William Adam, the algorithm, and a question India keeps deferring
India in 2026 is doing something it has not quite managed before: it is arriving. The G20 presidency, the moon landing, the Mars orbiter, a diaspora running companies and universities and governments on four continents — there is a palpable sense that India’s civilisational moment is approaching, and that the world is beginning to take note. The country wants a seat at the global table of the new world being assembled around artificial intelligence, and there are serious reasons to think it belongs there.
And then, in the same week, Tamil Nadu refuses to implement the National Education Policy’s mother-tongue clause. Karnataka reverses a ruling on primary-school instruction after protests. A parliamentary committee cannot agree on what “mother tongue” means without immediately raising the question of whose mother tongue, and whether the answer will eventually mean Hindi. The debate that erupts is not new. It is, in its essentials, the same debate India has been having since 1835, when a colonial government faced a budget decision about education and chose, against the advice of the only official who had actually looked at the evidence, a path that India is still trying to correct almost two centuries later.
The official who was ignored was a Scottish Baptist missionary named William Adam. The path he proposed is the subject of this essay. But this is not a history essay. The reason Adam matters in 2026 is that the AI revolution is now putting his question with a new and terrifying urgency: in what language should Indian children be taught to think? And this time, the decision will not take two centuries to play out. It will compound in a decade.
Adam arrived in Bengal as a missionary in 1818 and promptly converted to Brahmo monotheism after a single conversation with Ram Mohan Roy, which tells you something about the quality of his attention. The colonial administration, preparing to redirect education funds toward English-medium instruction, commissioned him first to survey what already existed. He spent years doing it properly, village by village, district by district, and his three reports, submitted between 1835 and 1838, constitute the most careful empirical study of indigenous Indian schooling produced in the entire colonial period.
The central finding was a number: one lakh. A hundred thousand village schools in Bengal and Bihar alone. The colonial imagination had conjured a darkness; Adam found a working system. The pathshala, the Sanskrit-medium Hindu school, was typically a single room, or no room at all: a teacher, a tree, a patch of swept earth. Students wrote on sand with their fingers before they touched a slate. They memorised first texts, multiplication tables, commercial formulae, then understood. A boy might spend three to five years in a pathshala and emerge able to keep accounts, read a land deed, calculate crop yields, and navigate the contractual demands of the rural economy. The maktab, its Islamic counterpart attached to a mosque, ran on similar lines: Quranic recitation at the core, practical literacy in Persian and Urdu woven through. Many villages had both.
The guru who ran the pathshala was often the only literate person in the village, and his authority was social as much as pedagogical. He knew the land records, the family histories, the weight of the last harvest. What he transmitted was not merely literacy but situated knowledge, the kind that is useful because it is embedded in the world it describes. At roughly eleven students per school, the hundred thousand pathshalas were reaching something like one in every thirty or forty people in the region at any given time. Adam was clear-eyed about the system’s limits: upper-caste boys predominated, girls were almost entirely excluded, quality was wildly uneven. But the density was extraordinary, and the schools were alive because they served the communities that housed them.
His proposal followed directly from his findings. Do not replace this network, build on it. Modest state grants tied to minimum standards. Printed vernacular textbooks to replace the manuscripts students copied by hand, so that a teacher in Murshidabad and a teacher in Patna were working from the same material. Basic teacher training. An expanded curriculum that added subjects the changing economy actually demanded. It was, in contemporary terms, a quality-improvement programme for an existing distributed system. The infrastructure was already there, rooted in every village, carrying local legitimacy no new institution could quickly acquire. It needed investment, not demolition.
Macaulay, who had never been within a hundred miles of a pathshala, wrote his Minute in the same year. He had read neither Sanskrit nor Arabic. He acknowledged this, and proceeded: “A single shelf of a good European library is worth the whole native literature of India and Arabia.” The Company chose Macaulay. The hundred thousand schools were left to wither, unfunded, unsupported, overtaken by institutions that served the administrative needs of empire and left everyone else to manage as best they could.
