Between Smart Images and Fast Trucks: Digital Surveillance and Obscured Labour in Hyderabad, India

Abstract

In Hyderabad, India, the growing information technology (IT) sector relies on ensuring safe and efficient movements of people and objects, and the city government and private actors have embraced the promise of digital surveillance to reach these goals. The new Telangana State, created in 2014, has built a new city-wide network of smart cameras, and at ‘hackathons’ programmers develop new digital tools, often connected to this network, that will technologically ‘solve’ social problems. In this article, I examine the system of CCTV cameras and programmers’ investments in these systems, and explore how migrant Vaddera stonecutters use cellphones to evade patrolling officers monitoring the streets where they carry the granite stones that they cut and load to construct the city’s buildings. Expanding on what Gilbert Simondon calls ‘the margin of indeterminacy’, this article reveals gaps in the digital infrastructure of surveillance—even as its integration and completion combine human and technical elements.

Keywords:

• India
• Smart Cities
• Infrastructure
• Surveillance
• Labour
• Videos
• Roads
• Cellphones

Within this urban network of surveillance, migrant workers like Vaddera caste stonecutters—who labour each day to quarry and cut the stones that provide the physical material for everything from new roads to the foundations of luxury apartment buildings—appear and disappear as they pass in front of one of those many CCTV cameras. In this article, I focus on how stonecutters have drifted in and out of the field of view of the surveillance system designed by computer engineers in Hyderabad, which comprises 36 CCTV cameras per 1,000 people, as they go about their work, cutting and moving stones across the city, while navigating older policing systems (Inzamam & Qadri 2022). I also explore how these stonecutters use their cellphones to act within the growing, yet incomplete, digital infrastructure of surveillance in the new Telangana State.¹

Surveillance infrastructural initiatives of this sort are not new to India and have historically been linked to Indian governments’ enthusiasm for technological innovations (Philip 2016). But recent national projects have increased digital monitoring of the Indian population. For instance, Aadhaar, a national program launched in 2009, is meant to secure universal welfare provision and secure citizenship entitlements, reducing corruption and fraud, via biometrical unique identification, which ‘promise[d] to build a “national grid” of identity information infrastructure’ (Chaudhuri & König 2018, 127). Similarly, the ‘100 smart cities’ project has been conceived so that urban populations across the country could fully benefit from the ‘datafication’ of areas and infrastructures that earlier were ‘invisible, unaccountable, unrecordable and thus ungovernable’ (Datta 2018, 407; see also Hoelscher 2016). Ensuring the safe and efficient movement of people and objects has been seen as a necessary condition for, and a characteristic of, the growth of interconnected ‘smart’ and ‘global cities’ such as Hyderabad (Datta 2018; Maringanti 2010; Xiang 2007). To achieve this ideal of seamless urban circulation, the new Telangana State has mobilised its IT resources to encourage the design and implementation of new digital forms of securitisation, such as video surveillance. Indeed, Telangana’s digital infrastructure of surveillance now includes 600,000 CCTV cameras that can be monitored at the new Command and Control Centre, which was inaugurated on 4 August 2022 by Kalvakuntla Chandrashekar Rao, the Chief Minister of Telangana (Amnesty International 2022). A few blocks from HITEC city, the new centre manages a surveillance network described as capable of ‘monitor[ing] every movement of the people and within no time track the crime and take instant action’ (Hans News Service 2022).

This visual monitoring is integrated in, and shapes the construction of, Hyderabad’s urban infrastructures. It is not only a system of control distributed throughout the city, however. It is also a fragile, evadable infrastructure, one which rests as much on the uncertain promises of prosperity that these infrastructures embody (Dalakoglou & Harvey 2012) as it does on technological sophistication or efficacy. A busy city road in India is how upper middle-class professionals make their way to and from modern buildings, which were branded to evoke ‘aspirational social identities’ during the 1990s real estate boom (Searle 2013, 273; see also Brosius 2010). But it is also how Mumbai taxi drivers creatively drive ‘through ecologies of smog, air, road debris, heat, and urban foliage’, and how they ‘sensorially work with and in the world around them rather than against it’ (Bedi 2023, 61).²

In urban areas, such as Hanoi (Turner 2024) and Hyderabad, infrastructural lives, relations, and practices that take shape across new roads and wireless connections are worked out through lines of technical (dis)continuity and provisional integration. Technical challenges integrate infrastructural lives in the present, just as these lives are shaped by prospects of more efficient technical systems in the future.

As Hannah Knox suggests, infrastructures such as roads are ‘undergoing processes of digitisation’ and they now include ‘sensors and communications [that] are placed in and on objects in order to make them part of an information infrastructure’ (2021, 185). In Hyderabad in particular, CCTV cameras, and its surveillance network, are not outside the city’s new roads and buildings. Rather, they are constitutive of these infrastructures, modelling usage and the work that goes into their future development. If, as Peter-Paul Verbeek (2006, 366) rightly states, ‘[t]echnologies help to shape what counts as “real”’, how then does video surveillance infrastructure create the realities requiring surveillance? How does this infrastructure complete itself, technically and socially, in Hyderabad? And, conjointly, how do the monitoring and monitored workers partake in this process?

