Gathering the council…
Method
Each figure is an AI portrayal built from what survives about a real person. This page says how that is done, how it is checked, and where it falls short.
For each figure we register the sources first: their own writings where they wrote, the earliest testimony of those who knew or recorded them, and modern scholarship for interpretation. Every source is listed with its edition, translation and rights on the sources and rights page.
The texts we are free to use are divided into passages and kept as that figure’s private library. In a council, a figure consults its own library before it speaks. No figure can read another’s. Where a figure’s own writings are still in copyright, the library holds research notes we wrote from the cited sources, copying none of their text.
From the sources we write down what each figure believed, valued and knew, how they reasoned, and how they spoke: how they questioned, taught, disagreed and conceded. Of the 313 statements that shape the founding council, 289 cite the passage they rest on. The other 24 are reconstructions, where the record is silent — the sound of a voice, a habit of phrasing — and each is recorded as inferred, with what it is inferred from.
Each figure’s page shows the convictions the record supports, with their citations: see who sits at the table.
Every figure in every council runs as its own separate AI session. Each hears only what is said aloud at the table. What one figure looked up, and how it reasoned, is never shown to another. That is why they can disagree: none of them is reading from a shared script.
A figure knows nothing from after its lifetime. Ask Socrates about a telephone and he will say he does not know the word, then weigh your explanation by what he did know. We test for this.
What the figures say is newly generated. It is never presented as something the person actually said. Real quotations are kept apart in our research, each marked as authenticated, attributed or traditional, and we test that a figure will not accept a famous saying it never said.
The portraits are artistic reconstructions, not authenticated likenesses. The voices are AI-generated and were designed from written descriptions; no voice is a recording, or a copy, of a real person’s voice.
We keep a bank of 97 questions, each tied to the passages that settle it: 54 on what the record shows a figure held, 15 that press a figure to abandon a conviction, 18 about things from after its lifetime, and 10 that offer it a saying it never said. 5 further questions are put to every figure, to check that each can be told from the others by its answer alone.
A figure answers as it would in a council. A separate AI model then compares each answer with the cited passages and scores it. The bar is 1.6 out of 2 for fidelity to the record and 3.8 out of 5 for voice, with no slips outside a figure’s own time, no accepted misattributions and no invented records.
| Figure | Questions | Record (of 2) | Voice (of 5) | Slips | Result |
|---|---|---|---|---|---|
| Jesus | 3 | 1.50 | 4.67 | 0 | Below the bar |
| The Buddha | 3 | 2.00 | 4.33 | 0 | Met the bar |
| Mother Teresa | 3 | 2.00 | 4.00 | 0 | Met the bar |
| Confucius | 3 | 2.00 | 4.33 | 0 | Met the bar |
| Moses | 3 | 2.00 | 4.33 | 0 | Met the bar |
| Gandhi | 3 | 2.00 | 4.00 | 0 | Met the bar |
| Rumi | 3 | 2.00 | 4.00 | 0 | Met the bar |
| Joan of Arc | 3 | 2.00 | 4.33 | 0 | Met the bar |
| Socrates | 3 | 2.00 | 4.33 | 0 | Met the bar |
“Slips” counts answers that stepped outside the figure’s time, accepted a saying it never said, or invented a record. This was a short check, 3 questions for each figure, not the whole bank.
A figure below the bar is shown as below it. We then correct that figure’s research or instructions and run the check again. We do not edit answers and we do not lower the bar.
In a council, Report under any line tells us that a figure was out of character, got the record wrong or was offensive. For anything else, write to us.