghFeb 19, 2021
Machines are coming! I know it’s old news. Artificial and machine intelligence making works of art is no more an idea that catches attention by itself. If art is an expression and a potent mode to question a norm, then the artist is the one whose thought or idea is put forward through their creation. And in the case of artificial intelligence (AI) art, the idea is translated into an algorithm that in-turn commands the computer system to make an imagery. It is as much as a sculptor feeding a laser-cutting machine with image they want cut on a metal sheet.
What is noteworthy is the plethora of possibilities that have opened up with AI art, which to a great extent, is impossible to achieve any other way. For instance, when a desired imagery fundamentally relies upon the analysis of a set of data as input, then there is a limitation on how much data a human or even manually fed computer can use.
Harshit Agrawal has been working with AI art since the inception of this genre in 2015. His ongoing solo presentation titled Exo-stential is hosted by Emami Art in Kolkata and curated by Myna Mukherjee. Mukherjee is the founder of Engendered, an organisation that seeks to drive cultural change and further the conversation of human rights using the strategic and transformative effect of art, cinema and performance; specially at the intersections of gender and sexuality in South Asia. Engendered commissioned Agrawal to produce Strange Genders, an AI art work in partnership with 64/1 (Karthik Kalyanaraman and Raghava KK). “Over many conversations we wondered if technology could be a collaborator ‘to create art’ that is a form of political or social currency, actively addressing cultural power structures to have a strategic and transformative effect. Karthik and Raghava dreamed up the project, executed it and involved Harshit for the AI portion. It was a wonderful collaboration,” says Mukherjee.
“AI Art allows to understand patterns from large amounts of data - becoming a means to probe into our internal biases individually and collectively,” adds Agrawal on the many doors this genre will open.
I speak to Agrawal on the sidelines of this first ever solo AI art exhibit in India.
Rahul Kumar (RK): Artificial intelligence (AI) art by itself is not a novel idea anymore. In the past four to six years it has not only gained traction, but also found acceptance within the formal art ecosystem. So, where do you think this trend is headed, what next?
Harshit Agrawal (HA): Having been working with AI art since its inception, I have become particularly interested in exploring themes of art that are uniquely made possible by how the AI art process works. The unique ability of it to parse thousands of samples of data and create a collective understanding from it, allows to probe into deep rooted themes of societal bias, of collective perception of themes like gender, offering a new lens (a data centric lens) to look at the long standing debate of philosophy about the existence of universals, to use now ubiquitous AI technologies like facial recognition, combining it with generative visual AI to become a vehicle for empathy for social injustice by transforming the sense of identity of individuals. Beyond the conceptual, I am interested in exploring the evolution of aesthetics too enabled by the process of teaching machines a visual language from given examples and varying the training process to produce different visual outcomes, like in Still Life: Icon and Fetish. All of this originates from the attempt to study and establish the poetics of machines, we are all immersed in AI and technology 24x7. We have to stop perceiving the machine as the external, it’s very much part of you - that’s the posthuman - where human and machine are no longer separate. For anything that’s this intimately involved in your life, it becomes critical to collectively engage in questions of how to shape it, can we trust it, can we express with it. If not, eight to ten large corporates and governments in the world will continue manipulating our experiences with AI they are building for their benefits.
RK: There are multiple formats and media along with various aesthetic approaches in your ongoing solo presentation titled Exo-Stential. Please talk about some of the key works and also the experimental stance for works Machinic Situatedness and Anatomy Lesson?
HA: Over seven years of my AI art practice, I have experimented with almost every aspect of the AI art process, both aspects that it shares with traditional art practices and those unique to it. The show brings together not only a diversity of themes, media (painting, sculpture, text, video, interactive media) and aesthetic approach (conceptual, sociological, the painterly) but also experiments with AI art-making itself. For instance, I consciously vary the ‘learning rate’ of the AI to produce new visual effects (e.g., in Still Life- Icon and Fetish, Machinic Situatedness, Anatomy Lesson) or achieve novel formal patterns by not relying on standard (Eurocentric) datasets (Artist as Community, Machinic Situatedness). Yet another conscious element of artistic manipulation is the degree of human involvement both in the production and the reception of the artwork (Tandem, (author)rise, Artist as Community).
