Summary: I want to use the Royal Society’s gradually developing institutions of publication and criticism to reflect on contemporary AI collaboration and the life of faith. This is an analogy across eras. It does not mean that modern peer review was invented in 1665, nor that every church response can be reduced to defensiveness. My question is: after content is generated, who checks it, and who carries responsibility for it?

As someone with theological training who also uses AI heavily, I easily find myself thinking about two scenes at once: how a research paper earns trust, and how a sermon becomes a word that a community can carry.

They are not the same kind of work, and their criteria are not identical. Both, however, make me wary of assuming that fluent content and a confident author answer every question.

I therefore want to read several historical and engineering materials together while preserving their differences, then state my own judgment of faith.

Do Not Draw the Battle Line Between Science and Faith in Advance

I do not want to assume that technology and faith can only oppose each other. Nor do I want to make the opposite claim that they have always been harmonious. Discussing a historical event still requires attention to the texts, institutions, people, and conflicts of its time rather than replacing them with one total story.

For that reason, this essay no longer uses “the nineteenth century invented the conflict” as a summary of all history. Nor does the participation of some believers in research prove that the entire scientific institution was a form of worship.

The starting point for cross-domain dialogue is to let different questions retain their own criteria.

For faith, the value of a person and the way a community lives are questions I need to face seriously. A testable claim still needs corresponding evidence. Placing them in the same essay does not allow either one to answer the other.

The Journal’s Founding in 1665 Did Not Complete Modern Peer Review

According to the Royal Society’s history of its journals, Henry Oldenburg founded Philosophical Transactions in 1665 and served as its publisher and editor. After the Society took control in 1752, it adopted a collective editorial process. More systematic expert review developed further in the nineteenth century.

This was an institution that formed gradually. It cannot be compressed into a claim that Oldenburg invented a separation between “author” and “judge” on one particular day, nor can we project today’s understanding of peer review backward onto every submission he handled.

The gradual formation of an institution cannot be compressed into a single completed invention.

What I want to borrow from this history is the reminder that claims should be open to examination. I am not asserting a proven, unbroken institutional lineage from an early journal to today’s AI workflows.

Refusing Authority Still Requires an Account of How a Claim Was Checked

The Royal Society’s historical account connects its motto, Nullius in verba, with resisting domination by authority and verifying claims through experimental facts. A conclusion does not become true merely because someone other than the author approves it.

That reminds me of my own habits with tools. I do not want to stop asking questions merely because a response sounds professional or because another model agrees.

Adding another reviewer still requires evidence that can be checked.

Two roles may miss the same issue, while self-review may uncover a genuine error. The useful questions are which information was examined, which criteria were used, and which parts remain uncertain.

AI Generation and Evaluation Are an Arrangement of Work

Anthropic’s “Building effective agents” describes an evaluator-optimizer workflow in which one LLM call generates a response and another provides evaluation and feedback. The document also stresses that the pattern is appropriate when the evaluation criteria are clear.

This can inform workflow design, but it does not require the calls to use different models and does not guarantee that their judgments are independent. I no longer describe it as proof that every model is incapable of self-evaluation or transfer a human psychological effect directly onto machines.

Separating generation from evaluation can arrange a check, but it cannot guarantee that the check is correct.

In my own work, I ask a role to report its findings and then return to the sources. The purpose is not to prove that one model is inherently more objective, but to give adoption, correction, and execution an explicit basis.

Churches May Be Asking Questions at Different Levels

Whether AI should help organize a sermon, how theological material should be checked, and whether members might treat a tool as their only source of companionship are not the same question.

Some people emphasize data accuracy. Others are concerned with communal relationships, while others explore whether tools can assist communication and service. Different limits should not be reduced to panic or defensiveness merely because they address different concerns. Whether a recommendation is appropriate depends on the problem it is intended to address and the effects it may produce.

Different faith responses need to be heard in terms of what each one is concerned about.

My own starting point is that a tool can help organize material, propose a structure, or polish wording. The person speaking still needs to check scripture and context, decide whether the material suits the community in front of them, and accept responsibility for using it. That is my understanding of pastoral work, not a rule automatically derived from an AI engineering test.

Four-panel comic: The starting point for cross-domain dialogue is to let different questions retain their own criteria. The gradual formation of an institution cannot be compressed into a single completed invention. Adding another reviewer still requires evidence that can be checked. Separating generation from evaluation can arrange a check, but it cannot guarantee that the check is correct. Different faith responses need to be heard in terms of what each one is concerned about. For me, a willingness to be examined is part of practicing faith.

What Remains Is the Faith I Am Willing to Carry

In my understanding of faith, a person’s worth should not be ranked by how quickly they write, how much they remember, or how much content they generate. I believe that human beings are made in the image of God, and that faith involves memory, suffering, hope, and mutual responsibility within an actual community.

AI may help me express something, but a better sentence does not mean that I have lived it. I still need to ask whether these are words I am willing to believe, practice, and submit to examination by the community.

In my historical notes on the Reformation and my reflection on incarnation, I have also asked how media, bodies, and faith relate. Those analogies may help me think. They do not mean that technology will repeat the same history, nor do they experimentally prove that a machine can never have some form of inner experience.

For me, a willingness to be examined is part of practicing faith.

This dialogue between technology and faith therefore does not have to begin with the question of who will replace whom. I would rather ask: Which statements already have support? Which are my beliefs? Who will be affected? What am I willing to take responsibility for?