The Message and the Messenger Part 3

Most of us have fixed a headlight on our car, followed instructions for putting together a shelf or replaced a faucet with the help of a stranger on the internet. We search, we find a set of instructions or a short video, and we follow along. I cannot remember ever stopping to wonder who wrote those instructions, whether he was kind to his family, or what his pastor thought of him. The faucet either leaked or functions as it was intended.

That ordinary experience sits awkwardly beside the sermon that started this series. My pastor warned that “the less access we have to someone’s character, the more cautious we should be about giving them sweeping authority over our lives.” In the first post I applied that caution to myself. In the second I applied it to the parachurch ministries that have shaped how many of our churches read Genesis. Both times I assumed there was a person behind the curtain, someone whose character we simply could not see from where we sit.  And in both cases, the character of that person mattered to the message.

I ended the last post with a question I want to answer now. What happens to the message and the messenger when the voice speaking to us may have no backstage at all? To answer it, I first need to explain why the widget instructions never troubled me, because I think the reason can help us determine exactly where to apply caution to this tale.

When Does the Messenger Matter?

My pastor drew two tests from Scripture. The Bereans tested the message against the Scriptures (Acts 17:11), and that test works at any distance. Hebrews 13:7 asks us to consider the outcome of our leaders’ way of life and imitate their faith, and that test requires access. The question is when we need the second test at all. As I have thought about it, four things seem to decide how much the messenger matters.

The first is the scope of the claim. “Tighten this bolt a quarter turn” claims very little.  In contrast, “this is what a faithful Christian must believe” claims a great deal. The wider the claim reaches into how we live, worship, and whom we trust, the more the person making the claim matters.

The second is whether we can verify it quickly. Widget instructions check themselves. If I follow them and the shelf collapses, I know within minutes. Other claims can be checked only over years, or never in this life, and for those we lean far more heavily on the character of the one making them.

The third is the stakes and whether they can be undone. A wrong turn in a recipe costs me a dinner. A wrong turn in counsel about whether to leave a church can cost a family its friendships, a child’s trust, and sometimes its faith. High stakes that cannot easily be reversed would seem to me to call for a messenger we know.

The fourth is the one Hebrews 13:7 points to directly: whether we are receiving information or a life to imitate. A paleontologist can teach me about a fossil without my wanting to become like him. A pastor teaches by example as well as by word, and that example is the part I cannot or find very difficult to test from a distance.

Put those four together and a spectrum appears. At one end is the arithmetic and widget instructions. Then come a recipe, medical advice, and a summary of the scientific consensus on some question. Further along the spectrum we may find the interpretation of Genesis, counsel about whether to leave a church, and finally spiritual shepherding. The messenger matters a little more at every step.

Where the Age of the Earth Sits

This spectrum helps explain the last post. As a scientific question, the age of the earth belongs near the widget end. The claim is narrow, the evidence is public, and anyone willing to do the reading can check the ice cores, the varves, and the radiometric data for himself. The messenger matters relatively little.

Well, that might be a bit oversimplified. For many people the messenger is trusted more than their own “research.”  They prefer to have an authority interpret the evidence for them. 

In the last post I described three situations that took that same question to the far end of the spectrum: recasting interpretation as a matter of trust, tying the question to the gospel, and evaluating churches from a distance. By the end of that journey, a question you could take backstage and check had become a test of whether your pastor is faithful. The spectrum shows why that relocation matters so much. It moves a question from the place where the messenger hardly matters to the place where the messenger matters most, and it hands the authority that belongs at that end to a messenger we may not be able to fully know.

But, What If There Were a Messenger With No Backstage at All!?

My pastor’s caution assumed a person standing behind the curtain whom we simply could not see. Artificial intelligence introduces three situations his sermon did not address, though I think its logic reaches all of them.

First:  A messenger with no backstage of the human kind. An AI chat tool has no marriage, no children, no private conversations, no history of handling criticism well or badly. There is no outcome of a way of life to consider, because there is no life. Here Hebrews 13:7 fails at the root. There is no faith to imitate.

