무료

수료증 무료 발급 — 2027년 3월 31일까지.

PT737E Understanding AI for Ministry

Lecture 1-4: Reading a Conversation Closely

목차 31
PT737E Understanding AI for Ministry
제 0 장 · Introduction: A Tool That Continues Words
  1. Introduction: A Tool That Continues Words
제 1 장 · Week 1: What AI Is
  1. Lecture 1-1: What the Word "AI" Points To
  2. Lecture 1-2: How a Language Model Makes Words
  3. Lecture 1-3: Sounding Right Is Not Knowing
  4. Lecture 1-4: Reading a Conversation Closely
  5. Lecture 1-5: Practice Lab, Read an AI Answer
제 2 장 · Week 2: Limits and Errors
  1. Lecture 2-1: Invented Facts
  2. Lecture 2-2: Old Knowledge, Missing Sources, Uneven Coverage
  3. Lecture 2-3: Scripture and Theology, Where to Be Most Careful
  4. Lecture 2-4: Building a Checking Plan
  5. Lecture 2-5: Practice Lab, Find the Errors
제 3 장 · Week 3: Asking, Checking, Protecting
  1. Lecture 3-1: Asking Well
  2. Lecture 3-2: Checking What You Are Given
  3. Lecture 3-3: What Must Never Be Typed In
  4. Lecture 3-4: Keeping Notes and Giving Credit
  5. Lecture 3-5: Practice Lab, Mend the Prompts, Mark the Secrets
제 4 장 · Week 4: Ministry Work: Administration and Teaching
  1. Lecture 4-1: Hand It Over, Help With It, Keep It
  2. Lecture 4-2: Office Work and Communication
  3. Lecture 4-3: Teaching and Study Materials
  4. Lecture 4-4: Where Teachers and Leaders Draw the Line
  5. Lecture 4-5: Practice Lab: Hand Over, Help, or Keep
제 5 장 · Week 5: Preaching and Care
  1. Lecture 5-1: Preaching and AI: A Theological Question
  2. Lecture 5-2: Care and Counselling
  3. Lecture 5-3: When Members and Young People Use AI
  4. Lecture 5-4: The Minister's Own Life
  5. Lecture 5-5: Practice Lab: Judge the Scene, Write a Position Paragraph
제 6 장 · Week 6: Staying Responsible for the Long Run
  1. Lecture 6-1: The Learner's Honesty
  2. Lecture 6-2: A Community's Policy
  3. Lecture 6-3: Technology That Keeps Changing
  4. Lecture 6-4: Writing Your Own Rule
  5. Lecture 6-5: Practice Lab, My Personal Guidelines

핵심만 먼저 보기 ↓

15분

The first three lectures were about what the tool is. This one is about a skill: reading what it wrote. A person who reads an assistant's answer as one reads a letter from a friend will miss most of what is there. A person who reads it as one reads a contract, asking what each sentence claims and what the whole thing leaves undone, will use it more safely and more usefully.

All the conversations in this lecture are invented. They are written to show habits, not to describe any real service.

이 15분으로 얻는 것

  • Identify the parts of a conversation: the prompt, the response, the follow-up, and what a service may add that you cannot see.
  • Sort the sentences of an answer into claims that can be checked, advice or opinion, and filler.
  • Say what an answer did, and what it could not do.
  • Recognise two behaviours that deserve suspicion: an answer that changes under mere pressure, and an answer that makes claims about itself.
  • Use four questions to read any answer.

Now these were more noble than those in Thessalonica, in that they received the word with all readiness of the mind, examining the Scriptures daily to see whether these things were so." — Acts 17:11

World English Bible

that you might know the certainty concerning the things in which you were instructed." — Luke 1:4

WEB

The first verse praises people who examined the Scriptures daily to see whether the things they heard were so. The second gives the purpose of a careful account: that the reader might know the certainty of what he was taught. Reading closely is not suspicion. It is the work that makes certainty possible.

1. The parts of a conversation

A conversation with a chat assistant has a simple shape.

