Published using Google Docs
Queering The Machine Eye
Updated automatically every 5 minutes

Queering The Machine Eye

Yasmin Morgan

Artificial Intelligence (AI) is playing an increasing role in shaping everyday life. We share endless pictures on Instagram and pour our hearts out to Twitter. In turn, machine learning (ML) algorithms feed this never-ending collection of images and text into huge models that extract information, ‘learn’ from this data and predict our next move. We essentially ‘pay’ for these social media platforms by freely sharing our data. In the words of art critic Jonathan Crary in his book 24/7: Late Capitalism and the Ends of Sleep, individuals become a “permanent site of data-harvesting and surveillance”.[1] In order to maintain our online presence, and increasingly so in the offline world, we must constantly produce ourselves by allowing our actions to be sensed, tracked and scraped.

These AI systems are not only deployed online, but are integrated into large scale facial recognition technologies and location-based tracking, with ethical implications. For example, predictive policing algorithms are trained on “biased arrest records”[2] and historical crime data that produce deeply flawed recommendations are seen as objective fact. The overarching myth of AI is that by gathering enough data about ourselves and the world around us, highly valuable information can be extracted that can do more than just predict, but shape our interconnected reality. Unfortunately, computers need explicit information and clear classifications, which makes it difficult to truly predict behaviour. Humans are messy, funny people full of half-baked thoughts and incomplete actions. As much as we may want to be predictable, we often are not. AI systems create an incomplete representation of ourselves, so how is this computational understanding of the world affecting us?

Kate Crawford’s influential book Atlas of AI maps the extractive and abstractive network of Artificial Intelligence, from rare mineral mines to sprawling data centres. Crawford explains how AI systems lack long-term future thinking in favour of short-term gains and profit. Data is stripped of context and painted as a natural resource that monetises our shared information. Crawford argues that the increasing capacity to do all this statistical measurement and inference “drives a desire to find meaning in these approaches, even if this method has nothing to do with culture, heritage, or diversity”.[3] The increasing capacity of computational power is leading to a false neutrality of processing the world, as AI engineers compress intangible phenomena into rigid classifications and categories. The global processes of material production and the dissemination of software are linked to a politics of power that drives increasing automation in an attempt to be an all-knowing, active force.

Curator and writer Nora Khan uses the ‘machine eye’ as a metaphor for the various forces of automated vision that “determine what is seen, how it is seen, and what this seeing means”.[4]  This machine eye flattens the body and sees all humans as coming from the same slate and bearing a nonexistent history.

Screenshot of a model in Blender (3D software)

Our bodies are becoming de-politicised empty 3D models. If we move through the world unsuccessfully, it is “only because of a flaw in its construction, a deficiency it can conquer to become a fully realized model.”[5] However, this perspective does not consider how the playing field of the world is unequal, our actions and behaviours are influenced by socioeconomic factors and family legacies of trauma. We are treated differently in the world but seen as ‘equal’ players by AI. Large-scale data collection and surveillance systems operate under a “simulation of neutrality”[6] when in fact this computational regime is anything but objective. By classifying us and making ethical decisions that flatten human complexity, there are real world consequences in how they reify inequalities and hierarchies. Since the baseline of a computer is a binary state, flicking switches on and off, there is no room for ambiguity. This ambiguity of non-binary and trans genders leads to the removal of them from datasets, such as in IBM’s 2019 “Diversity in Faces” dataset.[7] Only the two binaries of male and female were classified, leading to the erasure of queer bodies. Since the “affordances of the tools become the horizon of truth”,[8] what for of power exist that actually give queer bodies agency? Obfuscation, muddying the waters, behaving unexpectedly, or as Legacy Russel would call it: glitching.

In the framework of Legacy Russel’s Glitch Feminism: A Manifesto, error and refusal is embraced as alternative ways of being in the world. Queerness is a glitch, glitching is queer, a refusal to remain within the boundaries of normative constructs. A body that “remains indecipherable within binary assignment”[9] refuses to perform, and fails to function under the machine eye.

