Determinism Meets Dharma: What Ancient Indian Computation Teaches Modern AI

“We are taught to imagine a rational, computing West and a mystical, contemplating East. Spend an hour with Panini or an Indian astronomical treatise and the caricature collapses. This was a civilization that computed and contemplated with the same breath.”

---

The Caricature We Should Drop

Building a computational engine on top of an ancient Indian science taught me something I did not expect—not about astrology, but about a habit of thought so common we rarely notice it. We split the world into a rational, computing West and a mystical, intuitive East. It is a tidy story, and it is false.

The tradition I was drawing on was ferociously computational. It also never divorced that computation from meaning, ethics, or consciousness. That combination—rigor and reverence, held together—turns out to be exactly what modern AI is missing.

India’s Computational Tradition

Consider the evidence, none of it obscure.

Panini’s grammar. Around two and a half thousand years ago, Panini described the Sanskrit language as a formal system of roughly four thousand rules—a generative, almost algorithmic machine for producing valid language. Linguists and computer scientists study it precisely because it prefigures formal systems and generative grammar. This is source code, written in the fifth century BCE.

Ganita. Mathematics was pursued not as dry calculation but as a mode of inquiry into reality, as I explored in Ganita: Mathematics as Consciousness Exploration. Zero, place-value notation, the beginnings of the infinite—developed by a civilization that saw no wall between computing the world and contemplating it.

Siddhanta astronomy. Astronomical treatises built precise computational models of celestial motion, refined across centuries against observation. Jyotish, which I called the perfect test for honest AI, sits on this computational bedrock.

A civilization that gave the world zero, a formal grammar of language, and observation-corrected astronomical models was not choosing intuition over computation. It was refusing the choice.

This is the Indian knowledge system in its actual character: not a haze of mysticism, but a structured, computational inquiry that kept meaning in the room.

Pramana: The Original Hallucination Guardrail

Here is where it speaks most directly to our moment. Indian philosophy was obsessed, for millennia, with a question that has suddenly become the central problem of AI: how do you know that what you know is true?

Pramana concept

In Indian epistemology, a valid means of knowledge—the accepted ways a claim can be established as true, such as perception (pratyaksha), inference (anumana), and reliable testimony (shabda). Each was analysed rigorously for how it could fail and mislead.

The schools argued endlessly over which pramanas were valid and how each could deceive. This was not idle. It was a civilization building guardrails against false knowledge—an epistemic immune system. And that is exactly the discipline a hallucinating machine lacks. A language model has no pramana. It cannot distinguish what it has genuinely established from what merely sounds established. It has fluency without a theory of how it knows.

Everything I have argued about computation-first AI is, in a sense, a crude modern pramana: a rule about which claims may be trusted (the computed ones) and which may not (the guessed ones). The tradition was here first, and thought about it far more deeply.

Siddhanta Corrected by Drik

There is one more piece, and it is the most engineering-like of all. Indian astronomers distinguished siddhanta, the computed model, from drik, direct observation—and insisted the two be reconciled. When the model drifted from the observed sky, the model was corrected.

That is determinism meeting reality. A precise computation, held accountable to what is actually there, refined when the two diverge. Strip away the Sanskrit and it is the loop every serious engineer runs: model, measure, correct. The tradition institutionalised the humility that computation must answer to observation—the very humility evaluation demands of AI.

Dharma as Constraint

And then there is the word the computing story usually leaves out: dharma. Because this tradition never treated knowledge as ethically weightless. Knowing carried obligation. How you used a computation, whom it served, whether it was honest—these were not separate from the mathematics; they were part of the same fabric of right action.

An honest AI system—one disciplined to compute rather than fabricate—is, in a small and literal way, a system built with dharma in it. The refusal to deceive is encoded in the architecture. That is not mysticism bolted onto engineering. It is ethics expressed as engineering.

What a Civilization That Computed and Contemplated Offers AI

I did not set out to make a civilizational argument. I set out to stop a machine from lying about birth charts. But building on this foundation kept revealing that the questions I thought were new—grounding, evaluation, honest constraint—were old questions this tradition had already sat with, seriously, for a very long time.

The West gave modern AI its extraordinary machinery. What it is now scrambling for, under names like alignment and grounding and evaluation, is the other half—a theory of valid knowledge, a habit of correcting computation against reality, and an insistence that power be bound to right use. India worked on exactly that half for millennia. I think we would be foolish not to listen.

Determinism gave us the power to compute the world. Dharma reminds us that computing it is not the same as being honest about it—and that the second is the part that matters. The engine I built, Eternal Evals, is one small attempt to hold both at once. The tradition it draws on held them together all along.

---