What does a sensor know?

Abstract

This paper explores the limitations of sensor datafication by examining the distinction between “thick” and “thin” description. While sensors excel at capturing “thin” data - characterized in traditional ethnography as raw, observable behaviors akin to what a camera records - they struggle with “thick” description, the interpretation of intentions, meanings, and cultural context. Responding to Borbach and Kanderske's work on surveillance counter-practices, the paper questions the extent to which sensors can discern between accidental glitches and deliberate subversion. The paper concludes by considering the implications of emphasizing intentionality in resistance, suggesting that while such practices are crucial, they may ultimately contribute to the refinement of sensor accuracy and power and limit how surveillance scholars conceptualize surveillance subterfuge.

What does a sensor know? By some accounts, quite a bit. Sensors have become ubiquitous and continuous, transmitting data in constant flows. Sensors gather more information about people than people have about themselves, tracking where we are and where we’ve been, and even helping to predict where we will be in the future (Chon et al., 2011). From the perspective of surveillance studies and privacy scholars, the mass integration of sensors represents an incredible power imbalance and intrusion on civil liberties. Yet there are entire categories of data about which sensors know very little. They can track our location but not our intentions, they can create heat maps of our bodies and movements, but not (at least, not directly) our thoughts and feelings. Sensors can document, and their data can be used to predict and infer. But a sensor itself cannot interpret. This is a technical problem but it's also an anthropological one. It is in fact a problem deeply familiar to ethnographers, who often conceptualize it as the difference between thick and thin description. In thinking with Borbach and Kanderske about counter-practices of sensor datafication, I want to use the analytical division of thick and thin data as a heuristic for thinking through what sensors do and do not know, as well as the politics of their subversion.

Thick and thin description: an anthropological dilemma

For much longer than I’d like to admit, I fundamentally misunderstood thick description. Even after taking courses in ethnography and doing ethnography myself, I saw thick description as fundamentally about depth of description. Thick description was, I thought, mostly about gathering enough detail to immerse a reader in the field site and among participants, as opposed to the shallow, superficial and anecdotal observations of “thin” description. Providing this sense of immersion through rich detail was a sign of skilled observation, dedicated notetaking and a hallmark of solid ethnography. Of course, all of these things are true: it is important to convey the field site and its people to the reader. But detailed description isn't quite the same thing as thick description, because the difference between thick and thin description is of kind rather than degree.

In what has become a pivotal text on anthropological methods, Geertz (2017) used thick and thin description to highlight a foundational task of ethnographic work: identifying the underlying meanings, intentions or value to surface-level behaviors. To illustrate the distinction between thick and thin description, Geertz offers the example of closing one's eye on purpose versus incidentally, a wink versus a blink. Thin description observes behavior and thus cannot disambiguate a wink from a blink, while thick description entails explanatory power and can thus label one action as intentional and the other as incidental. Partly, this distinction is meant as a distinction of ethnographic work: Anyone can observe, but an ethnographer can interpret!

In his ethnography of African Hebrew Israelites of Jerusalem in New York City, Jackson (2013) issued a challenge to the persistent call for thick description in anthropology, arguing for a need to recognize the limits of what researchers can truly know about the people they study. In his division between these two approaches to observation, Jackson argued:

Thin description is what you can see with the naked eye. It is a raw and baseline empiricism, the necessary starting point for social investigation but not nearly enough all by itself. Part of an anthropologist's job is to contextualize social behaviors for readers, behaviors that are never purely self-evident and that always reward more careful scrutiny … ‘thick description’ is used like a mystic metaphor or methodological talisman that denotes an attempt at - an ambition for - rich, rigorous, and even full social knowing. (p. 13–14, emphasis in original)

For Jackson, thick description suggests a kind of hubris, a self-assurance of interpretative validity. While insisting on the value of ethnography as a method of gaining insight into human behavior, Jackson argued that thin description has been overlooked as a valuable (and perhaps more valid) source of information about the media and meaning-making work of the people, communities and practices we study. Writing about the affordances (and related obstacles) of ethnographic tools in literary criticism, Love (2013) provided a detailed analysis of how key figures in anthropology established a dichotomy of thick and thin description:

Thin description was an unadorned, first-order account of behavior, one that could be recorded just as well by a camera as by a human agent. Thick description, by contrast, added many layers of human significance, including attributions of intention, emotion, cognition, and depth, as well as cultural context and display. (p. 403)

Particularly relevant to the question of sensors, Love conceptualized thin description as largely technical, even mechanistic: “Thin description means, in effect, taking up the position of the device; by turning oneself into a camera, one could … pay equal attention to every aspect of a scene that is available to the senses and record it faithfully” (p. 407). Ethnographers’ allegiance to thick description comes in part from our resistance to seeing ourselves as mere devices recording thin description.