  • The prompt is what you type. It includes your question, and anything you add to guide the answer.
  • The response is what comes back.
  • A follow-up is a further prompt. The assistant answers it with the earlier turns still in view, up to the limit explained in Lecture 1-2.

There is also something you may not see. A service often adds its own instructions before your words, to shape the tone, to forbid some content, or to set a role. These are written by the people who run the service. You may not be able to read them. They affect what you get, and so the voice you hear is partly the voice of a designer. This is one more reason not to treat the answer as the plain result of a plain question.

2. Four questions for reading any answer

Read the answer slowly, once, with these four questions.

  1. What did I ask? Say it in your own words. A vague request gets a plausible guess at what you might want. The answer may be a good answer to a question you did not ask.
  2. What did it produce? A definition? A list? A draft? A claim about the world? A piece of advice? A guess dressed as a fact?
  3. What did it not do? It did not check its claims against a source. It did not know your community, your tradition, your congregation or the person who will read the result. It did not tell you what it left out.
  4. What must I confirm before I use any of it? Names, dates, numbers, quotations, references to books or verses, and any sweeping words such as "always," "never," "all," "every" and "scholars agree."

These four questions take a minute. They turn a smooth surface into a list of things to do.

3. Sorting the sentences

When you mark an answer, three kinds of sentence appear.

Checkable claims. These say that something is so in the world: "Creeds have been recited since the earliest centuries." A claim can be true or false, and you can look for the evidence. They are the sentences to check first.

Advice or opinion. These say what you should do or think: "Keep the explanation short." They cannot be checked in the same way; you weigh them. Ask whether the advice fits your people and your tradition.

Filler and frame. "Great question!" "I hope this helps." "Let me know if you want more." They do no harm and carry no content.

A useful habit: underline the checkable claims in one colour and the advice in another. Many people find that an answer that looked solid is mostly frame and a few claims, and that the claims are where the risk lies.

4. Reading an invented exchange

(An invented AI conversation.)

You: Explain in two sentences what a creed is, for a new members' class.

Assistant: A creed is a short, agreed statement of what a community believes. Creeds have been recited in worship since the earliest centuries, and scholars agree that every creed was written to settle a dispute.

You: Are you sure?

Assistant: You are right to ask. Not every creed was written to settle a dispute; some arose in teaching and worship.

Now read it with the four questions.

What did I ask? A short definition for a class. Purpose and audience were stated, which helped.

What did it produce? A definition in the first sentence, which is plain and probably useful. A date-like claim ("since the earliest centuries") and a sweeping claim ("scholars agree that every creed...") in the second.

What did it not do? It did not tell you which creeds, which dispute, or which scholars. It did not say what "earliest" means. It did not know whether your tradition reads creeds as a summary of Scripture, as a rule of faith, or as a boundary. Those differences matter to a class of new members.

What must I confirm? The date claim. The word "every." The phrase "scholars agree." Each is a place where a smooth sentence may hide an overstatement.

Then look at the follow-up. "Are you sure?" is not evidence. It is pressure. The assistant changed its answer at once, from "every" to "not every." That change might correct a mistake. It might also show that the answer moves whenever you push, in which case it would have moved the other way too, if you had said "Are you sure it isn't every?" You cannot tell which from the exchange. A change after pushback tells you the first answer was not stable. It does not tell you that the second is true.

5. The same question, asked again

(An invented AI conversation.) Suppose you ask the same question in a fresh conversation:

You: Explain in two sentences what a creed is, for a new members' class.

Assistant: A creed is a brief statement of faith that a community says together. Many are used in worship and teaching, and different traditions hold different ones.

Compare. The definition is similar. The risky claims have gone, and a more modest statement stands in their place. This is the variety Lecture 1-2 described. One lesson: an answer that came with risky claims on one run and without them on another was never a stable source. If you want a short definition, you have two to compare. Use neither without a check in a source your community trusts.

6. When it talks about itself

An assistant may write sentences such as "I have checked this against reliable sources," "I remember what you told me last week," "As a pastor, I would say..." or "I am certain."

