Guide
Can ChatGPT do the I Ching?
A language model handed only a hexagram name has nothing to reason from but the name, so it free-associates — and many AI I Ching tools also cast with the wrong odds (a uniform 1/4 instead of the real 1/8, 3/8, 3/8, 1/8). The fix is not a better prompt; it is to compute the structure first and feed it to the model. And you should be able to check that structure yourself.
Where AI readings go wrong
One: the casting odds are often wrong. A real cast does not produce 6, 7, 8, 9 equally. Three coins give 1/8, 3/8, 3/8, 1/8; yarrow stalks give 1/16, 5/16, 7/16, 3/16. A tool that picks a line value uniformly at 1/4 each draws moving lines twice as often as any physical method — see three-coin probability.
Two: the interpretation free-associates. Given only a hexagram name, a bare model has no palace, no six-relative roles, and no use-spirit to anchor on, so it riffs on the imagery and can invent line texts or attributes. It produces fluent prose about whichever reading it drifts toward — sometimes a coherent answer to a question you did not ask.
What rule-grounding changes
YiGram works the other way around. A deterministic engine resolves the cast into concrete structure and passes it to the model as labelled facts; the model interprets given specifics rather than guessing them.
| Bare LLM oracle | Rule-grounded (YiGram) | |
|---|---|---|
| Casting odds | commonly a uniform 1/4 (wrong) | true 1/8, 3/8, 3/8, 1/8 |
| Cast structure | none — free-associates on the name | palace, najia branches, six relatives, shi/ying, moving lines computed |
| Use-spirit | not selected | chosen by question category, then read |
| Line texts | can be invented | anchored to the cast + public-domain Legge (1882) |
| Auditable? | no | yes — the rule tables are open-source |
You don’t have to trust us
The part that can be checked is the rule layer, so we publish it. YiGram’s static-layer najia rules — palace lookup, branch assignment, the six-relatives derivation, shi/ying positions, the use-spirit tree, and the casting probability — are open-source and versioned at github.com/AdrienSterling/yigram-najia-rules. The mechanical, structural rules follow the standard 京房纳甲 sequence and are not school-disputed; the use-spirit selection is a documented draft labelled unaudited_draft and cross-checked against 《卜筮正宗》. If an assignment is wrong, you can open an issue.
What it is, and is not
A YiGram reading is a structured aid for thinking through a decision — not a prediction of the future, and not a substitute for professional medical, legal, or financial advice. The engine makes the reading specific and inspectable; the decision, and its outcome, stay yours.
Questions
- Does ChatGPT do the I Ching correctly?
- A bare language model handed only a hexagram name has no cast structure to reason from, so it free-associates on the name and can invent line texts. Many AI I Ching tools also use the wrong casting odds (a uniform 1/4 instead of the real 1/8, 3/8, 3/8, 1/8). A rule-grounded engine that computes the structure first and feeds it to the model avoids both failures.
- What does rule-grounded I Ching AI mean?
- A deterministic engine computes the cast's concrete structure — palace, najia branches, six relatives, shi/ying, moving lines, and the use-spirit — and passes those as explicit facts to the model. The model interprets given specifics instead of guessing them.
- Can I verify the rules?
- Yes. YiGram's static-layer rule tables are open-source and versioned on GitHub, so you can check the palace, branch, six-relative, and use-spirit assignments against the classics yourself.
- Is an AI I Ching reading a prediction?
- No. It is a structured thinking aid for a decision, not a forecast, and it is not a substitute for medical, legal, or financial advice.