# Whose History Gets Written? Documentation Gaps, Caste Memory, and the Limits of AI

## An Open Letter

I write this as someone with a personal stake in this question. Decades ago, I was among the students who took part in a hunger strike protesting a caste atrocity in Andhra Pradesh — the kind of event that, when it happened, was lived and witnessed directly by people who had no institutional power to write their own account of it into the historical record.

Recently, I had a long conversation with an AI system while trying to get information about that period. It surfaced a widely cited detail about the community of the victims that does not match what I know firsthand, from being there. That mismatch is what prompted this piece — not because one detail on one page is, by itself, a crisis, but because it's a small, concrete example of a much larger problem.

## The Documentation Gap

Here is the structural issue: written history is not a neutral mirror of what happened. It is written by whoever had the literacy, the access, the language skills, the institutional standing, or the media platform to write it down — and then indexed, copied, and cited by everyone who comes after.

When a caste atrocity happens against a community with limited access to journalism, courts, and publishing, the account that survives in searchable, citable form is often produced by outsiders to that community — however well-intentioned. Details get simplified. Distinctions that matter enormously to the people who lived through an event (which specific community was targeted, for instance) can get flattened or mislabeled by those documenting it from the outside. Once that account is published, it tends to get treated as settled, simply because no one has publicly and visibly disputed it — not because it was ever verified against the people who actually know.

This is not unique to any one country, but it likely affects underrepresented and marginalized communities' histories more than most, precisely because the gap between who has lived knowledge and who has the power to publish is widest there.

## Where AI Fits In

Tools like AI assistants make this problem more visible, and potentially worse, in a specific way: they search and summarize what's already published, and they can present that summary with a tone of authority that a raw search result never quite has. "No conflicting source found" starts to feel like "this is confirmed," when it may really just mean "the account that would correct this was never written down anywhere searchable."

I do not believe this happens because of some coordinated effort to plant false information. It happens for a more mundane and, in some ways, more troubling reason: silence in the documentation record gets mistaken for consensus. The 99 people who know the truth and never publish it lose, by default, to the 1 person who did — not because that person is more right, but because they're the only one left a trace.

## What I'm Asking For

This is not a call to distrust AI tools wholesale, any more than it would make sense to distrust libraries or search engines wholesale. It's a call for more honesty in how these tools present uncertain or contested information, especially around historical and social topics where documentation is known to be thin, biased, or shaped by unequal access to power. Concretely:

- **Flag documentation gaps explicitly.** When a claim rests on a single source, or on an absence of disagreement rather than verified consensus, say so — don't present it with the same confidence as an established fact.

- **Weight firsthand and community accounts appropriately.** They are not automatically right, but they are not weaker evidence than a secondary or tertiary published source either — especially for events where the affected community had little power to shape the written record.

- **Build in humility for exactly this class of question** — caste, regional history, marginalized communities — where the gap between lived truth and published record is likely to be largest.

More broadly, I believe this deserves attention beyond any one company's product design. As AI systems become a primary way people — including the next generation of students, researchers, and policymakers — access information about India's social history, the risk is real: that whichever version of history happened to get written down first, by whoever had the access to write it, becomes the permanent, unquestioned version, repeated by every tool that comes after.

History belongs to the people who lived it, not only to those who had the pen.

Truthfully Yours,

Sarabhaiah Polakam