Second:  A front stage with nothing behind it. In Part 1 of this series I noted that the internet is very good at making distant voices feel close. AI takes that a step further. Beginning in March 2025, a YouTube channel called “JOHN MACARTHUR WORLD” posted nearly a hundred sermons generated by AI in the style and voice of the well-known pastor. By April 2025 the most popular drew close to 100,000 views apiece. Similar channels have imitated Voddie Baucham. These are front stages with no backstage whatsoever. A listener could apply the Berean test to such a sermon, but he could not know whether any human being stood behind a single word of it. (Protestia, April 21, 2025)

We are in the very early stages of this technology. The ability to pretend to be nearly anyone who has enjoyed some authority and following will only become more prevalent in the near future.

Third: A backstage of a different kind. Every AI has a backstage, though it looks nothing like a theater’s. It consists of the material the system was trained on, the design decisions of the people who built it, the instructions it has been given, and the interests of the company or ministry that operates it. Most of that is invisible to the person typing a question.

This is where the subject returns to the ministry I wrote about last time. In June 2024 I suggested that Answers in Genesis would likely build its own AI tools, trained on its own materials, creating “a siloed information ecosystem.” Within months AiG was raising $3 million for a project it called AI Genesis. Its stated reason was to produce an AI “that does not have the fallible assumptions the world builds into such technology but will be built from the ground up, based on God-honoring principles.” The project included an interactive AI Noah for visitors at the Ark Encounter, which was offered for a time; I cannot confirm whether it is still in use.

In January 2025 Ken Ham announced a test version of the AI Genesis chat tool on AiG’s website, explaining that “the AI will gather the answer from across our website, summarize it for you, and then provide a link for deeper study.” In other words, its backstage is one ministry’s archive. I want to give Ham credit for what he wrote next. He cautioned that the tool “has limitations and nothing replaces God’s Word,” and he encouraged users to be Bereans and check its answers against Scripture. On that point he and I agree completely. I should also note that in the middle of 2026 the AiG chatbot disappeared without explanation.

But I would add one observation. There is no such thing as an AI without assumptions. Every one of them carries the assumptions of its training material and its builders. The only question is whose assumptions, and whether the user can see them.

That applies to my side of the question as well. I use Claude, a general-purpose AI assistant made by the company Anthropic, and some other LLMs for a variety of uses in my work and personal life. If I were to ask it how old the earth is, it would likely give an answer I agree with. That agreement would be no reason to trust it more. If anything, it is a reason to check it more carefully, because agreement is exactly when I am least inclined to check. Its backstage is a technology company’s training data, design choices, and business interests, and I can see very little of it.

Five Questions for a Machine

In Part 1 I offered five questions to ask of any distant teacher, and I answered them about myself. In Part 2 I applied them to a parachurch ministry. I think we should put those same questions to a machine.

Who could correct it? Its makers can, by retraining it or changing its instructions. You likely cannot unless you have a large budget and some expertise. You can argue with it, and it may change its answer in the conversation, but little you say reaches its backstage.  That said, you can “train” in to some extent on your preferences and expectations but that can only go so far and even then you likely do not know how it is taking that input into account when it crafts a response to your inquiry. 

When does it say “I was wrong”? Often, and that turns out to be part of the problem. These systems are known to apologize and revise an answer when a user pushes back, even when the first answer was correct. A messenger who concedes whenever challenged has told you very little by conceding.  It also presents a danger such that you feel like you won and that it was the quality of your argument that won the day.

How does it treat people who disagree with it? There is no conduct under conflict to observe. The tool is built to be agreeable, which feels pleasant and teaches you nothing about character, because there is no character to learn about.

Can you check its sources? Sometimes, and you must.  If you don’t know how to ask you will likely not be able to know the source of its response and for some things it will remain a black box no matter how clever your prompting may be.  Every output has to be checked against the original before it goes anywhere near a post.

What is it asking you to do? Strictly speaking, nothing, since it has no will of its own. Its makers may want a great deal: your attention, your subscription, your loyalty to their framework. Keep that in mind every time you open the chat window.

Tally the answers and the verdict is clear. An AI fails the messenger test entirely, and it passes the message test only when you do the checking yourself. On the grid from Part 1, it belongs in the narrow-authority column, whatever ministry or company built it. It is a tool.

