Keywords

Drones kill. More precisely: drones in movies, TV series, novels, video games and artworks often kill humans. A scene in the Norwegian TV series Beforeigners (2019) is a typical example: a man, Navn, runs through the woods and is shown by a conventional photorealistic camera that films him from about the height of another human. We hear the sound of an approaching drone as the scene cuts to footage of the treetops shot by the drone. Navn looks up. The camera cuts to show the drone as Navn sees it, high above the trees, the sound of its motor close by now. The camera angle switches to the drone’s view again, but this time the view isn’t photorealistic. Now the screen is a monochrome green like a computer display. Navn is shown in green in the centre of the screen, surrounded by concentric circles that move until they settle upon him, marking him as a target. The camera cuts back to a photorealistic view of Navn seen from the side. The sound of shots punctuates the whirring motor of the drone as bullets pepper Navn’s body. His body shakes violently and falls to the ground. The final images in the episode return us to the drone’s perspective, showing Navn’s dead body circled by the rings of a digital bull’s-eye.

This chapter builds upon an analysis of 72 games, artworks and stories that feature drones to understand how the relationship between drones and humans is imagined in contemporary culture. Beforeigners is one of these 72 works, and I will use a close reading of this scene in combination with a distant reading of the larger dataset to explore this drone imaginary. The 72 works involving drones are part of a larger dataset of 500 artworks, games and narratives portraying machine vision technologies (Rettberg et al., 2022a). By machine vision, I mean the “registration, analysis and representation of visual information by machines and algorithms” (Rettberg, 2023, p. 3), for example, facial recognition, night vision cameras, surveillance cameras, AI-generated images and drones. When I wrote “drones kill” at the start of this chapter, I based that on comparing actions taken by drones and other technologies in these 500 works. Not only are drones far more likely to be killing, they are also the only technology in our sample to be portrayed as shooting and attacking—but also as attacked.

Beforeigners is not primarily about technology or about drones, and at first glance, the scene where the drone kills Navn seems to fit the picture of hostile drones killing people. A closer look shows a more ambivalent picture, though, both at the level of this particular scene and when we look at the larger dataset. The drone is not as all-powerful as it might at first seem, and neither is the human who controls it. My goal in this chapter is to untangle the relationship between humans and drones as it is imagined in these works.

Fiction, art and popular culture are key sites for public debate about new technologies (Rettberg, 2023). Unlike the debate we see in political deliberations or in newspapers and social media, fiction and art explore possible futures. By showing us what can go wrong and what might be possible, fiction helps us develop and negotiate possible futures as a society. Mary Shelley’s Frankenstein, George Orwell’s Nineteen Eighty-Four and Margaret Atwood’s The Handmaid’s Tale are examples of well-known stories that have become standard references in debates about biotechnology, surveillance and technology-assisted discrimination. By analysing 500 works we wanted to look beyond the most familiar works to explore the cumulative imaginaries that are expressed across the broader field of cultural expression.

The Machine Vision Situation

Navn being killed by the drone is a “machine vision situation” that is documented in the Database of Machine Vision in Art, Games and Narratives. My colleagues and I developed the database as part of a larger project to analyse how machine vision technologies are portrayed in art, literature and popular culture with a focus on the relationship between humans and technologies. Over the course of three years my collaborators and I watched, played, read and explored 500 movies, novels, TV series, artworks and video games that featured machine vision technologies in various ways. We identified 874 machine vision situations involving technologies like facial recognition, drones and deepfakes. Seventy-two works in the dataset involve drones: 6 artworks, 18 video games and 48 narratives (movies, TV series and novels), all published between 1998 and 2021. A data paper documenting the full dataset is available (Rettberg et al., 2022a), and the dataset itself is also openly accessible as a set of CSV files (Rettberg et al., 2022b). Drones take part in 259 individual actions such as killing or recording.