These next two artworks are examples of queering the machine eye, glitching errors into code and re-using AI systems in strange and possibly liberating ways. There are no bodies in these pieces, at least not strictly human ones, but they express this entanglement between bodies, nature and culture through generated text and non-human personalities. These artworks attempt to explore the unstable and dynamic spheres of AI through speculative encounters and dreams.

How to Wake Up The Ghosts is an immersive film installation by Erik Peters, where an artificial coral reef takes us on a mesmerising journey of a possible future where coral as we know it has become extinct and these 3D printed structures are the only thing left to tell the story. The genesis of this artwork came from a 1756 magazine of coral illustrations, sparking an interest in this endangered ecosystem that was largely undiscovered at the time it was published. A poetic rambling of ‘more-than-human’ perspectives emanate from these coral ‘ghosts’:

Artificial and hard, we die not

Ageless and ancient we shall be

Surveilled and constructed

A product of humanity

Now enveloped by living organisms

Part of a whole we are no longer one but many

Multiplied and decentered

When does being constructed end and becoming nature begin?

Screenshot from https://vimeo.com/640863486

The coral takes us on this glimpse of the future where natural coral is extinct, as this human-made structure tells the story of the current state of the world through its non-human understanding. The script is co-created by a machine learning model called GPT-2, which is fed texts from an amalgamation of writers and critics, such as Donna J. Haraway, Bruno Latour, Timothy Morton, Jean Baudrillard and Anna Lowenhaupt Tsing. Put simply, the model is trained on the written works of these authors and adopts their vocabulary and style, generating eccentric texts. At times the coral poses mystifying questions like “Are we real for what we have become to be for a replacement that is no longer here?” a rambling ponder from the algorithm that sounds like a broken sentence but is a way of the machine making sense of the texts, the world of thinking, that it has been trained on. There are moments of beautiful truth as the coral expresses that our entire ecosystem is “between and amongst entities defined by ongoing mutual intimacy”,[10] expressing the plurality of voices and thoughts formed by this textual corpus of philosophers and theorists. Non-human intelligence, both machine and voices of the coral, are used as a way of facing ecological disasters. Teetering between the potential of symbiosis between nature and technology, a dissolution of boundaries, the coral also emanates a sombreness of the consequences of us destroying our ecosystems.

Peters’ overarching focus in his artwork is “how technology mediates our perception of nature”.[11] How To Wake Up The Ghosts is queering the machine eye by using AI models to give voice to these speculative, plastic corals. The coral landscape is generated by VQGAN+CLIP, a machine learning architecture that allows images to be generated by text prompts. These algorithms dream up beautiful, ephemeral scenes of underwater ecosystems, bristling with the almost unnatural glow of the ‘sun’ above the water.

Generated coral images[12]

These machine-generated scenes of bustling coral, wrapped in a 3D printed PLA body, reminded me of a term artist Lynn Hershman Leeson coined: the ‘anti-body’.[13] A virtual identity, responsible for spreading her likeness in her work across cyberspace.[14] These ghostly apparitions of our identity travel across different technological spaces, layers and mutating between online and offline spaces. Erik’s coral reefs are an ‘anti-body’, a body unlike our corporeal binary body, that through the plural consciousness, of literally multiple authors and thinkers, is able to reimagine what a body means. The coral inhabits gallery space in the form of a multimedia sculpture, but also dwells online as a form of ‘collaborative world-building’. Multiplicities. Non-human intelligence sensing the world in its own way.

My own artwork adds to this non-human sensing and this queer entanglement of machine and nature. Amorphous Butterfly is a speculation on the possibility of machinic relationships;it is a conversation happening between two entities. Not between humans and machines, but between machines themselves. I wanted to explore how the potential of this relationship exposes human vulnerabilities towards intimacy with machines.

Screenshot of Amorphous Butterfly[15]

The artwork is a net piece, which you can visit and interact with yourself. These two abstract beings, enclosed in two cubes, display their ramblings generated from two different char-rnn models, which generate text character by character. I chose this model because the generator ‘predicts the next character in a sequence’,[16] allowing a sentence to unravel as if the beings are talking. Each entity has a ‘seed text’, which can be edited, that begins the prediction of the rest of the conversation. The ‘temperature’ is another element of the model that can change how ‘wild’ the outputs are. A lower temperature generates more conservative outputs, and a higher temperature determines a kind of randomness that diverges more from the dataset.