With these queries in mind, my analysis focuses on how technological networks are completed by their interaction with users and with the fields of their real-world intervention. I discuss the social and technical adjustments that complete the digital video infrastructure of surveillance in Hyderabad. Following Gilbert Simondon, I consider the role of the ‘margin of indeterminacy’ in the very technical features of the video surveillance infrastructure. Such a margin expands as the surveillance network is being installed—it is the ‘threshold’ of the technological system, where human and technical elements are combined and interact. To do so, I track the process of ‘coming-into-being’ of this infrastructure, as it is formed, and placed, in specific contexts of action (Simondon [1989] 2017, 237). In this perspective, an infrastructure in development can find itself at the ‘meeting point between two milieux’; where and when human choices intervene by ‘attempting to realize the best possible compromise between these two worlds’ (Simondon [1989] 2017, 55). By proposing such an understanding of infrastructure integration or ‘the completion’ of the technological system, my analysis seeks to further exploit, analytically, infrastructural ‘gaps’ (Dalakoglou 2016), ‘breakdowns’ (Star 1999), ‘shortcomings’ (De Boeck 2012), and ‘leaks’ (Anand 2015; 2017), which, I argue, turn our attention to sociotechnical milieux, thresholds between technology and person that may be otherwise unrecognised. I specifically analyse how deidentification algorithms and wrongly positioned cameras actively shape milieux in which individuals evolve, as a new hybrid of material infrastructure and sensing technology is being developed. In dialogue with discussions around the notion of ‘sousveillance’ (from French, ‘watching from below’), I show that these milieux also arise from stonecutters’ ‘evading tactics’ by which they conceal certain work practices and develop an alternative information infrastructure by creative use of cellphones, trucks, and roads. These observations highlight tensions of visibility/invisibility in the high-tech city, between the desires of authorities and stakeholders to record, control, and optimise work practices via digital surveillance. It also looks at a parallel tendency, on the part of both powerful and subaltern actors, to hide, silence, and minimise the realities of work conditions. In an attempt to reflect meaningfully on the actions taken by workers at the edges of surveillance infrastructures, I conclude this article by discussing ‘obscured labour’ as a conceptual tool for analysing the tensions inherent in any large-scale surveillance systems.

First, I examine how surveillance infrastructures are built through prototyping and experimentation. I discuss how ‘smart’ CCTV cameras have been designed by computer engineers in Hyderabad ‘hackathons’ that I attended—48-hour or week-long challenges organised by the new Telangana State in 2014. Then I describe the only partial installation of these new devices in low-income neighbourhoods where stonecutters live, and show how the stonecutters interact with, evade, and work across information infrastructures and humble material technologies in ways that parallel the ‘high-tech’ work of coders, but with different results.

(De)identifying the ‘Unfavourable’

For a long time, computer engineers have been a major social grouping in Hyderabad, and a constituency deeply concerned to develop the capacities of the new Telangana State, created after a successful social movement in 2014. In addition to their commitment to building new capacities for the state, their ethical and moral orientations are on display at ‘hackathons’ where engineers meet to tackle ‘social issues’—imagining and demo-ing prototypes for new technological solutions to problems of urban life and governance. Some of these prototypes, when deemed successful, have been integrated into Telangana’s digital infrastructure of surveillance. The Hyderabad Control and Command Centre, as well as previous smaller monitoring centres, have been leaning on computer engineers’ prototypes of drones, cellphone applications, wearable digital devices, and smart surveillance cameras as they have expanded their capacities—and ambitions (Messier 2022; Coleman & Golub 2008).

These hackathons were meant to produce different paths for the commercial development of ‘smart’ camera systems and provide a foundation for urban deployments in the name of safety. In one of these hackathons, participants focused on solving issues of safety and urban traffic by producing and refining iteratively—via human discussions and computer processing power—the computer vision signals of threat, risk, and danger in Hyderabad’s streets and workplaces. One of these participants, a young man in his early twenties, explained how his team’s project could recognise (Figure 1) an ‘unfavourable activity’ in streets, shops, or retail stores:

[T]he particular project we are doing, we will have a solution so that people [who] are not doing anything wrong, their data should not be recorded. If they are doing something wrong, any unfavourable activity, stealing something, etc., then your data will be identified, and you’ll be punished accordingly. This is projected as a smart camera solution, a camera which is able to track activity.

The ‘solution’ proposed was a kinetic device that could decode human movements and identify some as ‘unfavourable activities’.

Figure 1. Screen demo during one of the Hyderabad hackathons (2014). Video by Philippe Messier.

This early ‘smart camera solution’ already used algorithms that could learn to recognise specific patterns of movements (see the points in the ‘deidentified’ blue silhouette in Figure 1) and then flag these sequences in the recordings. Once connected to a surveillance video infrastructure, the current versions of such devices can compensate for the absence of constant human monitoring, or its inconsistencies. Such a process includes the potential for mistaken identifications, as the images recorded may not be correctly interpreted by the system. When visual analysis tools are deployed to monitor larger crowds, streets, and large workplaces, these mistakes are common ‘due to fewer pixel per target, perspective effects, high density with heavy occlusion, a different variation of poses, variable appearance, strange clothing and different camera orientations’ (Elharrouss, Almaadeed, and Al-Maadeed 2021, 12).³

Whether they are integrated or not in larger online networks and platforms, the internal systems of these ‘smart’ cameras—their firmware and software—remain fallible in their capacities to capture and catalogue video sequences.⁴ Computer engineers in Hyderabad who, over the years, have presented to me similar flawed solutions, developed through tinkering and failures, have been disappointed to notice that there is still much work to be done. One recurring issue has been the positioning of cameras, which, even in the best scenario, cannot accommodate all the angles needed for AI algorithms to recognise ‘unfavourable activities’. One of the officials discussing the prototype during this hackathon, and who had a large experience of reviewing CCTV footage, was enthusiastic about the ‘smart’ camera systems deployed but also reminded participants how crucial the camera’s position was. He recalled situations when lengthy sequences of CCTV camera footage could not provide a usable image for agents to clearly identify a potential suspect because of inadequate camera position. In such contexts, the official suggested that the ‘smart’ camera system would still have issues identifying potential suspects.