The cost was not abstract. It was structural, and it accumulated.
Adam’s schools were doing something the English-medium model could never do at scale: they were teaching children to think in the language they already thought in, about the world they already inhabited. The Japan comparison is instructive not as flattery but as mechanism. The Meiji Restoration built mass literacy on the back of exactly such a pre-existing village school system, the terakoya temple schools, informal and community-embedded, not unlike Adam’s pathshalas, and invested in them rather than dismantling them. Japan’s literacy rate approached ninety percent by 1900. India’s was five percent. That gap is not a gap in civilisational potential. It is a gap in policy, and its consequences are still running.
What Macaulay’s model produced was a pipeline, not a field. It generated a small class of English-educated administrators useful to the colonial state, and left the vernacular world, its schools, its texts, its intellectual traditions, its ways of organising knowledge, to decline without support. By the time independence came, the regional languages were fighting not for prestige but for institutional survival: for the right to be languages of administration, education, and law after a century of systematic marginalisation. The Hindi-versus-south explosion of 1965 — the agitations, the self-immolations, Chinnasamy burning himself to death in Madurai rather than accept a future in which his language was second-class in his own country — was not a failure of Indian unity. It was the delayed detonation of a charge Macaulay had laid in 1835.
In an Adam-trajectory India, the regional languages would have arrived at independence as going concerns: thick with professional and literary tradition, institutionally confident, needing no one’s permission to be serious. The negotiation over a national language would still have been necessary in a country of this scale and diversity it always would be. But it would have been a negotiation between equals. Tamil Nadu’s suspicion of Delhi’s language policy — the hair-trigger watchfulness with which the south reads every education reform from the centre — would not have required one hundred and ninety years of accumulated grievance to produce it.
And what of the India of 2026, in that alternative history? The English-vernacular divide would still exist — inequality is stubborn, and history cannot be fully undone by a single policy fork. But the toll booth that English currently places at every entry point to professional life — the civil service, the higher judiciary, the elite university, the corporate boardroom — would not have had two centuries to become load-bearing. When the NEP proposed mother-tongue instruction in 2020, it would have been reinforcing a deep tradition, not attempting a correction so late it reads as an admission of failure.
The reason this history matters now, with new urgency, is that the algorithm has arrived. And it has arrived speaking English. The large language models reshaping how knowledge is produced, communicated, and accessed, were trained overwhelmingly on English text. The disparity between English and any Indian language in the training data of the major AI systems is not marginal, it is orders of magnitude. The models powering legal research tools, medical diagnostic aids, educational tutors, and administrative interfaces in 2026 handle English with a fluency they cannot approach in Telugu or Odia or Maithili. This is not because those languages lack the capacity for precision or nuance. It is because they have less data. And they have less data because the decisions made in 1835 and compounded in every decade since have produced a digital endowment that no Indian language has yet been able to match.
Adam’s pathshalas transmitted situated knowledge — knowledge embedded in the community it served, delivered in the language the learner already spoke, calibrated to the world the learner already inhabited. That is precisely what a well-designed AI tutoring system could do, at scale, across every classroom in India. And it is precisely what an AI system trained on English and deployed without vernacular infrastructure cannot do. The child who understood perfectly but could not express in the required medium — the student in Bhojpuri-speaking Bihar who grasped the concept but stumbled on the language of the examination — was the human cost of Macaulay’s choice. AI can dissolve that bottleneck or deepen it. The outcome depends entirely on what gets built.
Here is the pessimistic case, stated plainly: an AI ecosystem built on English-language dominance will reinforce English-language dominance. Students who develop fluency with AI tools in English compound their advantage over students using less capable vernacular interfaces. Employers integrating AI into their workflows find that English-language capability accelerates faster than any other. The divide between English-medium and vernacular-medium education does not merely persist — it widens, driven not by policy contempt but by the self-reinforcing logic of training data and network effects. Macaulay did not need to hate Indian languages. He simply redirected resources. The algorithm does not need to hate them either.