Sensors, I would argue, excel at thin description. While sensors may be installed or integrated into our devices with an ethnographer's desire for “rich, rigorous, and even full social knowing” (Jackson, 2013: 14, emphasis in original), they face an interpretive limit. Sensors observe and record, but they cannot interpret. By connecting the observational power of sensors to thin description, I do not mean to suggest that sensor datafication is somehow less worrisome. Indeed, Jackson (2013) and Love (2013) return to the practices of thin description precisely because it had been devalued by ethnographers in the work of human observation. By taking an anthropological and ethnographic perspective on sensors’ capacity for thin versus thick description, I want to focus more closely on the limits of sensor datafication, particularly when it comes to the question of deliberate (rather than incidental) obfuscation.

Glitches, counter-practices and intentionality

The cases highlighted by Borbach and Kanderske illustrate key tactics for working against sensory datafication: Hiding in plain sight involves being illegible to sensors but (often) hypervisible to humans, thus resisting data capture while displaying resistance to other datafied subjects. Dis/simulation relies on camouflage and false data to make datafication more difficult and less reliable. Exploitation leverages analog technologies to disrupt sensor logics, introducing noise into transmissions of surveillance data. Each of these counter-practices illustrates that subjects can effectively interrupt what sensors do through deliberate acts of subterfuge.

Intentionality is central to Borbach and Kanderske's analysis of subverting sensory datafication. Borback and Kanderske are clear in their pivot away from what they see as an emphasis in science and technology studies on “accidentally occurring crisis” (p. 6). In turning towards counter-practices, the category of behaviors they analyze requires deliberation, or “the active usage of sensors aimed at subverting their automated sensing and subsequent sensemaking processes” (p. 6, italics in original). This emphasis on intentional practices of undermining surveillance leads me to a series of provocations: if sensors capture thin rather than thick description, can a sensor know when it's being subverted? When might it matter to sensors when they are confounded by a glitch versus a counter-practice? When might it matter to anti-surveillance activists?

Sensors are relentless in their capture of what we do, even as they remain naive about why we do things. This naivety can be manipulated and exploited via the tactics that Borbach and Kanderske describe. But although intentionality may be definitional for the authors, when and how does it matter to sensors? In aligning the work of sensors onto thin description, I am arguably setting up a false distinction, given the integration of thin description sensors into thick description devices. Borbach and Kanderske rightly note that “digitally interconnected media and sensors can actually no longer be separated from one another” (p. 2). As such, even if sensors are a “thin description” layer of a datafication stack (Bratton, 2015), there is certainly enough thick description detection in the surveillance assemblage as a whole. But one provocation for a discussion of counter-practices of sensor datafication is to ask whether a division between glitches and practices is a distinction without difference for sensors and (ultimately) surveillance.

From an activist standpoint, the intentions behind disrupting sensors may matter less than the long-term consequences. In framing resistant tactics as forms of appropriation or limit testing (p. 20), surveillance scholars must acknowledge how testing the limits of sensor datafication can ultimately make these technologies more accurate and powerful. The history of social movements provides a cat and mouse game of developing resistant tactics only to see those tactics newly regulated or neutralized. Consider how Critical Mass bike protests at the Republican National Convention in 2004 helped NYPD develop strategies for curtailing this form of collective resistance (Hill, 2004) or how college campuses in the U.S. have cracked down on encampments following a pro-Palestinian demonstrations (Blake, 2024). Testing the limits of sensory datafication may or may not yield deep insight into their inner workings and sensemaking, but they will almost certainly provide additional information on how these devices can become ever more accurate.

By definition, counter-practices require deliberation, but these are precisely the thick descriptions that elude sensor datafication, at least for now. Borbach and Kanderske's typology of counter-practices illuminate useful characteristics for investigating the capacities of sensor surveillance, and their work may make it easier and more generative to lay claim to future practices of subversion and resistance. Yet testing the limits of sensor datafication also risks inadvertently strengthening the surveillance apparatus. If what we want is for sensors to know less about us, what does it mean to focus on counter-practices that convey our intentions?