the Unknown Person: Post-Colonial Fictioning, Personal Stories and Surveillance

the Unknown Person connects the artist's family history to Britain's postcolonial “fictioning.” The project interrogates the gaze of surveillance and social control systems to explore the fiction of the self, data and liminal spaces of the City of London. The final output of this research is a video documentary that employs machine learning processes and facial recognition techniques to generate visuals to reveal the aesthetic value of a neural network. The project culminated as an installation of multiple screens mounted on a scaffolding structure.

In 1949, my grandfather left his family and joined the Malayan Communist Party (MCP) in a guerrilla armed struggle against the British-Malaya colonial government. Three years into jungle warfare, the British ambushed and killed the MCP's chief propagandist and two of his bodyguards. My grandfather was one of the bodyguards. The colonial authority publicly displayed their corpses in the town square to instill fear in the masses. My grandmother and her five children who witnessed the scene from afar had to suppress their tears. Following that incident, the British began to hound and persecute the family members of the Communist fighters. They interrogated my grandmother at gunpoint about her connection to my grandfather. An admission would lead to retribution on the entire family. She refused to cooperate and denied that she knew of such a person. Soon after, she fled the village with her children, and they were declared fugitives of the law.
In the film the Unknown Person, I connected Britain's post-colonial history to my own violent family history. The work compared the surveillance and social control systems of today to the rebellion that my grandparents lived through. My grandfather fell to counterinsurgency intelligence. My grandmother hid a single data point to flee military surveillance and saved the family. The modern-day mass surveillance that could be weaponized by artificial intelligence (AI) is able to track, collect and control infinite data points. Could my grandmother have survived today's all-knowing and all-seeing dragnet? I situated myself in the non-spaces within the City of London and used fictioning as a method to explore how my personal fiction might be woven into Britain's through these liminal zones. Playing these tensions off one another, I have co-opted machine learning-based techniques similar to surveillance and facial recognition techniques to bring these stories to life.

Fictioning as a Method
In their book Fictioning, David Burrows and Simon O'Sullivan categorize myth-functions for contemporary art and philosophy as mythopoesis, myth-science and mythotechnesis. Mythopoesis (performance-fictioning) speculates on the production of new worlds, people and communities to come. Mythotechnesis (machine-fictioning) looks at existing and future influences of machines on our life [1]. My fictioning practice combines mythopoesis and mythotechnesis to blur the boundary between my personal fiction and postcolonial reality with the influence of technology.
Burrows and O'Sullivan write, "The concept of mythopoesis involves a disruption of a more dominant fiction of the self-talks about the multiplicity that would deterritorialize our usual identification (or speaks to the potential self we also are) while addressing us as part of a wider collective-as potentially part of a community to come" [2].
Similarly, I filmed myself listening to relatives and family members back home retell the stories of my grandparents. O'Sullivan summarized mythopoesis as the art of "calling forth the something in us that isn't us." In the same view, my presence filled the physical space with anecdotes (data) that flowed through me. This mode of mythmaking had an agency that brought forth the lamentations of my ancestors into the present. The tension drawn from an interplay of small and large fictions resonated with the definition of a "guerrilla" as a "small member of a small group that fights against larger forces" [3]. Jean-François Lyotard used the term petit récit [4], or "little narratives," as a rejection of the grand narratives (metanarratives) that dominate our world. The little narratives of my family are thus nested within the City's. In other words, the insertion of one time-space within another has fragmented the continuity of a postcolonial metanarrative. Gilles Deleuze's concept of summoning a "people-yet-to-come" [5] bears relevance here. Considering how colonialism has shaped my family history, I suggested that I was one of these "invented people"-a minor character who has arrived at the present.

The City of London
The City of London-referred to by locals as the City-was historically and economically intertwined with the colonial ambitions of the British Empire. In order to understand the entanglement of the City with my personal fiction, the background leading to the Malayan Emergency needs to be explained. The Malayan economy at that time relied on the prices of rubber and tin. The prices depended on the movement of the world market facilitated on the trading floors in the City's exchange. As a result, there was widespread poverty and starvation in Malaya (especially among the ethnic Chinese). A striking quote from the official record [6] published by the London War Office stated " . . . the main cause of the recent disturbances is the lack of rice." In fact, the conflict was only labeled an "emergency" because colonial estate owners, whose assets were under threat by guerrillas, would not be compensated by the City's insurers if it had been labeled a "war" [7]. One could speculate that my grandfather fought for basic needs as much as he did for revolutionary ideals. In my previous essay, I proposed that the City of London is a fiction-generating machine [8] with tentacles extracting fiction of history, community and nationhood from Malaya, hence inextricably linking my personal story to the place.

The Non-Spaces of the City
The "hidden" parts of the City of London were the backdrop of the film as a conceptual decision to avoid filming in a heavily scrutinized location. The work assumed ubiquitous surveillance cameras embedded into the fabric of the City to be a normative feature in our urban environment. Marc Augé described non-spaces as insignificant spaces of "transience," perceived subjectively to be "crossroads of human relations" forming our super modern world [9]. In the film, I explored corridors, transit passages, nooks and crannies of the City. The historical importance of these locations was subdued in order for mythopoesis to happen. In Augé's text, he described "the space of non-place creates neither singular identity nor relations; only solitude and similitude" [10]. When entering the space of non-place, the individual can enjoy a "passive joy of identity-loss." This luxury was futile under the gaze of surveillance that penetrated any liminality (Fig. 1).

