The Political Economy of Online Crowdwork and Caste

Introduction to Online Crowdwork

The blending of human labour and computer software to create seemingly automated digital services is not a new development. Remote digital work of various forms has existed since the mid 2000s and can be seen as the latest form of work in automation’s long march (Gray and Suri, 2019). Historically, the march in question has led to the creation of new types of labour to meet the needs of emerging technologies. The work that is mediated by software today is similar to factory jobs during the industrial revolution, the piecework performed by women and children on 19th century farms, and the outsourcing of call center work to the Global South with the expansion of the internet and liberalisation in the 1990s.

These serve as forerunners to online crowdwork, which involves a system of fragmented tasks distributed across the internet. Like their precursors, these tasks are characterised by being small, repetitive and isolated. They offer little stability and are mostly done by people whose labour economists classify as “low skilled”. The same labour, which is often low-paid and invisible, makes AI “intelligent”. For example, Filipino workers manually identify pedestrians in video footage used to train self-driving vehicles (Tan and Cabato, 2023), or people in Kenya work for less than $2 per hour to make ChatGPT less toxic (Perrigo, 2023). These data workers are often described as “humans in the loop” or “ghost workers” (Gray and Suri, 2019). They work on online labour platforms, such as Amazon Mechanical Turk and Clickworker, which subdivide digital tasks into smaller units of work to be completed by a ‘crowd’ of remote workers. Much like application-based gig workers, crowdworkers do not receive formal employment contracts or support. Furthermore, the tasks performed are often fragments of much larger projects, about which the crowdworkers lack any information.

Locating Online Crowdwork in the Indian Context

By 2028, India is expected to have almost a million crowdworkers (Lochab and Roy, 2024). Since the contingent nature of online crowdwork amplifies existing inequalities, it is essential to understand how caste discrimination shapes labour realities for crowdworkers from oppressed caste backgrounds. AI-powered digital products rely on a global supply chain starting with low-wage data preparation in the Global South. This process spans markets in multiple countries, and transforms raw data into assets for maximum profit accumulation, producing racialised subjects. However, in India, this international division of labour, together with uneven capitalist development consists of a sub-stratification based on caste, which in itself is a division of labourers (Ambedkar, 1944).

Online crowdwork attracts workers with flexible working arrangements. However, this corresponds with a wider unemployment crisis in the country, that limits the choices of work available to them. For workers from historically marginalised castes, any technological employment signifies upward social mobility through improved working conditions, greater regularity, and higher wages compared to traditional caste-based occupations. Moreover, the perceived dignity of ICT work creates an aspirational appeal. Feminist analyses show that women perform crowdwork by balancing it with non-negotiable domestic duties to uphold household “honour” (Gurumurthy et al., 2021). These standards of respectability are bound to vary across caste lines.

Despite being referred to as “remote”, online crowdwork requires physical hardware- computers, tablets, or smartphones, on top of stable internet, electricity, and a workspace. These necessities effectively exclude half the world's population (GSMA, 2018; ITU, 2017). Studies show that ICTs intensify dominant caste exclusion in India, creating new barriers for subaltern communities and hindering their socio-economic advancement in both formal and informal sectors (Kamath, 2023). Crowdwork also creates hierarchies between part-time freelancers and a growing cohort of full-time workers, particularly in India, where most crowdworkers are situated in rural and semi-rural areas (NASSCOM, 2021). These workers, lacking access to the required ICTs, are forced to find work at AI, cloud farming and other lower-end IT firms. Given that caste shapes the spatial access of labour in such sites, it follows that entry into ICT-adjacent professions such as crowdwork is not entirely straightforward.

Working Conditions in Online Crowdwork and Their Relation to Caste

Among online crowdworkers, the labour of data labelers and content moderators in particular makes seamlessness of social media platforms and generative-AI tools possible. They have to work on a huge volume of data and are often exposed to violent imagery, pornography, and other graphic material (Abedin, 2025). Indian moderators reported suffering from severe psychological trauma, including PTSD, as a result of working on disturbing content. On its own, this practice can be categorised as forced labour and intentional infliction of mental harm. Researchers describe the work of women crowdworkers who view harmful videos of rape, child abuse and deepfake porn as a form of “visual sexual violation” (Samal, 2026). Similarly, for a crowdworker from a marginalised caste background, exposure to such content mirrors caste-based violence. Further, it is possible for these workers to come across media including casteist slurs or caste atrocities such as honour killings, of which there is a lot on the internet due to the frequency of these incidents. The same has the potential to be traumatising to an individual worker. On a societal scale, this defining feature of online crowdwork can be described as a form of computerised “manual scavenging” as it involves the cleaning of highly violent and toxic materials to ensure the smooth working of the digital environs and that it is sanitised for the end users.

