Poetics of Garbage will be a single-screen film shot in the streets and hinterlands of Delhi that explores how single-use plastic designed into a performance costume is rationalized by a machine to produce language.
Image Source: greenpeace.org
The project will reflexively produce a commentary on the human-garbage relationship. Here, waste is considered an opportunity, an untapped resource waiting to be harnessed. Some forward-looking fashion brands have been designing with waste materials, including Spanish brand Ecoalf who aim to create clothes with 100% recycled materials including discarded fishing nets, post-consumer plastic bottles, worn-out tyres, and used coffee grinds. The India based brand, Doodlage, source fabrics that have been left over or discarded by the large manufacturers, which account for “wastage” in export terms. Veena Sahajwalla has pioneered the world’s first ‘green steel’ technology, which uses old car tyres in the steel making process instead of coking coal. She is a pioneer in producing a new generation of green materials, products and resources made entirely, or primarily, from waste. In architecture, recycled and reused materials are slowly becoming attractive alternatives to conventional building materials. Manav Sadhna, an activity center in Ahmedabad is built from multiple recycled materials, sourced from municipal and domestic waste. The House of Mixed Hues in Mumbai utilises recycled doors and windows from demolished houses in the city, and courtyard walls are built from the ‘waste’ generated in stone cutting yards. Contrary to the increasingly outmoded views of the linear materials economy, we can begin to appreciate how disposed material can be redesignated with value: waste can take form as fashion, architecture and energy. In Poetics of Garbage, spent things are given cultural agency and transformed into ‘writing machines’.
Here, waste is considered an opportunity, an untapped resource waiting to be harnessed.
As part of our project, the accumulation and sanitizing of plastic waste over the previous weeks is understood as a ritual—a metaphor for the burden and focused attention on the management of materials. This discarded stuff, reclaimed from the realm of the ‘useless’, has motivated a performance action in Delhi. The action involved the transportation of the waste material by wearing, carrying and dragging it across various locations.
These ‘manipulations’ of the waste have been filmed and are now being translated via computer vision and object recognition software into text, which will become part of a larger language generation process. In a way, this is opposite to the concept of ‘performativity’ by the philosopher of language, John L. Austin, who defines the specific capacity of speech and communication to act or to consummate an action. In Poetics of Garbage, the action accomplishes the ‘utterance’.
The performance on screen is similar to a ritualistic act of pilgrimage, a journey often into the unknown, where a person goes in search of new or expanded meaning. We imagine this sense of altered meaning being achieved through the actions of the body and its interplay between waste material and predictive technology.
Sacred River—Yamuna Ghat Delhi Kalindi Kunj
The Yamuna river, a primary source of water for Delhi, is becoming increasingly polluted with high levels of ammonia, an indicator of the inability of the river water to support aquatic life. As a potent symbol of the collision between heavy industry and spiritual belief systems, Suresh K Pandey’s photographs show the Yamuna not only foaming with ammonia but full of broken idols of Gods and Goddesses.
Images from Suresh K Pandey’s Instagram
This particular site is a traditional place to seek spiritual atonement by immersing idols and bathing in its waters, a source of many waterborne diseases and skin infections. We became interested in this site as the final destination and culmination of the performance action, and in anticipation, a costume created from recycled biomedical IV bottles was fabricated by a tailor in Delhi.
A sense of altered meaning being achieved through the actions of the body and its interplay between waste material and predictive technology.
In Poetics of Garbage spent things are given cultural agency and transformed into writing machines
Schematic of the project process
Text labels produced by object detection are incorporated into language prompts that are consequently ‘autocompleted’ by a computer language model (GPT-3) into a series of statements, a form of commentary that alludes to the picture in surprising and poetic ways.
Contemporary Chance Operations
The reality of artificial intelligence (AI) is that many algorithms labeled as such are, in fact, imperfect models that are at risk of making erroneous inferences. Algorithms employed to identify objects in images are contingent on the limited ‘worldview’ of a dataset. In other words, the software won’t be able to identify a cactus if it hasn’t already been trained on images of cacti. But it might falsely identify a cactus as a toothbrush because it looks a bit like a toothbrush at a particular angle, or lighting condition. In Poetics of Garbage, when an ‘erroneous inference‘ is made, it is nevertheless included in the formation of a text prompt—alongside any accurate ‘prediction’—to stimulate a language model. This is understood as a new form of ‘chance operation’. Uncertainty is introduced into the artistic procedure in a new way, making links to a long history of the accidental in art.
Never before have objects, the body and its behavior, been such an explicit interface. AI technology has a stark political reality that continues to be critiqued by artists concerned with its social implications, for example through the brilliant work of Crawford and Paglen (1) and Crawford and Jolar (2). Artists can also critically engage with its erroneous inferences, by creatively harnessing its failures as a new form of chance operation. Marcel Duchamp explored the aesthetic implications of randomly dropping pieces of thread onto the floor in 3 Standard Stoppages—now artists can explore the shortcomings of software as a new method of inviting chance.