We defined a machine vision situation as “a moment at which a machine vision technology is seen to make a difference to the course of events” (Gunderson et al., 2024). We drew upon Lauren Berlant’s concept of “a situation” as a moment of disturbance in something that was ordinary, as “a state of things in which something that will perhaps matter is unfolding amid the usual activity of life” (Berlant, 2011, p. 5). In the situation from Beforeigners, the drone’s arrival is such a moment of disturbance: it is a moment where the technology comes to matter.

Within each of the 500 works we identified situations like the drone killing Navn. We developed an analytical model to interpret each of these machine vision situations in a structured way that would allow us to compare what each actor in the situation is doing. Actors could be human or nonhuman, and we identified three types of actors: characters, technologies or entities. The next step in the analytical model is to assign active or passive verbs to each actor in the situation (Gunderson et al., 2024). In this situation, the drone is locating and killing, and Navn is surprised and killed. It is important to note that these verbs, like all the actions in the dataset, are interpretations made by our research team: none of them are automatically extracted from the movies or novels themselves. They are also by necessity quite reductive. The statements “a drone is locating and killing” and “Navn is surprised and killed” tell you far less about the scene than my description at the beginning of the chapter. However, this simplification gives us structured data that allows us to compare hundreds of different machine vision situations across many works of fiction and art. I will return to a richer analysis of the Beforeigners situation later in this chapter, but first let’s look at the big picture of how drones are represented through data analysis.

The overall dataset contains art, games and stories involving many different kinds of machine vision technology, and the research team assigned an overall sentiment for the work describing the attitude expressed to the technology. For instance, the sentiment towards technology in George Orwell’s Nineteen eighty-four would be hostile, because the novel strongly shows how oppressive surveillance can be. In the video game Duke Nukem 3D on the other hand, machine vision technologies are portrayed as helpful but also flawed. The player can use night vision goggles, which make it easier to see enemies. The player can also create holograms so enemies will attack the hologram instead of the Duke, but they do not work very well and enemies are rarely fooled. So, machine vision technologies in this work are framed as helpful but also flawed.

Unsurprisingly the 500 works in the dataset portray many different attitudes towards machine vision technology, but overall, the most common sentiment is that machine vision technologies are helpful—although hostile is the second most common sentiment. Figure 2.1 shows how the 500 works express different attitudes towards different technologies. Only four technologies are portrayed as more hostile than helpful: drones, facial recognition, surveillance cameras and machine learning. In contrast the technologies that are portrayed as most positive are holograms, augmented reality, non-visual spectrum (like night vision goggles) and general AI (we used this to refer to the fictional AIs in many works that are often sentient or at least very intelligent).

Fig. 2.1
9 stacked bar graphs plot six attitudes toward technologies such as holograms, augmented reality, and drones. Holograms, augmented reality, surveillance cameras, A I, object recognition, and nonvisible spectrum are highly helpful while others are hostile.

The overall attitude towards machine vision technologies in 500 words, sorted by which technologies are represented in the works

Full size image

Games are particularly likely to portray technologies as helpful—often the player can use machine vision technologies to improve their skills or access information that would otherwise be inaccessible. However, games also often portray machine vision technologies as flawed or untrustworthy, as with the holograms that do not quite work in Duke Nukem. Ragnhild Solberg’s article “‘Too Easy’ or ‘Too Much’?” explores this in an analysis of the games Call of Duty 3 and Cyberpunk 2077 (Solberg, 2022a).

Interestingly, none of the six drone artworks in our dataset portray drones as helpful. Visual technologies are used and explored a lot in contemporary digital art, and artworks often interrogate problems and flaws in the technologies. For example, Linda Kronman argues that “art hacks” function as a kind of artistic auditing of machine vision technologies, much as researchers “audit” AI systems for AI bias (Kronman, 2023). A drone in a game is probably something you can control, so helpful, or something that is attacking you, so hostile. In an artwork, a drone is likely to be something to be critiqued.