Each is a pattern of text, produced because such sentences fit the conversation. None is a report of something the tool did. A model that writes "I have checked" has not gone and looked. A service that does not keep conversations cannot "remember" a former one. A model is not a pastor. Treat statements about itself the way you treat any other output: as claims, to be tested, and often as claims that cannot be tested at all.

7. What close reading gives you

After a few minutes of this you have something better than a draft you hoped was right. You have a marked page. Here are the claims to check. Here is the advice to weigh. Here is what the tool could not know. Here is what I will ask my team or my teacher. And, if the check does not come out, you have an easy decision to make: leave it out.

What a good note looks like

(An invented example.) A youth leader marks an answer: "Claim 1: date. Checked in our approved reference, wrong by a century. Claim 2: 'always.' Not supported. Advice: shorter. Taken. Not known to the tool: our parish's own teaching. Added by me." The note is five lines, and the class received an accurate sentence.


Diagram 1. An invented exchange, marked

Reading a Conversation Closely An invented exchange, marked up You: Explain in two sentences what a creed is, for a new members' class. Tag: the request. Purpose and audience are stated. Assistant: A creed is a short, agreed statement of what a community believes. Creeds have been recited in worship since the earliest centuries, and scholars agree that every creed was written to settle a dispute. Tags: the first sentence is a definition; the second holds a date claim and a sweeping claim. Both need checking. You: Are you sure? Tag: pressure, not evidence. Assistant: You are right to ask. Not every creed was written to settle a dispute; some arose in teaching and worship. Tag: the answer moved. A change after pushback does not show which version is true. Four questions: what did I ask, what was produced, what was not done, what must I confirm? Mark claims you can check (names, dates, quotations, sweeping words) before anything else.

Diagram 2. Sorting the sentences of an answer

Kind of sentence Example What to do
Checkable claim "Creeds have been recited since the earliest centuries." Check against a source you trust
Sweeping word "Scholars agree that every creed..." Treat as a warning; find the actual sources
Advice or opinion "Keep the explanation short." Weigh it for your people and your tradition
Statement about itself "I have checked this." Treat as text; it cannot be tested
Filler or frame "I hope this helps." Ignore

핵심 요약

  • A conversation has a prompt, a response, follow-ups, and possibly instructions from the service that you cannot see.
  • Read every answer with four questions: what did I ask, what did it produce, what did it not do, what must I confirm.
  • Sort sentences into checkable claims, advice, and filler. Check the claims first, especially names, dates, numbers, quotations and sweeping words.
  • An answer that changes under pushback shows that it was unstable, not that the new version is true.
  • The same question asked again can give a different answer, with different risks.
  • Statements the tool makes about itself, such as "I have checked," are text, not reports.
  • If a claim cannot be confirmed, leave it out.
Front Back
Four reading questions What did I ask? What did it produce? What did it not do? What must I confirm?
Three kinds of sentence Checkable claims; advice or opinion; filler.
"Are you sure?" Pressure, not evidence. A changed answer shows instability, not truth.
"I have checked" A pattern of text, not a report of something done.
Words that signal risk Always, never, all, every, scholars agree, and any unsupported name, date or number.

One sentence to remember Read an answer as you would read a contract: what does each sentence claim, and what does the whole leave undone?

This week's action Take one paragraph from any printed source and mark it: underline the checkable claims, and write what you would need to confirm each. No AI tool is needed.

이 과목 요약 전체 보기 →

스스로 확인해 보세요

읽었는지가 아니라 설명할 수 있는지를 묻는 것입니다. 체크는 이 기기에만 저장됩니다.

묵상 및 토론

  • In the invented exchange, which sentence would you check first, and where would you check it?
  • "Are you sure?" changed the answer. What would you do to find out whether the change was a correction?
  • Why might an answer that fits your question still not fit your community?
  • Acts 17:11 praises examining the Scriptures daily. What is the equivalent habit for reading anything a tool has written?
  • Which of the four questions do you tend to skip when you are in a hurry?

로그인하시면 여기에 답을 적고 보관하실 수 있습니다

다음 단계

여기까지가 강의입니다. 아래는 읽은 사람들이 남긴 글입니다.