Will AI abolish India’s caste system or encode it?

In 2027, India will conduct an exercise of almost unimaginable scale. Hundreds of millions of people will be counted in what the government describes as the country’s first fully digital census. Alongside information about education, migration, fertility and employment, enumerators will collect another kind of information from every individual: caste. It will be the first Indian census since independence to enumerate caste across the entire population. Scheduled Castes and Scheduled Tribes (historically disadvantaged people) have long been counted, but most other caste identities have not appeared in the national census since 1931. The decision should give policymakers a clearer picture of social inequality. Yet it also exposes one of modern India’s deepest paradoxes. The state must count caste in order to correct inequalities produced by caste. To create a society in which inherited identity matters less, it must first establish precisely who belongs to which inherited group. At the same time, India is entering the age of artificial intelligence. If automation transformed work and every citizen received a universal basic income, or UBI, could these changes finally dissolve the material foundations of caste? Or would AI simply automate one of the world’s oldest systems of social classification? Varna is not jati Any answer must begin with a distinction that the English word “caste” frequently obscures. Varna refers to the familiar fourfold order of Brahmins, Kshatriyas, Vaishyas and Shudras, with some communities historically placed outside the scheme. It is a normative and philosophical map of society, described in classical texts and interpreted differently across time. Jati refers to the thousands of hereditary, localized, and usually endogamous communities that have structured actual social life. Jati has influenced marriage, occupation, kinship, status, diet, neighborhood, and access to social networks. It is far more complex than the four varnas and cannot be neatly derived from them. Can these thousands of jatis nevertheless be grouped under the four varnas? Only very loosely. Some jatis have traditionally identified themselves, or been identified by others, with a particular varna. But the fit is neither comprehensive nor consistent. A community may claim Kshatriya standing in one historical setting while occupying a different position elsewhere. Local rankings can be disputed, and many Dalit communities were historically treated as avarna, outside the fourfold order altogether. Ideal vs. Reality: The first column describes a normative ideal, the second a lived practice. In history, the ideal largely became hereditary—which is how varna became caste. Some classical texts describe the fourfold order in terms of aptitude and action rather than just birth; the Bhagavad Gita famously explains that it arises from guna and karma, or quality and work. But in lived history, that ideal was converted into heredity: aptitude was assumed to follow birth, and difference hardened into hierarchy. A map of human diversity became a cage of inheritance. In public discussion, varna and jati are often collapsed into “caste,” which makes caste appear to be a single ancient system when it has repeatedly been reconstructed by political and administrative forces. Even the English word is a foreign import: it derives from the Portuguese casta (lineage, breed), which early European traders applied to Indian social groups, folding varna and jati together into a single category. Independent India abolished untouchability and prohibited caste discrimination. It also created reservations and targeted programs for historically disadvantaged groups. Applicants for certain educational places, public positions, or benefits may therefore need certificates establishing membership in a Scheduled Caste, Scheduled Tribe or Other Backward Class. This recognition is intended to compensate for accumulated disadvantage, not to endorse hierarchy. Nevertheless, an identity the Constitution seeks to make less consequential must remain visible to the agencies remedying its effects. What AI and basic income could change Caste was never only a belief system. It was embedded in inherited occupations and unequal access to land, education, credit, and social support. Leaving a caste-associated occupation could mean losing an entire network. AI and UBI could weaken some of these bonds. Automation may reduce the importance of occupations historically associated with particular communities. Digital education and remote work can connect people to skills and employers beyond caste networks. A guaranteed income could give workers greater freedom to reject degrading work or leave places where dominant groups control economic opportunity. India’s 2016–17 Economic Survey called UBI a “radical new vision,” partly because targeted welfare often failed to reach its intended recipients. A universal payment would recognize the citizen before the category. But universality carries its own problems. A payment made to everyone can be used as an argument for doing nothing more for those who have been specifically wronged: if every citizen receives the same floor, why maintain reservations, scholarships, or targeted programs? Used in that spirit, UBI would not complete the work of social justice but quietly retire it. A universal floor must supplement remedial programs, not replace them. Used well, the change could be quietly revolutionary. If survival no longer depended on accepting an inherited place in the division of labor, one of caste’s practical foundations would weaken. AI might make traditional occupational boundaries obsolete, while UBI supplied the minimum economic security needed to cross them. Caste after work Social reformer B.R. Ambedkar understood caste not simply as a division of labor but as a division of laborers. He identified endogamy, marriage within one’s group, as central to its reproduction. This helps explain why modernization has weakened some caste practices but has not eliminated caste. Marriage remains an especially resilient boundary. The India Human Development Survey estimated that approximately 95% of marriages occur within caste. A professional may work for a global company and participate in an international culture, yet still be expected to marry within a particular community. Caste prejudice is an attitude, but caste is also a structure. Even when individuals no longer consciously “look down” on others, marriage patterns, family networks, property and inherited opportunity can continue reproducing hierarchy. A basic income would not erase inherited differences in property, schooling, family connections, or political influence. Caste can survive caste-based occupations because it also functions through identity, memory and social reproduction. A post-work economy might even strengthen inherited identity. If automation disrupts occupational communities, religion, ethnicity and caste may become more important as alternative sources of belonging. Material abundance does not guarantee social openness. When caste becomes data There is another reason to resist technological optimism. AI learns from the society that produces its data. An algorithm does not need a box labeled “caste” to discriminate on the basis of caste. It can infer social background from surnames, postal codes, language, schools, occupations and social networks. Hiring and lending systems trained on historical decisions may convert accumulated advantage into apparent merit and historically produced poverty into evidence of individual risk. In some domains, the automation of caste is already visible. Matrimonial platforms routinely allow users to filter prospective partners by caste and sub-caste, and recommendation systems then prioritize matches who resemble past choices. What was once mediated by family and community networks can now be encoded into search filters and default settings. The 2027 caste census makes this danger especially relevant. Better information can reveal disadvantages, improve policy, and clarify who has been excluded from opportunity. But as caste becomes more precisely recorded and linked to other digital systems, it may also become more machine-readable. The same data that enable social justice could, without strong safeguards, enable profiling, political targeting, or automated discrimination. AI may make caste economically obsolete while making it digitally legible. Social transformation This doesn’t mean AI and a universal basic income are irrelevant to caste. On the contrary, material security matters profoundly. A person who can obtain education and live without the permission of a dominant community has acquired real freedom. Technology can expand choices and UBI could provide those choices with an economic foundation. But neither can it substitute for social transformation. Ending caste also requires equal institutions, protection against discrimination, greater social interaction, and a weakening of endogamy. It requires people to stop treating inherited identity as a measure of human worth. No algorithm can perform that moral work on society’s behalf. The central question is therefore not whether India should count caste or refuse to see it. A state that ignores caste may merely ignore inequality. The challenge is to use classification provisionally, transparently and for clearly defined remedial purposes, without allowing it to become a permanent digital destiny. Census 2027 will produce the most detailed national picture of caste in independent India. The AI systems emerging alongside it will help determine how that picture is used. They could expose exclusion, widen access and loosen the relationship between birth and livelihood. They could also convert inherited hierarchies into data, and then turn that data into automated judgments. AI and a UBI may help create the conditions in which caste can decline. But they will not abolish caste on their own. The decisive struggle will remain human: whether Indians use new technologies to escape inherited categories or teach their machines to preserve them.