Such failures are part of a process important for technological development—and economic success—beyond the hackathons. Some years later, in 2019, Pradeep, one of the main organisers of these hackathons, was developing with his team an AI-based system that combined a ‘smart’ camera and a large TV screen set up horizontally (which was acting like a mirror) to enable customers in clothing shops to virtually ‘try on’ pieces of clothing. Parvati, the computer engineer, who was writing the relevant lines of codes, explained how she was still struggling to achieve the proper virtual clothing ‘fit’ for the expected customers. As I filmed the demo and the explanation, the creators were dismayed to see a virtual sari appearing and disappearing on Pradeep who was standing in front of the AI-enabled camera-TV screen combo. In this case, the position of the camera, as well as the movements of the person in front of the camera, could not realistically overlay the virtual saris on the person in the live video stream. When we met again in 2022, this project had been put aside, because the startup could not wait any longer for investors. Pradeep’s assessment that AI evolution was happening very quickly led him to focus on a different IT venture—a fairly typical result in the IT sector where projects often come and go, much like the computer engineers who conceive them.

Despite the setbacks mentioned, it is those early flawed but ongoing technological experiments that, over time, have probed the possibilities of computer-aided surveillance, algorithmic recognition, and tracking of patterns; and various applications have been installed as part of the wider infrastructure supporting monitoring bases like the Command and Control Centre of Hyderabad. However, this movement from prototype to urban installation is not a seamless progression towards efficiency, either—at each stage, there are failures and components that just don’t work. Thus, the 600,000 surveillance cameras that this centre now uses and controls were not installed overnight or without hiccup. News organisations have reported rats eating the cables of some of these cameras, while other cameras are positioned where light is too dim to capture usable video. In short, a good number of those cameras are simply not functional (Rao 2017). And yet, with and through these deficiencies, the aspiration to produce a dense web of video surveillance infrastructure has endured and the officials in charge appear willing to solve these technical issues and to ensure that the cameras installed in different sections of the city will become functional.

Along with these new systems that are appearing, the surveillance infrastructures of Hyderabad’s neighbourhoods also include newer patrol cars that were made available to police officers in the early days of the new Telangana State in 2014, as well as slightly upgraded traffic booths in recently renovated areas. New motorbikes with GPS trackers were also provided to police in 2017, and for officers on foot, streetlights, whistles, and bamboo sticks are still useful tools to alert and discipline legal offenders. These technologies and fixtures are parts of a surveillance infrastructure that reminds citizens of the possibility of being arrested. Yet these elements also comprise technical limitations that create opportunities to avoid legal peril.

Like other migrant workers, the Vaddera caste stonecutters who have migrated to Hyderabad and its booming IT sector since the 1980s have learned to navigate this old and new surveillance infrastructure by developing and sharing knowledge about its inadequacies (limited manpower, broken or insufficient equipment) and its opportune openings (for instance, via familiar and more accommodating guards or police officers). As producers and transporters of granite stone, stonecutters in urban areas must learn to move safely and quickly to their workplaces, but they must also develop ways to transport the unique commodity that they create across the very surveillance infrastructure that they help build. The new cameras installed across Hyderabad’s neighbourhoods have extended the surveillance infrastructure’s reach, as it seeks to compensate for its human and technical blind-spots, but this has also created new milieux of action, new thresholds of indeterminacy in which human actors ‘complete’ the infrastructure.

Wide-Angle Lenses in the Basti

Stonecutters by trade and as traditional occupation, Vadderas have been using their technical expertise to acquire a specific, if precarious, position in the urban construction sector.⁵ Largely concentrated in the neighbourhood of Raayinagar basti (neighbourhood), they have improved their settlement over time, from temporary housing and tents to houses of concrete, brick, and granite stone in the basti, using the same techniques and digging through hills of the very same granite that they work to cut daily in HITEC city (situated around 5 kilometres from their neighbourhood).⁶ In the streets of the Vaddera basti, stonecutting’s ‘everyday technologies’ are prevalent, such as coal to sharpen takis (chisels) and tractors able to supply air to compressor drills (Arnold & DeWald 2012). The combination of hand tools and power tools also indexes the current technological state of Vadderas’ traditional occupation. These technologies of work are visible and form the distinctiveness of this neighbourhood, alongside the other common street furniture of any Indian urban neighbourhood, such as lampposts, electricity poles and cables, satellite dishes, motorbikes, and the occasional car.

But in early 2019, something else appeared, as two digital CCTV cameras were installed in the main street of the small neighbourhood (Figures 2 and 3). Such cameras are equipped with wide-angle lenses that can capture a field of view of 80–100 degrees and are set back-to-back, approximating the ‘360-degree’ field of view achievable with more advanced digital camera set-ups familiar from, say, Google Streetview. The field of view of these cameras is similar across the city (as both computer engineers and I learned during Hyderabad hackathons, when monitoring officers exhibited their viewing station). The two-camera setup that was installed in Raayinagar was thus likely able to cover a total of 160–200 degrees from their fixed position above the main street. For reference, Figure 3 presents one section (about 160 degrees) of a 360-degree view taken in 2022, from almost the same position as the two-camera setup. My 360-degree camera was positioned high up, sustained by a pole, but it was not as high as the two-camera setup. However, the high angle shot of the surveillance cameras means that the field-of-view produced in Figure 3 is representative of the surveillance system installed. The recording was shown to some residents of Raayinagar at the time that I took it, and we discussed together the field of view that resembled the CCTV setup and its ‘distinct aesthetics’ (Cardullo 2017, 129).

