Why Vedic Astrology Was the Perfect Test for Honest AI

“I did not choose astrology because it was easy to make an AI sound wise about it. I chose it because it was brutally easy to catch an AI being wrong about it.”

---

Why This Domain, of All Domains?

When I decided to build a system that would prove AI could be honest—that it could compute instead of guess—I could have picked any field. I picked Jyotish, Vedic astrology, and people assume that was a spiritual choice. It was partly a practical one.

Jyotish is almost perfectly designed to expose a lying machine. The facts are exact and instantly checkable, and the entire market is built on faking exactly those facts. If you want to test whether an AI is grounded or merely fluent, you could hardly design a cleaner laboratory.

Jyotish Is Computation, Not Fortune-Telling

Start with what the tradition actually is, because the caricature gets in the way. Jyoti means light; Jyotish is the science of cosmic light and timing. And its foundation is not intuition—it is arithmetic. I have written about this at length in Understanding Jyotish from my Perspective and Jyotish Beyond Prediction, but the core point is simple.

Jyotish (Vedic astrology) concept

An ancient Indian computational science that determines the positions of celestial bodies at a given moment and interprets their significance. Its computational half—establishing where the planets are—is pure astronomy and mathematics; its interpretive half is the layer of meaning built on top of those exact positions.

A birth chart is a precise astronomical fact. The Moon occupied one exact sidereal longitude at the instant you were born. That is not a matter of belief; it is a calculation against centuries of observation. Our tradition treated this seriously enough to build mathematics itself as a form of inquiry—ganita—and to refine astronomical models over millennia. The interpretation is where philosophy enters. The positions are where the mathematics lives.

That split—exact computation underneath, interpretation on top—is exactly the computation-first architecture I believe honest AI requires. Jyotish did not just tolerate that separation. It demanded it.

The Domain That Punishes Hallucination Instantly

Here is what makes it the perfect test. When an AI fabricates a planetary position, the error is immediately checkable. Compute the real chart and the lie is exposed in seconds—the planet is either in that sign or it is not.

Compare that to the domains where hallucination usually hides. When a chatbot invents a plausible-sounding historical detail or a soft business platitude, who checks? The error slides by. But a fabricated chart collides with arithmetic instantly. There is nowhere for a confident guess to hide.

Astrology is unusual: it looks like the softest, most subjective subject in the world, and underneath it is as hard and checkable as an ephemeris table. That contradiction is exactly what makes it a good test.

If you can build an AI that never fabricates a chart—in a field where fabrication is this easy to catch—you have proven something that transfers directly to fields where fabrication is far harder to catch, and far more dangerous.

And Yet It Is the Most-Faked Domain of All

The cruel irony is that astrology is simultaneously the easiest place to catch a lie and the most heavily lied-about subject online. I unpack this fully in The Machine That Refused to Guess, but the shape is familiar: enormous content farms publishing generic articles that never compute your chart; apps reading only your Sun sign; and now chatbots hallucinating whole charts in flawless prose.

An entire market had quietly agreed not to do the one checkable thing—compute the actual chart—because computing it is harder than writing beautifully about it. That gap was the opportunity. Not a spiritual gap. An engineering-honesty gap.

What Building Here Forced Me to Get Right

Choosing a domain this unforgiving imposed a discipline I could not have faked my way around. It forced the hard computation–interpretation boundary to be real, not decorative:

• The engine had to be deterministic—same birth details, byte-identical chart, every time.
• The positions had to be computed from real ephemeris data, not approximated.
• The language model had to be fenced off from the numbers entirely—free to interpret, forbidden to invent.

In a soft domain I might have let those slip. Here, any slip was instantly visible. The unforgiving test made the system honest by leaving no room to be anything else.

The Tradition Already Demanded This

What moves me most is that none of this discipline is foreign to Jyotish. It is native to it.

Classical Indian astronomy distinguished siddhanta—the computed theoretical model—from drik, direct observation, and insisted the two be reconciled. When a model’s predictions drifted from the observed sky, the model was corrected. That is not mysticism. That is empirical computation, checked against reality, refined over centuries. It is the same instinct as a modern engineer validating a system against ground truth.

And underneath sat the deeper concern of Indian epistemology: pramana, the valid means of knowing, scrutinised for how each could fail. The question “how do you know this is true?” was load-bearing in this tradition long before hallucinating machines made it urgent again.

What Generalises

So astrology was the laboratory, but the result is not about astrology. If you can make an AI honest in a domain where the facts are this exact and the temptation to fake them is this strong, you have a template for every domain where truth is computable and the stakes are higher—the ones I map out here.

I chose the test that was easiest to fail in public. Passing it in the open, in a field full of confident fabrication, was the point. You can see the result at eternalevals.com—and, in the spirit of the tradition, check it against the sky yourself.

---