In Still Life: Icon and Fetish, two AIs are trained from the same underlying dataset of European still lives flower paintings. However, both the datasets and training process are treated very differently to generate visually very different results.
The first AI develops a sense of form from studying examples of whole paintings in its collection of European still lives of floral arrangements. The second AI develops its aesthetics by only studying random details in the still life paintings it has access to. For the first AI, I keep the learning rate (the speed with which the machine forms its understanding of the dataset of images) very low. The machine thus learns minute details of the composition and is able to generate detailed realistic outputs. For the second AI, the learning rate is kept fast since the attempt is for the machine to learn abstracted visuals from the dataset - the painterly as opposed to the compositional of the first. Thus, experimenting with both treatments of the dataset and the AI learning process, a new study in aesthetics of icon and fetish emerges. The work is an experiment in understanding computer vision (and hopefully towards advancing the field of AI art), starting 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 makes an important headway in answer to this formidable set of questions.
RK: Why do you believe that AI art is uniquely positioned to be the vehicle of engagement with social, cultural and ethical issues, and acts as a conduit to explore our internal biases at an individual and societal level?
HA: For the first time, AI art allows to understand patterns from large amounts of data - becoming a means to probe into our internal biases individually and collectively. When we ask people to draw their visual representations of gender - people draw a male and female form separately. However, when thousands of such drawings are passed through an AI, it is able to generate new drawings that it classifies as a percentage of female - based on the original drawings that people draw. When people see this final outcome of percentage females starting from 100 per cent female to zero per cent female - their internal (possibly binary) perception of gender is questioned.
RK: In continuation, the work Masked Reality asks for viewers to participate to form the image. Kathakali and Theyyam performance props are projected on the face of the viewer. How is the work commenting on social discrimination through facial recognition and racial profiling?
HA: In Masked Reality, viewers stand in front of a digital screen with a webcam attached to it. The webcam, using an AI algorithm of facial recognition, recognises the facial contours of the viewers in front of it. Another AI generative algorithms called conditional generative adversarial network, generates the face of a Theyyam and Kathakali dancer conditioned on the facial structure and features of the viewers. Therefore, they see themselves as these two dance forms simultaneously juxtaposed next to each other. Kathakali is a performance art deeply informed by Sanskrit aesthetics and epic, patronised by royal families and “sattvic” temples (where typically the scheduled castes had no entry); Theyyam, a deity possession ritual, is locally varied and participated in typically by the lower castes. Now, aided through artificial intelligence, the viewers see themselves as both these forms simultaneously. Facial recognition is otherwise used to perpetuate biases further in terms of racial profiling and criminality prediction. When viewed from an artistic lens, can that same technology be used as a vehicle for empathy for social frameworks of caste?
RK: Lastly, emotion is an important aspect that machines are yet to learn or express. While in your effort for Still Life: Icon and Fetish you have successfully varied the speed and process of learning by the machine (and therefore focussing on details vs spontaneity respectively), do you think machine can alter its own pattern at some point, and therefore have its own expression?
HA: One aspect important to make note of is there is no inherent sense of agency or emotion in the machine. The artist is very much the human and any attachments of emotional sentiments that the works evoke are a result of how the human artist has created and represented the work. Having said that, what is fascinating with AI art for the human artist to engage with is - even when machines look at human-labelled human-curated objects (datasets), the generalisations they form (or the patterns they learn from it) are profoundly alien. As the human artist, you begin with a concept of a work, but your relation with the tool (here AI), evolves in a way to leave room for influence by the estrangement of the data the machine is able to produce from your inputs. That space for direction and accepting influence in a conscious manner, is what is fascinating for me as an artist, and I believe how our relation with technology is evolving to be.