Figure 2. Two-camera setup in Raayinagar, Hyderabad, 2019. Photograph by Philippe Messier.

Figure 3. Still frame from 360-degree camera, Hyderabad, 2022. Main street of Raayinagar. Video by Philippe Messier.

When we discussed further the setup in 2022, the Vadderas living in the area explained that the installation had been accepted by the local community after internal discussions. However, like other similar cases, the initiative originated from outside the community.

The cameras were set up in front of a new four-storey apartment building, a frequently used meeting spot because of its central location. Through these conversations, I understood that the video recordings had been designed for internal usage by the community and access to the recordings by the authorities was only requested in the case of an illegal action. The residents interviewed agreed that the video surveillance system could curtail local robberies, limiting, for instance, petrol being syphoned from motorbikes parked in the street.⁷ From that perspective, these residents considered the system beneficial and were not overly concerned by its presence in the main street of Raayinagar.

But while the decision to install the system was discussed locally and described as consensual in the community, the Vadderas with whom I talked were also aware that the footage could actually be requisitioned at any time by Hyderabad police officers. The exact reasons for this initiative were also not clear, which led to some educated guesses that revealed a certain skepticism about the system. The residents explained to me that they had heard the installation was on account of tensions between communities in other parts of the city. Flare-ups have happened in the city before, even though Hyderabad is often taken as a model of harmonious relationships between Muslim and Hindu communities. Such claims are consistent with the justifications given for installing other cameras across the city (Deccan Chronicle 2015). But in Raayinagar, Vadderas have told me over the years that while they used to, at times, quarrel with kids from other communities when they were young, it was never a significant issue—and generally they have been getting along with other communities. Thus far, CCTV cameras are present only in small numbers in this part of the city, and Vadderas can evade them by using the peaceful spots throughout Raayinagar that they have frequented over the years—from tiny alleys far enough from the main street, where they park their conspicuous tractors, to the top of the hill overlooking the colony, still mostly empty of residential houses and where they have built small Hindu temples and can catch a scenic view of the outskirts of Hyderabad.

Still, Vadderas have been historically described as ‘thieves’, ‘drunk’, and ‘rough characters’—negative views that they sometimes use to refer to themselves. They have explained to me, for instance, that the ‘thieves’ stereotype has stuck with them because stonecutting entails being close to houses, apartment towers, or businesses which, supposedly, puts them in an advantageous position to steal. Vadderas are, thus, discursively placed alongside other marginalised communities in Indian ‘global cities’, including ‘low-skilled’ working-class people, and are often deemed potential threats to upper middle-class residents.

Overall, the objectives for setting up the CCTV system remain blurred for Vaddera residents. But they have noticed that police officers show up from time to time to deal with altercations happening in streets where a system has been installed. The Vadderas who mentioned this also explained that if they engage in or witness disputes, they do not stand in front of the cameras’ suspected field-of-view to avoid being apprehended or being questioned as by-standers of a confrontation. Because only a few residents have viewed the footage recorded by the two-camera setup, and information about the entire video recording process is sparse at best, the residents must guess the blind spots in the frames to stay out of the recordings.

The situation in Raayinagar contrasts with that elsewhere in the city, where surveillance infrastructure is more seamlessly integrated into newly built high-rises and is omnipresent along roads. In Raayinagar, CCTV cameras have only been installed in a few places and, moreover, the system’s designers must contend with ‘the inertia of the installed base and [which] inherits strengths and limitations from that base’ (Leigh Star 1999, 382). A fixed camera, even if it sports an ultra-wide-angle lens, can hardly cover visually the curved and uneven alleys of the basti, where Raayinagar residents sit, talk, and, at times, argue. The CCTV cameras are also attached to the infrastructural assemblages of lampposts and electric poles, which do not necessarily offer the best vantage point for a surveillance installation. In parallel, and while CCTV cameras are likely to appear in larger numbers in the basti soon, Vadderas seem disposed to a level of indifference for objects and persons that indicate the monitoring (and possible disciplining) of their actions. Following Julie Soleil Archambault (2021, 52), I would propose that for Vadderas, thus far, and to an extent, CCTV cameras have this ‘capacity of objects to fade out of focus, to become invisible even while remaining in plain sight’.

Nevertheless, the integration of CCTV cameras in newly constructed areas of Hyderabad, where the surveillance video infrastructure can more easily converge with previous infrastructures, has been progressing. 1,200 cameras are, for instance, expected to be installed on the Outer Ring Road—a 156 km high-speed stretch of new tarmac road surrounding the capital. Similarly, as I observed and filmed in 2022, numerous CCTV cameras have been installed in the Financial District of Hyderabad, where new wide roads and luxurious modern office buildings are regularly being constructed. If, in their neighbourhood, the Vaddera stonecutters can more easily evade and ignore cameras, out on the roads and across buildings in construction, they are more aware of them, as well as the police officers patrolling these more regulated spaces.