So far this supports my claim in the first sentence of this chapter: drones kill. But clearly there are exceptions: drones are also often helpful, especially in games, and they can be fun as well. Looking at the topics that are most common in works about drones might provide some clues: artworks, games and stories featuring drones tend to deal with topics like surveillance, dystopias, AI, conflict and physical violence. That seems to support the simple “drones kill” framing. However, companionship and free will are also common topics. Companionship and free will are topics that connect to ideas about human agency—do we control drones? Are we complicit in their killing?

To better understand how drones are represented across the dataset, I identified the most frequent actions taken by drones across all situations they appear in. Drones are significantly more likely to record, kill, transmit, target, fly and be controlled than other machine vision. They also are the only technology in the dataset that is shooting, attacking and attacked.

Table 2.1 shows the ten most common actions taken by drones. The columns after each action show the distribution of each action for drones in comparison to the whole dataset, that is, all the machine vision technologies we looked at. So, to start at the top of the list, the most common action taken by drones is to be recording. Around 6.2% of all actions taken by drones are recording, whereas only 2.2% of actions taken by all the machine vision technologies are recording.

Table 2.1 Most frequent actions taken by drones
Full size table

These verbs suggest drones are represented as tools that humans control. However, a more qualitative reading of individual situations, such as Navn’s murder in Beforeigners, shows a more complex picture.

Navn, the Drone and the Viewer

In Beforeigners there are two sets of relationships: one between Navn and the drone that kills him, and one between the viewer and the drone. The first is diegetic, or internal to the fictional world, and the second is extradiegetic, extending outside of the fiction to the viewer sitting and watching the show. The dataset focuses on the diegetic representation of machine vision, but a close reading of each situation finds many cases where the viewer (or player or reader) is involved or called upon in different ways.

Navn looks up at the drone with a bemused expression before he is violently shot and falls, dead, to the forest floor. He is surprised and killed, and the drone is locating and killing. When the drone’s eye view shifts to the monochrome green display with concentric circles targeting Navn, we snap out of the photorealistic conventions of cinema. The videogame-like display is a liminal space between the diegetic and the extradiegetic levels. We could interpret it as diegetic: perhaps we are now seeing the action from the perspective of the drone operator. But the visual aesthetics of the screen are those of a video game or a control panel, not of a TV series. The visual interface puts the viewer in the position of the drone operator, but we are not given the opportunity to control the drone.

This is the first time in the series that the image breaks with the realism of television, and it signals a shift in agency: the technology and its operator take control of the situation. Up until this point Navn is portrayed as a powerful character. Beforeigners is a TV series premised on the idea that people from the past wash up on the shores of contemporary Oslo, leading to a refugee crisis of involuntary time travellers instead of war refugees. The refugees come from different periods: the Viking era, the nineteenth century and, like Navn, from the stone age. Navn has done very well for himself in contemporary Oslo. He has married an influencer and lives in an expensive house by a forest. Immediately before the forest scene he is shown looking out at the forest through the floor-to-ceiling window in the large modern house he shares with his wife, who is an influencer who has become famous for blogging about being married to a stone age man. In the next shots he is shown as a hunter, master of the forest, running naked, carrying nothing but the stone he intends to use to kill the rabbit. The drone’s sudden appearance is a plot twist, a moment of change: the hunter becomes the hunted. Technology wins over this natural stone age man who is powerless to intervene against this drone far above him.

The Beforeigners scene (Fig. 2.2) is reminiscent of many movies and video games: the camera’s point of view shifts to the drone, aligning the viewer with the drone. This shift to the drone’s eye view is more than a change in camera angle: it changes the modality of the image from representational to operational. Operational images are images that do something, that are data to be processed. While representational images are primarily about perception, operational images invite action (Parikka, 2023; Rettberg, 2023, pp. 15–17). Even though the viewer cannot really control the drone or act upon this image, the drone’s-eye-view perspective promises control and interaction.

Fig. 2.2
A screenshot of a drone's view of a man lying in the middle of the forest, with a zoomed-in view of the body in the bottom right corner.