Where AI May Help

None of this means Christians must avoid these tools. I serve as one of my Universities AI Ambassadors, leading workshops for faculty about AI in their teaching. Those workshops exist to help faculty talk honestly with one another about what these tools can and cannot do. How to design assignments both to avoid student use of LLM tools and to promote proper student use of LLM tools.  In those conversations, and in my own work, I have found AI useful for tasks whose results can be checked.

It can summarize a long paper so that I know which sections to read closely. It can help me locate a source I half remember. It can check my arithmetic, translate a passage, or reformat a reference list. It can lay out the strongest objection to an argument I am making, which is a discipline every writer needs and few of us practice. In each case the output is something I can test, and the authority I grant the tool is narrow.

Used this way, AI can be a remarkable aid to the Berean test. Ordinary believers can now trace a claim back to its source, find the paper behind the headline, and see what the backstage of a claim really says, more easily than at any point in history. That is the opportunity. The danger lies in letting the same tool drift toward the far end of the spectrum, where we begin asking it what to believe, whom to trust, or how to live.

If I could leave readers with one line from this post, it would be this one. AI can help you test the message. It cannot be the messenger.

The Importance of “Among You”

My pastor told us that the key words in 1 Thessalonians 1:5 were “among you.” Paul sent the Thessalonians a message, and he also lived among them, so they knew what kind of men he and his companions proved to be.

Even so, Paul knew the limits of a letter. Writing to that same church a few verses later, he says that after being torn away from them, “we endeavored the more eagerly and with great desire to see you face to face” (1 Thess. 2:17). The apostle John felt the same way: “Though I have much to write to you, I would rather not use paper and ink. Instead I hope to come to you and talk face to face, so that our joy may be complete” (2 John 12). If the men who wrote Scripture felt the distance of paper and ink, we should not be surprised that a screen feels distant too.

Now, at this point, you may object at this point. We never met Paul or John either. Does this whole line of reasoning undermine the authority of Scripture? It does not, and the Westminster Confession explains why. The authority of Scripture “dependeth not upon the testimony of any man or church; but wholly upon God (who is truth itself) the author thereof” (1.4), and our full assurance of its truth comes “from the inward work of the Holy Spirit bearing witness by and with the Word in our hearts” (1.5). Scripture’s authority does not rest on our access to its human authors. And the churches that first received those letters did know the men who wrote them.

For me, the deepest answer to this whole series lies in how God himself chose to speak. “The Word became flesh and dwelt among us, and we have seen his glory” (John 1:14). In Christ the message and the messenger are one. There is no curtain between his front stage and his backstage, and the disciples who spent three years with him could write of “that which we have seen with our eyes, which we looked upon and have touched with our hands, concerning the word of life” (1 John 1:1). God sent us more than a message. He came among us.

That is why I expect the local church rather than YouTube pastors and apologists to grow more valuable as synthetic voices multiply. A congregation that has watched its pastor’s family for ten years has a kind of knowledge that no algorithm can counterfeit. Christians have always used new technologies to carry the gospel, from letters to printing presses to radio, and there is no reason to panic about this one. What changes is how much of our trust can rest on what we hear from a distance. The answer has always been: less than we think.

Let me look back briefly over the ground we have covered in this short series. In the first post I turned my pastor’s principle on myself and asked how much authority a distant voice like mine deserves. In the second I put the same questions to a parachurch ministry and traced how a question anyone can check became a test of faithfulness. In this one I asked what happens when the voice has no backstage at all.

The same answer runs through all three. Test the message, wherever it comes from. Know your messengers, and give sweeping authority only to those who live among you.

There is one more messenger I should mention before I close, since I am a biologist and spend a good deal of my time outdoors. “The heavens declare the glory of God, and the sky above proclaims his handiwork. Day to day pours out speech, and night to night reveals knowledge” (Ps. 19:1–2). Creation is a messenger too, and its backstage stands wide open. Rocks, tree rings, and fossils have no curtain. What you see in the field is what is there, and anyone willing to look can go backstage. Fieldwork, in that sense, is going backstage in God’s creation. Rocks have no curtain.

Stay curious. Check my work.

Blessings,

Joel