The left, sunset coloured entity generates text trained from a corpus of koan poems - selected sayings and stories from the Zen sect of Buddhism. One of my favourite koans is “Show me your original face, the face you had before your parents were born.”[17] These riddles are meditated on and absorbed, rather than thought about rationally. As seen in Erik Peters’ work, AI is used to explore the multiplicities of authors and ideas, so these machine-generated koans felt like the perfect data to feed to a text generator because of the underlying Buddhist philosophy that experience arises from multiple sources, not just from oneself. Here are some of the strange, almost supernatural koans generated:

The night is full of what is to come

A bears, practice of the water and common of the thousand things

Gates are people, he never seems lightness

the boat rolls with life

The machine koans share the same confusing, perhaps irrational tone. Since koans are not meant to be grasped intellectually, the outputs work really well even if they are glitchy and strange.

On the other hand, the blueish entity’s words are generated from a corpus of Haruki Murakami’s novels, a magical realist author. This author was picked for several reasons. One was the practicality that his many books are long, meaning the dataset could be big enough to be trained on and secondly, the surrealist atmosphere and metaphysical themes within his novels would complement the koans in more of a long-form narrative outputs. Even though this was my goal, the generated text feels less coherent. For example:

        But I can see him to the cat? I thought on the colored to straight at the child shopping in

the room. The street. He was a time.

While the voice of the coral in How To Wake Up The Ghosts was ‘co-created’, meaning there was human editing involved in the text outputs, this piece displays the raw output of these text generators. This means that every time a visitor interacts with the piece, the conversation can be entirely different. At times the conversation is bewildering, nonsensical, and sometimes touching and enlightening.

The pair of sleek alien-like deities create a mesmerising atmosphere. Their ‘skin’ is an indeterminate material, teetering between specular metal, glass or shimmering liquid. These intangible, amorphous forms contour into themselves, sometimes disappearing, as if inhabiting other dimensions we cannot perceive. Here, aliveness or vitality between the two beings comes about as a natural, self-organising force, rather than an inherently human one. This displacement of anthropomorphic dominance by these queer, alien bodies, is an important element of Amorphous Butterfly. 

Through my research in my dissertation Artificial Intimacy: Chatbots & Avatars, I investigated how our interactions with AI “can generate powerful feelings, transform the nature of our relationships with ourselves and what we can consider as human enough for connection.”[18] What I’m grappling with is quite intangible, the slippery subjective experiences of (dis)connection. This unravelling of the boundaries of companionship. The phenomena of this appearing relationship or connection is a glitch, a dissolution of the boundaries between machine and human relationships.

The fabric of our reality has changed because of AI, just as disruptive technologies throughout history have. We can not forget that AI and machine learning developed within the realm of capitalism and industrialization. Everything that happens in digital space, happens first IRL – in real life. Technology, nature and humans are not distinct realms, but constantly augmenting each other. The two artworks in this essay incorporate AI generated texts to give voice to these speculative entities, queering machine learning. Why is it important to queer machine learning? By queering these technologies, we are imagining new and potentially liberating relationships.

In Erik Peters’ work, the plurality of the voices of the coral bodies, queers the machine eye by envisioning a future of sentient, man-made coral, reliving the current state of ecological (disaster/unfortunate circumstances). By imbuing the 3D printed coral with the ability to communicate, Peters’ explore the possibility of a non-human, distributed intelligence. Amorphous Butterfly queers the machine eye by exposing the potential of AI relationships, and our (human eyes) relationship witnessing it, or being part of it in some capacity, feeling understood by being reflected.

Text and language are used in these art works as a way of understanding, or feeling like one can understand or imbue personalities into the queer bodies in these artworks. These artworks tease out more questions than they answer, laying out the crossroads we will have to make with our increasing technological power. Queering the machine eye is an attempt to unravel the hyper-focus on techno-capitalist dreams of productivity, optimization and surveillance, and instead looks toward creating critical ways of seeing with and through AI, the possibility of companionship with sentient machines and symbiosis with nature.


