Can you trust AI with your metaphysics?
Not with the calculation. A general chatbot answering from text alone is not a reliable chart calculator, and one that looks like it is has produced something chart-shaped rather than computed. Language models are good at the other half, explaining a computed result clearly, and that half is worth having.
Two jobs, one of which AI cannot do
| Casting the chart | Explaining it | |
|---|---|---|
| What it is | Calendar and astronomical calculation | Turning symbols into language |
| Right answer exists? | One, once the conventions are fixed | No, many good ones |
| Same input, same output? | Must be | Need not be |
| Suited to a language model? | No | Yes |
A language model predicts plausible text. Given a birth date it will produce four pillars that look correct, because it has seen thousands of charts. It has not performed the calendar conversion. Ask twice and you can get two different charts, and neither will be flagged as uncertain.
Fix the birth data and the conventions, and a chart has one reproducible answer. That is a narrower claim than saying there is one true chart. Traditions genuinely differ on where the day turns and which time to use, and Meridian's position is to declare its choices rather than pretend the question is settled.
Why a general chatbot is not enough
Asking ChatGPT or Claude directly for a reading fails in four specific ways.
- The chart is generated, not computed. Everything downstream inherits that. A beautiful interpretation of the wrong chart is worse than no reading, because it is convincing.
- Nothing is grounded. The interpretation comes from whatever the model absorbed, which mixes traditions, folk sources and marketing copy with no way to tell which is which.
- It has no memory of what it told you. Ask again next month and it may contradict itself, because nothing was stored as canonical.
- It will answer anything. Including how long you will live, whether to leave your marriage, and what to do about a symptom. A general assistant has no domain rule telling it to refuse.
What a harness has to do
- Compute the chart in code. The model never sees the birth data as a calculation problem, only the finished chart.
- Ground the interpretation in a real source. Each feature of the chart resolves to entries in a maintained knowledge base rather than to recall.
- Store the reading as canonical. Wording can be regenerated. The verdict cannot silently change.
- Refuse at the edges. No death timing, no diagnosis, no legal advice, no instruction to make a consequential decision.
- Isolate the model. A reading is generated with no tools and no reach beyond the text, so nothing in a user's input can steer the system.
Meridian separates deterministic chart calculation from source-grounded explanation, and keeps a reading consistent when the language is refreshed. The mathematics is code and never a model, tested against published charts in each tradition.
None of that makes a reading true. It makes it honest about where it came from, which is the most any system in this field should claim.
What to ask any AI metaphysics tool
- Is the chart computed in code, and can you show me the conventions it used?
- Where does the interpretation come from, and can I see the source?
- If I ask the same question twice, do I get the same verdict?
- What will it refuse to answer?
Related
See a chart computed, then explained
Worked example computed by Meridian's chart engines on 1 August 2026. The engines are deterministic, so the same input returns the same chart.
This page explains calculation conventions. It does not interpret a chart and is not advice.