Michlaude Discipleship Project

Why It Took So Much Research to Decide How to Read One Word

By Michael W Winslow ·

On this page (10 sections)
  1. The short version
  2. Why one word is a problem
  3. How a rule gets tested
  4. The hardest part, in plain words
  5. The rounds
  6. When to change the wording
  7. What the chatbots showed
  8. What is still weak
  9. The decision
  10. What this means for you

Beta. This post describes work in progress. The author makes every decision, including the one below. Nothing here is a proof.

The short version

The Bible uses "die," "dead" and "death" for more than one thing. A rule for reading them must say which thing a verse means, and it must say it from the text, not from what the reader already believes. Writing and testing that rule took seven versions in about two days. This post tells what changed in each round, what outside readers did with it, and where the rule stands now.

Why one word is a problem

Genesis 5:5 says Adam lived 930 years, "and then he died." A life ended.

Romans 6:2 asks, "How can we who died to sin still live in it?" The people asking are still alive.

Ephesians 2:1 tells readers they "were dead in your offenses and sins," and the next verse says they "formerly lived" in those ways.

Same word family, three different things. A reader who decides in advance what "dead" means will read every verse the same way, and the Bible does not. So the rule asks one question: what does the text itself say about this use of the word? If the text settles it, the rule gives an answer. If not, the rule says "the text does not tell" and stops. Anything the reader believes beyond that must be labeled as the reader's own belief.

How a rule gets tested

A rule stands only if Scripture itself shows it, with named examples, or the text forces it. No rule rests on silence. Every rule must say what would make it lose.

Then the rule goes to outside readers: AI chatbots. Each gets the same message with the rule and the verses, and is asked to apply the rule to real passages. Where the answers differ, the author looks at the wording. The chatbots decide nothing.

The hardest part, in plain words

In John 11, Martha says her brother "would not have died" (11:21). Jesus answers, and says the one who believes "will live even if he dies" (11:25). The rule says the second "dies" gets the same kind of death as the first, a body that stopped living. But only when three things are true. The original-language word is the same. Two different speakers are answering each other, with no trip, arrival or long wait printed between them. And both speak in one account. Most of the work went into those three conditions.

The rounds

Versions 5 to 6. The wording about two speakers was fixed. The rule also began to name one verse it does not settle, 1 Corinthians 15:54, where a saying about death is linked to a future event but the rule has no label for it.

Version 6b. One change: a group counts as one speaker whenever the text names it as one. Three chatbots read it: Claude Sonnet 5.5 (Extra setting), Claude Opus 5.5 (High setting) and Meta AI. In John 8, where several groups are named, the two chatbots that reached the passage found two readings that gave different results. All three asked something the rule had not covered: do the narrator's own sentences count as someone speaking?

Version 6c. Three changes: narrator sentences, what kind of meaning may be copied, and what "the same word" means. Three chatbots again. At the John 8 passage two readings still existed in all three, but now they printed the same label. None showed a wrong label.

Version 6d. The author made choices on five points, such as whether groups named in different words are different speakers. Two chatbots read it, Claude Sonnet 5.5 and Claude Opus 5.5. Each said that if "each to the other" is read very strictly, the rule fails its own John 11 example, because the English does not print who is spoken to in 11:23 and 11:24.

Version 6e. Three more answers from the author: no printed addressee is needed for each speech, the rule is read on the English text, and narrator remarks count as speaking. The same two chatbots read it. The John 11 example now worked in both. A new split appeared at John 11:14 to 11:16. Jesus says "Lazarus has died." Then Thomas says to his fellow disciples, "so that we may die with him." One reading copies Lazarus's kind of death onto Thomas's word. The other does not. They print two different labels, so the wording had to change.

Version 6f. A check found that 6e had left out a phrase from 6d, a drafting slip. The fix: the first death-word must be said to the second speaker, and the answer to the first, while speeches in between need no printed addressee. That fits the text. 11:21 says "to Jesus," 11:25 says "to her," and 11:16 says "to his fellow disciples," so Thomas's "die" is not copied.

When to change the wording

Before the 6e readers ran, the author chose a stopping rule. The wording changes only when a chatbot shows a wrong label, an answer of "the text does not tell" where the text does tell, or two readings that print two different labels on one passage. If two readings print the same label, that is written down as a stated limit and the wording stays. Parts of the rule that are neither seen in Scripture nor forced by the text are marked as the author's own choices, and they change only by the author's choice.

What the chatbots showed

They agreed on the easy cases and split on hard ones: groups of people, narrator remarks, and how strictly to read "to each other." The splits that mattered changed a label. Splits where different reasoning led to the same label did not change the wording. Each chatbot is reported separately, none is ranked, and an AI answer is not proof of anything.

What is still weak

The "later word copies the earlier word" part has one worked example from Scripture, the John 11 conversation. Other cases are listed as limits, not solved: groups named in different words (Genesis 23), how far narrator remarks go (John 8), and how many levels of quoted speech are allowed (Luke 16:30 to 31). John 11:26 is held in line only by a sentence in another part of the rule. And no chatbot read the final wording, 6f.

The decision

The author made this decision. The rule for reading "die," "dead" and "death" is accepted with stated limits, which the project calls "Passed limited." It is not on hold. "Limited" means three things. One worked Scripture example carries the "same word, later answer" part. Several edge cases are listed as limits and not solved. And the last piece, the 6f wording, was not re-tested by readers, because the author chose to stop testing here. No reader runs were done after the last wording change.

This is not "proven" and it is not "final." The rule can be reopened if someone finds a verse where it gives a label the text does not support.

What this means for you

You do not need the rule to check a claim about a Bible word. When someone says what "death" means in a verse, read the verses around it. Ask who is speaking, to whom, and about what. Ask whether the answer uses the same word as the question. If the passage does not say, "the text does not tell" is an honest answer, and what you believe beyond that is yours to label as belief.