Marcel Duchamp ‘3 Standard Stoppages’ (1913-14)
The first few weeks of the commission were spent developing a software pipeline to process the video footage. This coincided with the proliferation of generative text-to-image and text-to-video software that is understood to be pushing AI into a new phase. Instead of looking at the visuals from models like DALL-E 2, Midjourney, and Stable Diffusion as fixed and isolated objects, perhaps it’s interesting to consider how these tools might operate as part of a larger creative process. At the centre of Poetics of Garbage is an interest in the prompt as a way of interacting with the generative technology that’s fast emerging. There’s also an appreciation of a conceptual link to the history of Instruction Art from the 1960s avant-garde onwards. Instruction Art is defined as a set of instructions created by the artist for the spectator or performer to follow, or as a score for the owner to reproduce (as in Fluxus Event Scores, or Sol LeWitt’s directions for his giant wall drawings). In AI/machine learning enabled art, the instruction can be provided as a way of stimulating or prompting further action by the software to generate something—images and text. In our case, as collaborating artists researching creativity in the context of environmental crisis, we are working with waste material as a form of ‘instruction’—something that can be redesignated with value—as a constituent part of a prompt to generate a text output.
Algorithms employed to identify objects in images are contingent on the limited ‘worldview’ of a dataset. In other words, the software won’t be able to identify a cactus if it hasn’t already been trained on images of cacti. But it might falsely identify a cactus as a toothbrush because it looks a bit like a toothbrush at a particular angle, or lighting condition.
Currently we are in discussion over the best way to represent the language model’s output in an Indian language. This will be spoken as a voice over (VO) and soundtracked onto the film. We will accompany the VO with English subtitles.
Before the Delhi performance action in-January 2023 we had to imagine possible outcomes from object detection in order to explore the creation of the text prompts. In the examples below, speculated text labels from object detection were dropped into contextual sentences as a way of stimulating the language model.
Prompt: A woman is walking in the street with a costume made from cell phones, cups, bottles, a person, fire hydrants and wine glasses. AI Language Model Response: This woman is likely participating in a protest or street performance that is calling attention to the dangers of excessive consumerism and wastefulness. The costume, made entirely out of trash, is meant to shock onlookers and make them think twice about their own habits. The message behind this costume is that we need to be more mindful about the things we use and throw away—everything has value, even if it’s just in the form of scrap metal or recycled plastic.
Prompt: Robert Frost writes a dark poem about a machine looking at a woman in the street. The machine thinks she has fire hydrants, bottles, cups, wine glasses, and cell phones attached to her.
AI Language Model Response: I saw a woman in the street with fire hydrants, bottles, cups, wine glasses, and cell phones attached to her. She was a walking disaster and I couldn’t look away. I wanted to help her, but I was just a machine and all I could do was watch as she stumbled and fell into the darkness of the night.
Prompt: A machine mistakes a woman’s body for cell phones, cups, bottles, people, fire hydrants and wine glasses.
AI Language Model Response: In one particularly alarming incident, the machine, which was supposed to be sorting recycling, began violently shaking the woman’s body in an attempt to separate the different materials. The woman was fortunately unharmed, but the incident highlights the potential dangers of these machines.
Approaching the Prompt
A significant effort has been made in sourcing the plastic waste (with assistance from Toxics Link). This material was sanitized and made safe to handle and manipulate into costume elements to be worn as part of the performance. Some of this studio activity has been used as a video source to help create an automatic method of translating the object detection into verbal language.
We intend to play with distinctions between automation and augmentation
We are also exploring another approach that is less automatic, and more curated. The curation and editing of outputs are becoming generally accepted amongst artists working with language models. This attitude is expressed by K. Allado-McDowell talking about their GPT-3 generated novel “Amor Cringe” that perhaps signals a move away from an extremely systematic machine aesthetic and towards something more freeform.
In our case, we imagine eventually combining both approaches in the making of our film in an effort to help articulate the process, and to play with distinctions between automation and augmentation—automation at the risk of creating ‘word salad’ instead of coherent, human sounding language, and augmentation that combines new tools with human intuition and spontaneity allowing a wider creative net to be cast farther afield.
The images above and below are screen grabs from a text generating experiment that used a ‘cloak’ made from plastic bottles as a way of stimulating the object detection algorithm and language model.
- Kate Crawford and Trevor Paglen. (2019) Excavating AI. Available at: https://excavating.ai
Kate Crawford and Vladan Joler. (2018) Anatomy of an AI System. Available at: https://anatomyof.ai/img/ai-anatomy-map.pdf