The shots of Navn seen from the perspective of the drone that kills him look like they are from a video game. Beforeigners, s01e03 (2019). Copyright: Rubicon TV/HBO Nordic, reproduced here under fair use provisions. The CC-BY licence does not apply to this image; all rights reserved

Full size image

I watched Beforeigners on the TV in our living room and up until this point the visual aesthetics was conventional, like any other TV show. With the switch to the drone view, the screen suddenly looked as though it was hooked up to a gaming console instead of streaming a TV show. The shot represents the first-person view of a military drone, but to a non-military viewer it looks more like a video game. As I watched the series from my living room sofa it seemed as though my TV screen momentarily becomes a video game. I felt complicit, as though I was killing Navn—but I was not holding a controller. Seeing as the drone, I watch, helpless, as it targets and kills its victim.

Seeing from a Distance

The desire to see from a safe distance is an ancient human dream. The first technologies that allowed us to see at a distance—telescopes, opera glasses and cameras—were held close to the human body, held to the eye like a prosthesis. Satellites with cameras provided a nonhuman point of view, both allowing us to see the beauty of Earth from a new perspective, and giving a sense of control and the “view from nowhere”, that “god-trick” of imagined objectivity that Donna Haraway so emphatically tells is impossible (Haraway, 1988).

Drones are neither held in the hand as an extension of the human body nor are they still, silent objects far from human control, like satellites or gods. Drones are lively. They can be piloted by humans but can also be automated, flying independently. When you drive a car or ride a bicycle the vehicle becomes a part of your body. Perhaps it makes as much sense to say you become part of the vehicle. You do not hold a car in your hand, as you do a camera: you sit in the driver’s seat, or place your feet on the pedals, and become a driver or a cyclist, part of a human-technological assemblage. But anyone who has piloted even a toy drone knows that the sense of control is fleeting and precarious. The drone is separate from you, far above and perhaps even out of sight. The sense of being one with the drone might be something we long for, but it is easily lost.

Humans express the liveliness of technologies by anthropomorphising them, that is, by thinking about them as though they are people. Historically people have named ships and swords, but not “ordinary” technologies like cups and plates or screwdrivers. Thomas Stubblefield writes that we zoomorphise drones, imagining them as animals rather than as people (2020). They are a bit like pets or birds, possible to tame but different from humans and always a little wild. Unlike cars, ships and swords, drones also allow us to see differently, expanding our human perception of the world (Rettberg, 2023, pp. 60–82). This liveliness of drones is both enthralling and terrifying: they escape our control.

By anthropomorphising or zoomorphising technologies we form social and emotional bonds with them. Humans are social creatures, and we have evolved to depend on each other, to respond to each other, to be aware of each other and adjust our behaviour to those around us. Some technologies are explicitly designed to tap into human sociality. Conversational artificial intelligence like Siri, Alexa or ChatGPT are good examples. In the Beforeigners scene the juxtaposition between two camera angles mirrors the way interaction between two human characters is conventionally represented in cinema: the camera shows each character from a specific angle. The visual grammar of cinema reinforces the relationship between the two.

Although drones rarely have conversations with humans, their physical interaction with us plays into the same realm. Our bodies recognise the drone as another creature, as an opponent or perhaps a companion. Our relationships to drones are embodied, regulating and regulated by our bodies, by our hormones, endorphins and our breath. Navn stops, and looks up. The relationship is often stressful, more like that you might have with a lion or a stranger following you at night: you know they could hurt you although they might mean no harm. You know that your actions might set them off. You need to be careful. People sometimes fall in love with AIs. People are sometimes killed by drones.

Humans can encounter drones as drone pilots or as drone targets. In the scene from Beforeigners we are thrown back and forwards between these two positions. When we see the green control screen, the drone’s eye view, we are positioned as a remote pilot sitting at a computer. But as television viewers, rather than gamers, there is no way for us to control the drone. Nor does the drone’s eye view show interface tools to allow the implied fictional drone pilot to control the drone. Rings of concentric circles are centred on Navn, clearly indicating him as the target. Perhaps it is fully automated. The anonymity of the lack of control also speaks to Navn’s experience as he looks up from his stone age hunt to see this utterly incongruous technological object. Neither Navn nor the viewer knows where it comes from or who controls it. But it has immense power.