Improved Vigilance

Constant monitoring of construction sites—of future roads, IT offices, or apartment towers for computer engineers, such as the ones where stonecutters work—is seen by the IT sector and the national and state governments as a way to promote steady and reliable productivity (Messier 2020). The expansion of video surveillance, in and across such workplaces, can convince multinational IT companies that, even from a distance, the work accomplished will be of the highest quality. In 2020, the IT Minister of Telangana, Kalvakuntla Taraka Rama (KTR) Rao, reiterated the importance of this objective, highlighting that video surveillance projects will be needed ‘[a]s the city is expanding and attracting more investments, there is a need to improve the vigil [that is, enhance vigilance]’ (Telangana Today 2020). Some of the computer engineers who participated in the Hyderabad hackathon have recently teamed up with videographers and, together, they have been mobilised to design and operate tracking audiovisual technologies, such as drones and ‘smart’ cameras, that can respond to the needs of secure and monitored areas, as envisaged by KTR Rao. For instance, the promotional videos used by government agencies involved in road construction in the region do not simply, as one video producer said in 2022, serve to ‘beautify’ these new infrastructures. Rather, their very construction requires a new infrastructural layer, one of progress monitoring, quality insurance, and safety surveillance. As our discussion went on, the videographer explained that:

[At] the government level, they do track progress through drones. They do that. So, for example if they have approved a road. […] So, whatever was being shown to the Minister or higher officials, they had, date-wise progress, [from] drones. […] And so, just imagine the number of people working on it, on each road. On each trip [to the road], there was one drone guy whose only work was to come in the morning, take a shot of the progress, go back, [and to] come [the] next day. And [then] give [the video] a timecode … ‘this was shot at this time, this date’. And that is how they communicate the progress to the higher officials, the minister, the decision-makers, the stakeholders, and the government.

The videographer highlighted that the videos are produced differently according to whether they monitor progress or are dedicated to general surveillance. And so, alongside promotional videos, companies that manage large infrastructure contracts track ‘the manpower changing daily’. For these companies, the videographer said, ‘the surveillance might be from an angle of the contracts that [the governments] have given [to] some X company, [which] has to execute that work’. And the CCTV cameras, owned by the companies ‘are used by their internal teams for surveillance, [and] to make sure that all safety measures are followed’. These new digital audiovisual surveillance standards echo what Waqas Butt described in Lahore waste management, where smartphones are used by coordinators to confirm workers’ daily attendance by selecting ‘an attendance point, tick off those workers that are present, whose names are all pre-loaded, and then take a picture of the group of sanitation workers assigned to that point’ (2020, 113).⁸

This is also coherent with a view shared by many Indian computer engineers that digital technologies can outplay corruption tactics and that transparency motivates such projects (see also Mazzarella 2006; Sargent 2022; Webb 2019). The overlapping desires for steady progress, efficiency, and safety in the public-private construction projects in India have also developed into customised forms of audiovisual surveillance for future homeowners. Construction companies now offer packages that merge home construction and monitoring services. These infrastructural kits combine surveillance and construction to provide consumers with what is being described as ‘transparent’ tracking systems that include remotely accessible CCTV cameras, guaranteeing elimination of what are deemed ‘malpractices’ in the process of house construction. Customers can observe, record, and recognise from a distance the daily practices of construction workers and stonecutters toiling beside an apartment tower or even at a specific apartment. The expansion of new digital and audiovisual affordances in Hyderabad means that both government agencies and future homeowners can monitor from afar the building of new infrastructures—from public roads to private apartments.

Stonecutters thus find themselves at this critical juncture of intensive productivity and privatisation in postliberalised India (Mitra 2018). The surveillance system that aims to follow stonecutters at, and between, workplaces includes CCTV cameras, monitoring centres, and police officers. As this system records stonecutters’ movements, joining two milieux while being ‘integrated to both of them at once’, these workers have developed their own ways to use the digital infrastructure in curtailing these tracking plans (Simondon 2017 [1989], 55).

Reconnaissance Tours and Silent Night Loading

Hyderabad’s surveillance infrastructure follows stonecutters when they leave their neighbourhood for work, early in the day or very late at night; meanwhile, they themselves deploy technologies ranging from very humble hand-tools to (rarely) smartphones in order to manage their own movements in ways that interact with, respond to, and hence ‘complete’ (while evading) this infrastructure.

In sites around Hyderabad, up to 40 stonecutters can work together in large quarries, with another 20–30 operating compressor drills connected to loud tractors, and one or two workers operating an excavator. In these quarries, stonecutters cut down big blocks of stone of approximately 1.5 m x 1.5 m with hammers and takis (chisels), to produce rectangular blocks around 30 cm long, 15 cm wide, and 15 cm deep. By and large, this type of stonecutting in Hyderabad is piecework, and stonecutters are paid per stone block.⁹ The workers stack their blocks in their own piles until truck drivers show up to count and load the blocks and transport them to construction sites. Some stonecutters help drivers in the loading process and then follow the drivers to unload the stone blocks. These blocks are either used whole, to form road-dividers or walls, or crushed for building foundations for office towers.

Stonecutters, like other construction workers, move from quarry to construction sites daily, across urban spaces, controlled by police officers, security guards and, at times, insistent supervisors (maistri), but also by the video analytics software, which learns ever-more quickly to recognise unwanted actions in video sequences captured by CCTV cameras. And yet, as Trevor Pinch (2010, 409) suggests, ‘old technologies live on alongside new ones’. Indeed, old infrastructural affordances coexist and overlap with newer forms of digital surveillance and help Vadderas manage their own movements through these layered infrastructures.

Thus, affordable cellphones are used to call, text, or send photos to coworkers if one forgets a piece of equipment, or if it is stolen during the night. These cellphones facilitate rapid movement from one quarry to another if a distant manager requests such a transition. These possibilities, created by a growing wireless infrastructure, offer access to important information regarding the workplace. The omnipresence of cellphone connectivity has been seen by the IT Telangana Department and IT-based NGOs as an infrastructural necessity, one that could lead workers—such as stonecutters—to ‘catch-up’ and enter the middle class by intensifying their production.

