Encoding Marginality: AI Supply Chains and Linguistic Data Work in India

India is a country marked by a considerable degree of linguistic fragmentation. This is particularly apparent in rural areas, where literacy rates are low and people usually speak only local dialects and a major regional language. Rural communities, mainly composed of socially marginalized groups, face varying degrees of exclusion from India's economic and social structures, with limited access to education and formal employment. Against this backdrop, recent efforts have emerged as attempts to encode India's linguistic diversity into AI systems. Rather than substantively addressing structural barriers to education and social inclusion, however, these initiatives primarily aim to integrate Indian languages into small and large language models to enable cross-linguistic communication, positioning AI as a means of delivering digital services to disadvantaged groups. At the same time, these efforts are linked to broader strategies of economic integration aimed at incorporating segments of the informal economy into data-driven digital markets.

This transformation is driven by a complex ecosystem of startups, public and private platforms, research institutions, and subcontracting firms, which depends on the labor of marginalized communities for the production and annotation of linguistic data. Unlike models trained on pre-existing data, this form of AI development is, in fact, based on the active construction of datasets by members of these communities. The resulting AI systems are then redeployed within those same social contexts, with the primary aim of integrating them into wider circuits of value production, especially—though not exclusively—through the deployment of technically formalized microfinance. This points to an emerging process of technological expansion into areas that had thus far remained relatively “unencoded” due to data scarcity, and also helps explain India's ambition to position itself strategically as a leader in AI development across the Global South. The underlying assumption is that AI can streamline information flows across a culturally and socially fragmented landscape, opening new avenues for the expansion of formal markets into previously marginal spaces.

Apart from questioning whether this transformation offers fair working conditions and meaningful economic prospects for data workers, the broader concern is whether AI-based inclusion can effectively improve the conditions of marginalized communities, or whether it will instead reproduce and amplify the negative effects commonly associated with microfinance—such as debt entrapment and the erosion of social ties. Both issues are analyzed through reports from government bodies, startups, and research institutes, as well as semi-structured interviews with researchers involved in this process of linguistic codification.