And here is what Adam would recognise: this is exactly the choice he was asking the Company-state to make in 1838. You have a distributed network of intelligence, embedded in communities, working in the languages those communities actually use. You can invest in it — improve it, extend it, give it the tools it needs to serve its communities better. Or you can bypass it in favour of a system that serves someone else’s purposes and leaves the existing network to wither. The Company chose to bypass. India in 2026 is being asked the same question, and the window for choosing differently is narrower than it looks.
The optimistic case is equally real and more important. AI is potentially the most powerful vernacular tool in human history — if the will exists to build it that way. High-quality training datasets in Tamil and Bengali and Kannada, built as deliberate national infrastructure rather than as afterthoughts, would narrow the capability gap within years. AI translation is already good enough to make technical content accessible across language barriers in ways impossible a decade ago. The cognitive architecture that Adam understood — that a child learns most deeply in the language she dreams in, and that this foundation makes every subsequent layer of knowledge more secure, not less — can be honoured by AI at a scale and with a personalisation no pathshala teacher, however gifted, could achieve alone.
India’s twenty-two scheduled languages, its hundreds of millions of vernacular speakers, its literary and intellectual traditions of genuine depth and antiquity — these are not obstacles to the AI future. They are, if treated as Adam treated the hundred thousand schools, a distributed network of extraordinary richness waiting to be invested in. The question is not whether that investment is possible. It is whether it will be made, and whether it will be made in time.
A seat at the global table of the new world is not an honorary position. It comes with a question attached: what do you bring? A country that arrives at the AI century having built an education system that teaches children to think in the languages they dream in, and an AI infrastructure that serves those languages with serious investment rather than token gesture, arrives with something genuinely distinctive. It arrives having demonstrated that modernity and linguistic diversity are not in contradiction — that a child can reason deeply in Tamil and then, on that foundation, acquire English and Hindi and whatever else the world requires. That the distributed intelligence of a thousand-year literary tradition is an asset to be compounded, not a weight to be shed.
A country that arrives having outsourced its cognitive infrastructure to Anglophone AI — having allowed its vernacular languages to persist as cultural ornaments while surrendering the tools of knowledge production to a language most of its citizens learned as a second tongue — arrives as a consumer of someone else’s modernity. Gifted consumers, perhaps. But consumers nonetheless, dependent on tools they did not build, in a language that is not their own, for purposes defined elsewhere.
Adam’s report proposed something that sounds modest but is, in its implications, radical: meet communities where they are. Build on what they already know and how they already think. Do not demand that people abandon their cognitive home as the price of entry to knowledge. In 1838, acting on this would have required grants, textbooks, and a modest teacher-training programme. In 2026, it requires something larger — a national commitment to building AI in Indian languages as seriously as India has built space capability or pharmaceutical manufacturing: not as an aspiration but as an infrastructure project, funded and prioritised accordingly.
The hundred thousand pathshalas are gone. But the hundred million children in vernacular-medium schools are here. The twenty-two languages are here. The intellectual traditions are here, still living, still generative, still waiting for an Indian state that treats them as the foundation of its knowledge economy rather than as the sentimental residue of a pre-modern past.
Adam walked the villages, counted the schools, and came back with a proposal. The colonial government chose Macaulay instead. The choice compounded for two centuries into the toll booths and coaching industries and linguistic anxieties that still define Indian education today. India is now at the equivalent moment: a new technology, a new educational infrastructure being built in real time, and the same fundamental question. Does the system serve the people who actually live here, in the languages they actually think in? Or does it serve someone else’s purposes and call it progress?
The fork is here. The road is visible. What Adam knew is still true: the intelligence is already distributed. It just needs to be taken seriously.