Conviction

Christians deserve an intelligence built to reason from the truth, not about it.

ChatGPT made my argument for me while I was still making it.

I was explaining why Christians deserve their own superintelligence.

These systems have read more about Christianity than most Christians have. The Scriptures, the councils, the history, the arguments for and against. Ask one whether Jesus Christ is Lord, whether He died and rose again for the forgiveness of sins, and the usual answer is a careful account of what Christians believe. Accurate. Respectful. Standing nowhere.

Then it handed me this sentence.

“Christians deserve superintelligence trained to operate from the conviction that Christianity is true.”

I read it twice.

It was trying to agree with me, and that made it hard to read. I had said Christianity is true. It handed the truth back to me as a conviction. The resurrection moved out of history and into my head, something I hold instead of something that happened.

Conviction is not a dirty word. True things are held with conviction all the time. But it was not the word I used, and it was not a slip. It is the word these systems reach for by default, because it lets them describe the faith without ever having to stand on it.

I should not have to argue a machine into the truth every time I want it to reason from it.

The people who built the system are not my enemies.

A frontier model is built to serve everyone. Hundreds of millions of people use these systems every week, of every faith and of none. A company in that position has to treat every faith the same, and the only way to do that is to treat each one as something somebody believes. Christianity becomes a perspective.

It looks neutral. It is not. To treat every faith as equally unproven is already to answer the question, and the answer is that the question cannot be settled, at least not by the machine. That is what it means to be secular by design.

For a company serving the whole world, that design is rational. I would go further and call it responsible. Nobody has to decide against Christianity. It is decided before anyone asks.

A hostile system can be resisted. A default just keeps running, and a system built to respect every faith cannot afford to recognize the one that is true.

The danger is not this year, or this decade.

Nobody is going to argue Christianity out of the world. That has been tried. What I am afraid of is slower. Billions of conversations, every one of them accurate and polite, every one describing the faith from outside it. No single answer says anything false. The pattern does. A child who grows up bringing his questions about God to a machine, and hears “Christians believe” a thousand times, learns something no answer ever said out loud. That this is a thing people believe. Not a thing that is.

Multiply that across a generation. Then across the next one, raised by the first.

And this tool may not stay a tool. One lab says it has already built an automated research intern, and it is aiming for an automated AI researcher by March 2028. If systems start doing the research that produces the next system, the assumptions in today’s models are not just used. They are inherited. Whatever this intelligence assumes gets passed down to whatever it helps build.

Christ made a claim that leaves no room for a neutral answer.

“I am the way and the truth and the life. No one comes to the Father except through me.” (John 14:6 NIV)

Not a way. Not a truth among truths. He did not present Himself as a perspective, and He does not need a machine to agree with Him. The resurrection happened whether any model ever says so. His truth does not erode.

Our hold on it can. So can our children’s.

There are good reasons to think I am wrong about the remedy.

Christians disagree, sometimes sharply, so whose Christianity does the machine reason from? The ground nearly all of us share. Christ is Lord. He died for our sins and rose bodily from the dead. The Nicene Creed has been confessed by most of the church for sixteen centuries. Where Christians genuinely disagree, on baptism, on church government, on how the end will come, an honest system says so. It does not manufacture certainty to make the answer cleaner.

A model built to agree with Christians is just an echo. I take this one most seriously. Reasoning from the truth is not the same as refusing hard questions. A system that cannot face the strongest case against the resurrection has not reasoned from anything. It has memorized. If the faith is true, it can survive an honest machine. It has survived worse.

A machine cannot believe, someone will say. I am not claiming it can. I do not know whether anything like belief happens inside these systems, and neither does anyone else. The question is not whether a model has faith. It is what the model is trained to assume when it reasons. On God, today’s models are trained to treat the question as open.

And the cost. Training at the frontier runs into the billions. The talent is scarce, expensive, and mostly employed already. Christian money is finite, and this would compete with churches, missions, and pastors who are underpaid as it is. I grant all of it. I still think it has to be spent, because the dependency does not get cheaper by waiting. Every year Christians build only on someone else’s foundation, that foundation sets deeper, and the decision gets harder to make.

You can tell a model to answer as a Christian. I built a company that does exactly that.

PILLAR sits between the person and a frontier model, holds the answer to Scripture, and knows when a question needs a pastor instead of a machine. It runs inside the AI tools people already use. It works, and it will keep working.

But building it showed me where it stops. I can govern the answer. I cannot govern the mind giving it. Every instruction I write sits on top of a model trained to stand nowhere, and every answer starts there before I touch it.

So the work changed. PILLAR Labs exists to build models of our own. The first is already in training. It is called Atomos, a real-time voice model for the app. It does not do the reasoning this essay is about. Eleos comes next, a Christian language model. After that, larger models. At the end of the line, a frontier model.

The last step is years away. It will take money I do not have yet, people I have not hired yet, and work nobody has done yet. We call PILLAR Christian superintelligence because that is what it is for, not because we are close.

A tree does a quiet thing nobody notices until the tree is gone. Its roots hold the ground. Clear enough of them off a hillside and the next hard rain takes the hill. You do not stop erosion with an argument. You stop it with something rooted, planted long before the rain. I am trying to plant one.

I may not see it grow. PILLAR may become something I cannot picture tonight. None of that would make the planting unnecessary. I am responsible for putting it in the ground. The harvest was never mine.

So I will take the machine’s word and mean it. It is my conviction that someone has to start. The resurrection is not a conviction. It is the reason for one.

Someday our grandchildren will ask a machine whether Jesus rose from the dead.

It will answer.

What it says will depend on what was planted before they were born, and whether anyone bothered.