Non-transparent algorithms are used to coordinate online crowdwork i.e. to match workers and requesters and facilitate transactions between them. These algorithms are also used to evaluate and monitor the workers. The design of digital platforms makes it difficult for crowdworkers to communicate and resolve disputes with the requesters, exacerbating the unequal power dynamic between the two parties. The logic of caste-based discrimination may not play out directly in such a scenario, but it can shape the nature of the transaction between a requester and a crowdworker. The former feels entitled to the labour power of the latter, who is invisible and has no means of negotiation in the event of any dispute. Additionally, crowdworkers tend to avoid such situations as failure to complete an assigned task negatively impacts their evaluation. This takes the form of a rating, which the crowdworkers have to keep improving to receive more tasks on the platform or to become eligible for higher paying ones (Gupta et al., 2014). The presence of such gamified reputation systems not only increases the level of uncertainty for the workers, but also creates a hierarchy through competition, akin to Ambedkar’s conceptualisation of “graded inequality”. While this may not look like the role-based hierarchies in IT workplaces, it echoes the narrative that such professions are inherently “meritocratic and free from caste-based inequalities” (Shakthi, 2023). These systems also create a strict regime of reward and punishment. The crowdworkers must balance accepting or rejecting a task, and the prospect of qualifying for better pay. For example, some workers may get suspended for accepting a task not “meant” for their skill level. On the other hand, the only way to earn more wages is to complete as many tasks as possible. The platform’s arbitrariness, together with the myth of meritocracy perpetuates voluntary servitude. The elusive approval ratings and qualifications keep the already caste-conditioned workforce toiling endlessly.

Recent studies have found that the labour of Adivasi women is channelled into the lowest-paid and most monotonous segments of the digital economy, raising concerns of “epistemological exploitation” (Ranjan, 2025). The images and texts worked on by Adivasi women labelers carry their knowledge. However, it is engineers in the West that set the overall framework of meaning for the data. Researchers believe that AI language models, shaped thus, reproduce epistemic hierarchies by erasing, misrepresenting and appropriating indigenous categories. These knowledge systems have long been treated as inferior by colonial and imperialist powers, while they also benefit directly from the expropriation of the same. This form of epistemic marginalisation happens to oppressed caste workers as well through the historically constructed dichotomy between “theoretical brahmins and empirical shudras” by relegating groups like Dalits into occupations that have no possibility of creating or disseminating knowledge (Guru, 2017). Though developed for Indian social science research, this dichotomy closely reflects the organisation of work in online crowdworking.

Political Mobilisation in Crowdwork and Caste

Political mobilisation has been gaining ground in platform work, as seen with the 2026 New Year’s Eve strike (Inamdar, 2026). However, uncertainty of employment, gender and caste are causes of inhibition and shape organised resistance. In gig, service and platform workers’ unions, it was found that dominant caste workers are more likely to put forth their grievances about being mistreated by managers. In contrast, those from scheduled castes or migrant communities tended to be less vocal (Kalia, 2025). This could be due to the fact that not all workers see “dignity of labour” as something to be demanded and fought for. Another observation suggests that direct organisational methods such as strikes were mostly called by workers who owned land, or some other stable property. A blocked account or even suspension is less threatening to such workers than to those who have no other means of sustenance to rely on. Online crowdwork complicates these dynamics, by lacking a concretely defined ‘workplace’ upon which crowdworkers can form a class consciousness and undertake traditional forms of organising. While such methods are able to target tangible entities like the state and company owners, the presence of algorithmic management on digital platforms makes conventional strategies inadequate. However, recent research has shown that platform workers adopt several tactics to try to improve their working conditions. First, they cultivate “algorithmic subjectivities,” and “hack” the platform’s algorithms (Stringhi, 2022) to earn more, and exert greater control over their schedules. Second, online communities such as Telegram groups for crowdwork platforms help them build social connections. Consequently, it is important to investigate the role caste plays as a potential barrier to be overcome on the way to crowdworkers’ solidarity. It is likely that a crowdworker with a high degree of comfort and experience with a platform will be in a position to experiment with its algorithms and affordances. As mentioned earlier, caste can shape the level of uncertainty for a crowdworker. If they’re from an oppressed caste, the worker would feel less confident in adopting these strategies due to the fear of suspension. Similarly, the pattern regarding the dominant caste workers being more outspoken than their oppressed caste colleagues can be applied to the online communities as well. However, the anonymity afforded by the same might enable the oppressed caste workers to speak up more.

Conclusion

According to technology policy analysts, it is difficult to assess the size of the Indian online crowdwork industry. The Periodic Labour Force Survey includes “data entry clerks” but fails to account for modern data workers making them “statistically invisible” (Arya, 2025). No studies have examined the dimensions of caste in crowdwork, due to the lack of publicly available information. However, this does not rule out caste’s function as an embedded form of discrimination in platforms such as AMT or its influence on the already precarious position of crowdworkers. This article has employed caste as an analytical category to frame digital platforms as environments where caste-based power dynamics and social hierarchies are actively reproduced. This lies in contradiction with the dominant narrative of technology positioning itself as innovative and meritorious and with globalisation’s promise of alleviating caste-based oppression. The reality however, is of a system of subjugation that continues to shape the organisation of work in India.

It is clear that AI and online crowdwork benefit from the divisions enabled by caste as the labour power of oppressed caste and tribal workers can be expropriated more efficiently. These workers, much like the ones in Kenya and the Phillipines, occupy the lowest rungs of the supply chains that make the digital technologies of today possible. More broadly, caste shapes aspirations towards a more “humane” form of work, the everyday subjective experiences, and possibilities for formal employment. Much like in the realm of app-based gig work, it is possible for caste to inform the class consciousness among crowdworkers and the extent of political mobilisation. All of these effects must be interrogated in detail to develop a more holistic understanding of crowdwork and caste in future of work discourse.

Instagram: eyes_ofa_bluedog