Any murder portrayed on screen situates the viewer as witness. When the point of view is that of the weapon itself, the viewers’ role as silent bystanders becomes more complex. We are positioned as the killer, not simply as a witness. But we have no agency—we can only watch. In video games, we pull the trigger ourselves. The combination of distanced and unemotional murder with our own embodied complicity as we see with the drone in a kind of cyborg vision, to use Ragnhild Solberg’s term (Solberg, 2022b), is a paradox that seems to both disturb and delight us.

In video games machine vision technologies like drones are often presented as a tool that empowers the player, even giving the player superhuman abilities. The monochromatic images from Beforeigners, where we see Navn from the drone’s perspective, emphasise the augmentation in the drone’s view: it sees not only from above, but it sees more than humans, differently from humans. The monochrome green suggests infrared vision rather than human vision, and the automated targeting circles show computational assistance. The letters on the overlay are nonsense, but suggest automated real-time analysis. In her analysis of machine vision technologies in the video games Call of Duty 4 and Cyberpunk 2077, Ragnhild Solberg argues that this kind of “superhuman vision” supports a “superhuman fantasy of domination” that is “also tied to the idea of player agency”. The “Death from Above” mission in Call of Duty 4 gives the player control of drones with infrared vision that are “overwhelmingly superior” to the enemy on the ground (Solberg, 2022a, p. 554).

Control

The question of control is central to the representation of drones in contemporary science fiction and video games. Navn loses control of the situation when a drone appears. The digital aesthetics of the drone’s eye view promises but also refuses control to the viewer. In many other works people are portrayed struggling to retain control of drones or using tools to disrupt drones. For example, police drones are countered by drones belonging to protesters in Cory Doctorow’s Little Brother (2008) and Tim Maughan’s Infinite Detail (2018), and so the police drones lose control, or at least have to fight for it. In Nadar Assor’s art film Lessons on Leaving Your Body (2014) footage shot by the drone is chaotic, wobbly, and the drone crashes. The shots composed by the human photographer stand in stark contrast: visually balanced, calm and beautiful (Rettberg, 2016). Some stories are told from the point of view of sentient drones. In Annalee Newitz’s Drones don’t kill people (2014), which is discussed in more detail in Gunderson et al. (2024), drones that have been used for assassinations gradually become self-aware, realising the horror of what they have done. They wrest back control over their own actions, refusing to be controlled anymore. In Yudhanjaya’s Wijeratne’s The Salvage Crew (2020) the ship AI, a digitised, once-human consciousness who is also the novel’s narrator, uses drones and spiders to be able to explore the planet with the ship crew, but their visuals glitch and the overfitting makes their image recognition “paranoid”; the drones and spiders break and are lost. Here the drones are part of the AI’s sensory apparatus and give the ship more control, until they no longer work and cause a loss of control.

In this chapter I approach drones as a technology that participates in an assemblage with human and nonhuman characters and with other entities such as the military. This is not the only way of analysing drones. While I view them as a single technology that works differently in different contexts, I could separate domestic drones from military drones, as Anna Jackman does in Domestic Drone Futures (2022). I could treat drones as media, as Julia Hildebrand does in Aerial Play (2021), or as art, as Thomas Stubblefield does in Drone Art (2020). By focusing on them as technologies I draw upon science and technology studies (STS) theories of how technologies and humans shape each other and on posthumanist theories of assemblages between human and more-than-human entities. By analysing these drone-human interactions through this lens I hope to open up a less anthropocentric perspective. The scene where the drone kills Navn can easily be read as a simple example of how drones kill. Reading the scene in more detail, and alongside many other portrayals of drones, allows us to see how the relationship between drone and human is becoming more complex. Navn is killed, but the scene also shows the agency of the drone as being outside of human control. We don’t know who controls it, and the visual aesthetics simultaneously offer control to and withhold control from the viewer.