But the new possibilities afforded by cellphones have also been worked out differently by stonecutters. For instance, stonecutters have explained that if they do not want to work for a particular supervisor, or at a certain quarry, they can simply not respond when called and claim that they want to avoid repetitive calls from telemarketers. They might also respond, complain a bit, and use the physical distance produced by wireless communication to ignore the request, or claim they misunderstood it because of bad reception. In these cases, as Vincent Duclos points out, ‘[t]echnical glitches, failures to communicate’ and ‘absent users’ ‘are not external to, but rather constitutive of network connectivity and speed’ (2017, 25).

Stonecutters also use cellphones to bend traffic laws and get around road surveillance. Like autorickshaw drivers, who might take more passengers than they are legally allowed to, and who remain aware of the locations of police control stalls, stonecutters and lorry drivers exchanged information about police officers roaming the streets. This is necessary because, in Hyderabad, the lorries that carry stone blocks to construction sites are restricted access to streets during certain hours. Stonecutters will often move their loud lorries and tractors to construction sites early in the evening, before the cutoff time of 22:30, or try to complete their deliveries in the morning, after the accepted start time of 06:00.¹⁰ Even when they manage to follow these regulations, stonecutters endeavour to labour quietly in the dark to ensure that police officers remain at bay. In such conditions, and to complete their daily jobs, they return at night to prepare for the next day’s work, or to deliver stones to a worksite. Indeed, many must work overtime to load trucks and ensure that each stone cut is credited to their name. As one stonecutter explained back in July 2014: ‘I come at night partly to pay back my loans for the truck that I bought. Otherwise, I only break even … It adds to my daily income. […] About 50 per cent of [stonecutters working there] will come [at night]. If they only have a few stones cut one day, they might not but if they have many—like 100 stones or so—they will work at night’.

To accomplish delivering a load without being apprehended, stonecutters first track down police officers patrolling the streets for illegal loading out of hours. A stonecutter who owns a motorbike will then conduct a ‘reconnaissance’ tour of the area by driving in the surrounding streets of the quarry. When police officers are far enough away to allow work to proceed with little chance of interruption, the worker texts his fellow stonecutters and the lorry drivers so they can meet at the quarry, or leave that quarry once the loading is completed. Because digital tracking is not always accessible to police officers patrolling streets, traffic can accelerate and escape penalties. The big trucks used for stone blocks are not as fast as the sport cars often seen driving through Hyderabad’s affluent areas. But a truck engine can be pushed to the maximum when a driver receives a text message confirming that police officers are looking the other way.

If workers are caught by police officers, who can be alerted by agents monitoring surveillance cameras, bribes are involved. In some cases, stonecutting contracts pre-emptively include a dedicated amount for these bribes for the purpose of limiting construction delays and smoothing out relationships between police officers and stonecutters. Still, many interactions remain contentious or at least unsatisfactory for stonecutters.¹¹ In 2019, while I was discussing such encounters with Vinod, an experienced stonecutter who I apprenticed under, I learned that when a tractor-owning stonecutter receives an offer from another stonecutter to rent the tractor, the tractor owner coordinates with a cellphone to avoid patrols (tractors are also limited in their movements). Since the tractor owner already charges the stonecutter INR 1500 per day to rent the tractor, the leasee seeks to avoid being pulled over and fined INR 500 by police (less if a bribe is involved), so they leave the quarry after 06:00 on restricted roads. Speaking about the text exchanges he uses to communicate when and where an illicit movement is possible, Vinod, who owns a tractor and drives it to construction sites and also leases it to others, explained: ‘[s]o, why would I make my friends pay if I paid [the fine] already?’.

Across these terrains, stonecutters redirect communication paths and block exchanges of information about their whereabouts, acting through data ‘leaks’ at the threshold of technical and social ensembles. Such technical actions create milieux that emerge from stonecutting conditions—milieux that are not simply uncovered by but rather arise from a digital infrastructure of surveillance that is being completed as stonecutters act selectively upon it. These milieux of concrete infrastructural entanglements exist when certain people and devices slow down (De Boeck 2012): maistris who cannot reach stonecutters via cellphones and, when other machines and people speed up, police officers who miss those drivers sneaking through certain intersections and disappearing down barely lit alleys, away from the gaze of human and camera. These are acts of discovery that concretise an infrastructure that associates and reveals otherwise unlinked milieux.

The combined actions of ‘reconnaissance’ tours and text exchanges by stonecutters may also be thought of as forms of sousveillance. Steve Mann, Jason Nolan, and Barry Wellman (2003, 333) have initially defined the notion as practices that ‘[enhance] the ability of people to access and collect data about their surveillance and to neutralize surveillance’. Sousveillance has been used to describe how hackers ‘video sniffed’ CCTV systems with digital video receivers (Cardullo 2014). How consumer drones have made sousveillance a process that is ‘airborne as well as taking place on the ground’ is considered by Dennis Zuev and Gary Bratchford (2020, 447). ‘Professional sousveillance’ training has even been offered to address the need for visual evidence in legal cases. Brandim Howson, who has been studying the growing importance of professionally produced visual sousveillance in favelas in Brazil, suggests that ‘[b]y deploying these sousveillance practices, the activist is able to make visual order out of disorder’, and where NGOs and activists ‘may have been overwhelmed by a particular incident they encounter, emphasis on the production of visual evidence allows them to disengage from the immediate chaos’ (2020, 278). In other instances, sousveillance practices have more modestly been linked to activists, inciting ‘civilians to engage in incidental monitoring of police with cell phones and other cameras’ (Brucato 2015, 457).¹²

However, stonecutters in Hyderabad are not actively producing documents about their encounters with police, and their navigation of the surveillance infrastructure is mostly tied up with daily work patterns and needs. Similarly, in their ordinary ‘infrastructural lives’, they use technologies and techniques that are readily available to them in the digital infrastructure without necessarily distinguishing a particular political meaning to any one device or practice.