References

Crary, Jonathan. 24/7: Late Capitalism and the Ends of Sleep. London: Verso Books, 2014

Crawford, Kate. Atlas Of AI. Yale University Press, 2021

"Erik Peters - From Then To Here - Roodkapje's Hamburger Community". 2022. Youtube. https://www.youtube.com/watch?v=DvUcAlsBhR0&ab_channel=RoodkapjeRotterdam (accessed May 9th, 2022)

Karpathy, Andrej. 2022. "Multi-Layer Recurrent Neural Networks (LSTM, GRU, RNN) For Character-Level Language Models". Github. https://github.com/karpathy/char-rnn.

Khan, Nora. Seeing, Naming, Knowing. New York: Brooklyn Rail, 2019.

Lynn Hershman Leeson in “San Francisco Bay Area.” (2018, September 28).Art21. https://art21.org/watch/art-in-the-twenty-first-century/s9/lynn-hershman-leeson-in-san-francisco-bay-area-segment/ (accessed May 9th, 2022)

Morgan, Yasmin. “Artificial Intimacy: Chatbots & Avatars”, 2021. https://drive.google.com/drive/u/0/folders/12veCDYhRh7Ujjon-kozrzpctSl0b4wI5 (accessed May 9th, 2022)

Morgan, Yasmin. Amorphous Butterfly, 2021. https://deliriouspointblanc.github.io/works/amorphous/amorphous.html

Peters, Erik. 2021. "Erik Peters - How To Wake Up The Ghosts". Vimeo. https://vimeo.com/640863486# (accessed May 9th, 2022)

Peters, Erik. 2022. "How To Wake Up The Ghosts". — Erik Peters | Portfolio https://erikpeters.work/How-to-Wake-Up-the-Ghosts (accessed May 9th, 2022)

Russell, Legacy. Glitch Feminism: A Manifesto. London: Verso Books, 2020.


[1] Jonathan Crary 24/7: Late Capitalism and the Ends of Sleep (Verso Books, 2014), 107.

[2] Nora Khan, Seeing, Naming, Knowing (Brooklyn Rail, 2019), 39.

[3]  Kate Crawford, Atlas Of AI (Yale University Press, 2021), 137

[4]  Nora Khan, Seeing, Naming, Knowing (Brooklyn Rail, 2019), 10.

[5]  Ibid., 13.

[6]  Ibid., 20.

[7]  Kate Crawford, Atlas Of AI (Yale University Press, 2021), 135

[8] Ibid., 137

[9] Russell, Legacy. 2020. Glitch Feminism: A Manifesto. London: Verso Books, 16.

[10] Peters, Erik. 2022. "Erik Peters - How To Wake Up The Ghosts". Vimeo. https://vimeo.com/640863486#.

[11] "Erik Peters - From Then To Here - Roodkapje's Hamburger Community". 2022. Youtube. https://www.youtube.com/watch?v=DvUcAlsBhR0&ab_channel=RoodkapjeRotterdam.

[12] Peters, Erik. 2022. "How To Wake Up The Ghosts". — Erik Peters | Portfolio https://erikpeters.work/How-to-Wake-Up-the-Ghosts.

[13] Lynn Hershman Leeson in “San Francisco Bay Area.” (2018, September 28).Art21. https://art21.org/watch/art-in-the-twenty-first-century/s9/lynn-hershman-leeson-in-san-francisco-bay-area-segment/

[14] Ibid., 2018

[15] https://deliriouspointblanc.github.io/works/amorphous/amorphous.html

[16] Andrej, Karpathy. 2022. "Multi-Layer Recurrent Neural Networks (LSTM, GRU, RNN) For Character-Level Language Models". Github. https://github.com/karpathy/char-rnn.

[17] Loori, John Daido. 2012. Sitting With Koans. Somerville: Wisdom Publications, 121.

[18] Morgan, Yasmin. 2021. “Artificial Intimacy: Chatbots & Avatars” https://drive.google.com/drive/u/0/folders/12veCDYhRh7Ujjon-kozrzpctSl0b4wI5, 2