Technical Research
The machine learning technology that I've implemented isn't particularly novel. I first came across an experiment by Chris Harris where he uses a neural network to detect and remove vehicles from the street [11]. Another artist, Michail Rybakov, had done tests using machine learning to delete bodies [12]. He combined two open-sourced GitHub repositories-Mask RCNN for body recognition and Generative Image Inpainting with Contextual Attention-to fill the gap back in. To emulate AI surveillance systems, I experimented with object detection models such as YOLO and Mask RCNN following the GitHub repository of Mark Jay [13].

Machine Learning
For the machine learning model, I first experimented with installing the Mask RCNN and DeepFill Inpainting GitHub repositories. The tutorial series by Mark Jay was an immense help in getting started with some early test shots with still images and videos. Although I successfully got the model up and running on the local graphics processing unit (GPU), it struggled to run smoothly on my machine because the model outputs video at one frame per second, which wasn't visually impressive. I relied heavily on the remote GPU and model libraries of RunwayML (Fig. 2). The platform provided more models and flexibility for the workflow of producing image sequence files. I used the DeepLab model by Gene Kogan to extract semantic maps from objects in images. The model was able to capture multiple objects at various scales based on COCO-Stuff 10K/164K and PASCAL VOC 2012 datasets. I only needed inference from a single image-that of a Person. The model saved a black-and-white mask out of the image, which I then used as a source for segmentation input for the image inpainting task. For that, I used the Deep Fill model by feeding in the black-and-white images generated from DeepLab as segmentation input to fill the missing regions of the image (footage of myself ). The model comes pre-trained on the Places2 dataset by MIT, which contains a lot of images of outdoor places that would be able to bias the filling in of the "Person's" image toward more natural scenes of its surroundings.
One of the challenges with the image inpainting algorithm was the low-resolution output. I needed the output to be relatively high-resolution as it will be displayed on video monitors. The dataset came pre-trained with images of resolution 256 × 256. On top of that, the image resolution deteriorated with training time. The pixelated 360p video when broadcasted on video monitors was a concern. The solution was to run an upscaling-pass from 360p back to 720p that was added back into the original image information and mask. The inpainting redacted me with imprecision, but the imperfection endows the image with potency. A perfect rendering would mean there's nothing to see. It posed the question: Am I a figure hiding from the gaze or has the camera penetrated even the liminal spaces? The ambiguity brought to life the interplay between the self and the story while revealing the liminality of the space.

Face Interaction
For the interactive component of the piece, I built a face-detection algorithm in openFrameworks and modified the ofxFaceTracker2 add-on by stripping away the wireframe lines on the face-tracking mask and keeping the outline (Fig. 3). The add-on was based on dlib's face detection API, which generates various facial features (eyes, nose, mouth, etc.). My approach was to emulate the aesthetic of a facial dataset used in a surveillance face-detection algorithm. The algorithm was not trained on any dataset to classify objects it sees. The detection operated through a consumer Webcam positioned discreetly on the structure.
My purpose for adding the interaction component to the piece was twofold. Firstly, the Webcam feed running the face-detection model represented the City's surveillance apparatus. Secondly, it acted as an instrument for fictioning the subject of surveillance, in this case, the audience, into my personal story. On a more pragmatic level, its inclusion was built with the scaffolding structure in mind to entice the viewer to walk around the structure and experience the three dimensionalities of the piece (Fig. 3).
The detection frame was labeled "unknown person," the default label when the algorithm failed to classify its subject, suggestive of the inherent (un)bias of the facial recognition algorithm. The Unknown Person here clearly referred to my grandfather, whose identity remains cloaked in anonymity. He remained anonymous when his body was buried in an unknown location. However, his existence was redacted when my grandmother denied knowing him to the interrogating soldiers.

Installation
The installation (Fig. 4) ran on six monitors and four Raspberry Pi's. The Raspberry Pi's were configured to play media upon booting and were controlled via SSH with a laptop running the face-tracking code on openFameworks. There was a wide-lens Webcam for camera input. I relied on this video looper code [14] that turns a Pi into a dedicated video playback tool. It called on OMXplayer, the default video player for Pi's, and can play most videos encoded with the H.264 video codec and in an .mp4 format. This was the best solution for my needs because OMXplayer plays videos up to a resolution of 1080p.
The piece's industrial scaffolding structure measures 8 × 7 ft high (Fig. 4). I attached six stripped-down monitors to the bar, giving a cold industrial aesthetic to the work. The imposing scale of the structure represented the towering meta-fiction of the City-a latticework for my family fiction to hang on.

Conclusion
In 1960, the British utterly defeated the remaining Communist guerrillas, ending the Malayan Emergency. Each unclaimed body of a Communist fighter was officially classified as an "unknown person." We do not know where my grandfather is buried to this day. the Unknown Person demonstrated the potential for a new medium and method of personal storytelling, defined by machine learning-based surveillance aesthetics. I wanted to emulate the emotional connection to the brutal scenes my grandparents suffered during colonial rule, and through that process, unpack and share a complicated part of my heritage. To those ends, the work succeeded. The audience resonated with my family story the most. The power of narrative as a medium held all the computational processes together; weaving the story around the physical structure created a shared spatial experience among the audience as they interacted with the different parts of it. The scale of the installation serves as a reminder of the idea that grand historical narratives are often built on momentous stories of the individual.