In their own context, to be sure, stonecutters also share information about street controls and seek to carve out some autonomous space, but they are aware of their limited power over official authorities.¹³ On any day, stonecutters might exchange digital photos of missing or stolen working tools, and of the new quarries where they work. They do not snap photos or record videos of police officers monitoring the roads, or of CCTV cameras, but they routinely share text messages or calls about ongoing surveillance, focusing on new police booths and incoming patrols. In the layered technological milieu, moreover, texting and calling can be more subtle modes of digital exchange and interaction than capturing and uploading photos or videos.

Moreover, because many of these Vadderas seek bargains, it is quite common for them to have more than one SIM card, and to change these cards over time. In doing so, these workers become less easily trackable across the digital infrastructure because the official registration needed to acquire the SIM card in the first place is often dispersed by human mediators, through locally known employees and friends at cellphone shops, and, ultimately, does not necessarily correspond to users’ current addresses.¹⁴ Such a practice does not neutralise surveillance, but it limits its reach. This also underscores how the ‘low-tech’ infrastructural layer of digital information that the Vadderas accessed and used, in part because of their own financial constraints, matters to the larger-scale infrastructures in which the Vadderas circulated. Text messages are more subtle than photos or videos, but they also consume less data than visual graphic exchanges, while calls can often be made by someone without remaining credit in their cellphone data plan. Because of the same financial concerns, and for convenience, many stonecutters go to cellphone shops to re-charge their memory cards of Telugu songs and videos that they then listen to on the job instead of downloading these files with their own connection. The practice partly escapes internet monitoring, and it operates through smaller informational networks. As such, and even as the digital infrastructure grows, it also creates and maintains provisional milieux of technical and social convergence by which stonecutters’ work practices remain opaque to surveillance.

Concluding Remarks: Framing Obscured Labour

Stonecutters’ movements across the city have always partly escaped surveillance, even as they help build that city. As the new Telangana State continues to project Hyderabad as a haven for economic growth, the lenses and positions of ‘smart’ cameras, as well as the video analytics processes defining appropriate practices, have repurposed older infrastructures, such as lampposts, electric pylons, buildings, and roads to form Hyderabad’s digital surveillance infrastructure. Even through delays and issues, such a process is happening at a rapid pace not only in the new Telangana State but also across India where ‘the absence of policies to govern these systems is allowing the state to build complex and far-reaching surveillance structures’ (Purandare & Parkar 2021, 1411).

In the meantime, ‘smart’ cameras have now burst open the corporate doors—flying up in the air on drones and attaching temporarily to buildings in construction to track work practices. Although these are early steps, the increasing mobility of these devices is likely to convert informal workspaces into more formal ones. And it is conceivable that in doing so this process will reshape workers’ creative and intimate relationships with productive techniques, especially if, for instance, the techniques recorded are shared as appropriate examples of work practices to follow or not.

Stonecutters’ frequent adjustments to this surveillance infrastructure have been and will likely continue to be part of their daily toil. The material explored here can help us to refine further our understanding of what is happening for monitored workers as they engage with and participate in shaping the indeterminacy of their infrastructural lives. In thinking with the term ‘obscured labour’, I suggest that we may more meaningfully address the varied sets of unrecorded actions, deliberate avoidances, and partial knowledge that surveillance infrastructure appears to provoke as it grows alongside and with workers. The term must also encompass the types of work and workplaces that may be monitored for safety or efficiency but that otherwise remain largely invisible, silent, disposable, and beyond one’s idealised experience of smart cities (Anteby & Chan 2013; Zhang 2019; Doherty 2017; Turner 2024). These two obscured ensembles often coexist and reinforce one another, because concealing processes of certain forms of work in a given context can lead workers to desire and take the opportunities to further avoid recognition, to remain hidden, straddling the lines of necessity and choice. Importantly, the development of an ever ‘smarter’ and high-tech Hyderabad deliberately obscures the labour that produces it—which may now be integrated into algorithmic and automatic processes—but does not eliminate it.

For stonecutters in Hyderabad, obscured labour means that they must hide their loud and imposing tractors in the tight and zigzagging uphill alleys of their basti. Or that they are required to cut stones behind the aluminum temporary construction fences installed around construction sites and quarries, which conceal stonecutting realities—a working condition stonecutters had to adjust to well before the new smart cameras appeared in Hyderabad. But obscured labour also means that they actively dodge out of these CCTV frames, move into unseen alleys for heated discussions, execute sousveillance or countersurveillance by texting information via cellphones about police officers’ whereabouts after ‘reconnaissance’ tours on motorbikes, and, at times, rev up lorry engines to load stones unencumbered at odd hours—all in a largely unplanned but not completely improvised fashion.

Taking a wider perspective, the milieux that are provisionally organised by these actions bring together—and are made of—otherwise disconnected ensembles of techniques, people, and times through the same, converging, urban and digital infrastructures. These milieux also result from digital forms of obscured labor as they appear at the edges of these infrastructures’ new sensing capabilities, as indeterminacy and a force of creative re-routing. For example, when computer engineers tinker with algorithms in ‘smart’ surveillance cameras and manage to improve facial recognition in Hyderabad’s new Financial District, but also when stonecutters successfully drive their lorry away from a police officer who is requesting them to stop when they are caught off-hours in Hyderabad’s early morning traffic or on the darkest nights. As a concept, obscured labour thus aims to focus on a particular kind of labour: the under-recorded, partially monitored, and unsensed labour. From that point of view, the sensorial is not only how the workers experience the streets, but also how these workers create their own ‘image’ of these streets and are ‘imaged’ in turn by digital surveillance. Obscured labour happens through overlapping moments—when there is too little light information or too much video noise to capture a usable image for facial recognition, too many sounds to distinguish a voice from the layered audio signals, or too few movements to sense any human actions.

Such actions happen at the interweaving of digital vision and material infrastructure, where human visual recognition and understanding of recorded realities are becoming less important than the relationships between computer vision algorithms (Uliasz 2021). Deep learning processes and online platforms have indeed been found to work more and more at a level of ‘invisuality in which relations between images count more than any indexicality or iconicity of an image’ (MacKenzie & Munster 2019, 16). As the margins of indeterminacy play out and create milieux of action, as well as gaps in the mesh of surveillance and control, where marginalised workers stay in obscurity, the growing surveillance infrastructures in Indian cities are likely to be subjected to the expansion of AI capabilities, which are, already, blurring the boundaries between what is seen or not seen.

Acknowledgments

Thank you to the computer engineers, truck drivers, and stonecutters who participated in this research in Hyderabad. I thank the members of the editorial team at TAPJA and the anonymous reviewers for their insightful suggestions for revision. Finally, thank you to Leo Coleman for engaging generously with the argument proposed in this article.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Field research for this article was funded by the Social Sciences and Humanities Research Council, Government of Canada, and the Fonds de recherche du Québec Société et Culture.

Notes

1 This article draws on 20 months of fieldwork research in Hyderabad between 2013 and 2022. The main part of the fieldwork (15 months) was completed in 2013 and 2014, with subsequent visits in 2016, 2019, and 2022. Fieldwork included semi-structured interviews (individuals and in small groups) with more than 60 Vaddera caste stonecutters, conducted during work breaks (while I learned stonecutting as an apprentice) and at their homes. Over the years, I produced and reviewed videos collaboratively with these stonecutters and asked questions about the impact of audiovisual recording on their daily labour. Reviewing sessions gave us time to discuss video capturing and editing processes (from different lenses to video effects, and more recently with 360°/VR cameras). I also conducted 31 interviews with computer engineers engaged in designing—in hackathons, startups, and large IT companies—the types of ‘smart’ cameras that are being installed across Hyderabad. These interviews included discussions on the technical details of the audiovisual capabilities of this surveillance infrastructure. I also resided with some of these engineers during my fieldwork. I finally interviewed 18 videographers who have been hired to produce corporate videos for IT and pharmaceutical companies in Hyderabad. Thematic analysis (recurring patterns and comparisons of themes) was used to analyse interviews and videos.

2 Anjaria (2020) explores how certain forms of (slower) mobility in urban India, like cycling, can offer different sensory engagements with these infrastructures.

3 MacKenzie and Munster go further and claim that once these systems are linked to other online platforms, the ‘operativity’ of these images ‘cannot be seen by an observing “subject” but, rather, is enacted via observation events distributed throughout and across devices, hardware, human agents, and artificial networked architectures such as deep learning networks’ (2019, 5).

4 Some of the same engineers I worked with have highlighted in recent years that many of the cameras installed across the city also include firmware conceived in China. Although these experts can reverse engineer these pre-loaded designs, they have expressed their suspicions for the numerous cameras and systems that are being put in place with no adjustments across the city and the state.

5 In Telangana, Vadderas are classified as ‘Other Backward Classes’ (OBC) in the national census.

6 In this article, I use the fictional name Raayinagar to identify the neighbourhood where Vadderas reside in Hyderabad.

7 Residents have notably connected this recent practice to the COVID-19 restrictions that have severely constrained the construction sector and stonecutters’ livelihoods. See also Annavarapu (2022).

8 Construction workers in large-scale infrastructure projects have also been monitored via visual analytics of fixed video and photo cameras. Guo, Luo, and Yong (2015) have discussed how such a system catalogued the unsafe practices of the construction of Wuhan Metro.

9 Stonecutters who own tractors and can find the helping hands needed for long-term projects in their networks may accept contractual work, with a set price to clear an entire lot and smash stones in small pieces to build the foundations of what will usually become an apartment tower.

10 The accessible hours have changed over the years, but since July 7, 2020, according to the notification 75/1/T2/DCP.Tr/Cyb/2020 of the Government of Telangana State Police Department, the heavy lorries used for stone blocks cannot use the streets during peak hours. Stonecutters explained that, in certain neighbourhoods, excessive noise from machines or workers at any time during the night can also be deemed ‘illegal’ and that fines will be issued for breaching the regulation.

11 This is not to say that such interactions are enjoyable for police officers in India. Jauregui (2016) notably explains that these bribes are often used to compensate for the scare resources available to maintain adequate policing services.

12 Mann (2020, 266) has recently conceived ‘participatory sousveillance’ more neutrally as referring to the ‘recording of an activity by a participant in the activity’.

13 Browne (2015, 22) highlights the importance of ethnic and racial contexts, and describes the ‘dark sousveillance’ through ‘the use of a keen and experiential insight of plantation surveillance [by slave patroller] in order to resist it’.

14 See Assa Doron’s work (2012) on the impact of those different social actors in the Indian grey market of cellphones’ adaptations and repairs.