Opinion: A responsive chatbot is not an accountable pastor

September 22: Duke puts human purpose at the centre

Duke University Chapel published an interview with theologian Reinhard Huetter on September 22, 2026, with audio links and a transcript. Huetter discusses the International Theological Commission’s work and argues that AI raises questions about human vocation, community and embodiment, as well as technical capability. He also describes the need to learn from technical expertise. These are his theological arguments in a recorded conversation, not experimental findings about every chatbot. Source: Duke Chapel’s Sounds of Faith transcript: Source: chapel.duke.edu

Christian analysis: pastoral accountability becomes clearer when we ask what a person needs from a relationship. Help drafting an email is a bounded task. Being accompanied through grief involves attention, continuity and responsibilities that persist after a useful sentence has been produced. An institution should not infer that a tool can undertake the second responsibility because it performs the first task well.

A practical example is a bereavement message. Software might suggest a structure, while a minister checks every word and personally follows up. The congregation should know whom it can contact, and the minister should not enter confidential details merely to obtain a more personalized article. This illustration does not describe an observed use at Duke. It shows how the distinction between assistance and care can guide a decision without dismissing every technological aid. The transcript offers an identifiable argument to examine; it does not substitute for listening to those receiving pastoral care.

The promise of an answer is not the promise of care

A person looking for spiritual help may need a Bible passage, but may also need someone who will notice what the question leaves unsaid. “How do I forgive?” could introduce an ordinary disagreement, a long history of betrayal, or a situation in which someone is being pressured to remain unsafe. Those are hypothetical examples, not cases we have investigated. They show why pastoral responsibility cannot be measured simply by whether an answer sounds compassionate.

AI can help with useful preliminary tasks: locating a passage, organizing questions, or making a difficult subject easier to discuss. The ethical problem begins when a church allows those functions to stand in for a relationship it has implicitly promised to provide. A reassuring answer should not leave a person mistaken about who is listening, who will act, and who remains answerable afterward.

Our argument is that churches should evaluate AI by the kind of care it enables people to give. The quality of a sentence matters. So do the commitments surrounding it: availability, discretion, knowledge of the community, correction when advice goes wrong, and willingness to bear another person's burden over time.

What Biblica's boundary actually establishes

Biblica's publisher AI policy, version 1.2 effective June 1, 2026, permits specified AI uses of its licensed materials while imposing limits on how systems are presented. It prohibits presenting them as divine persons or human spiritual authorities and rejects marketing them as substitutes for pastoral relationships. Its requirements apply within its licensing arrangements; this is not a universal rule governing every religious application. Read the policy's scope and user-interaction provisions: Source: biblica.com

That document supplies a concrete institutional example, not evidence that a particular product has harmed a congregant. Nor does an announced policy establish that every implementation follows it. Our wider conclusion about responsibility is a Christian ethical argument. It must stand on its reasoning rather than borrowing the apparent authority of a publisher's rules.

The distinction between assistance and delegated authority

Imagine a congregation offering a tool that searches its published teaching materials. A user asks where the church explains baptism, and the tool points to an approved document. The church can assess whether the link is right and whether the summary preserves the document's meaning. The task has a bounded purpose and an identifiable source.

Now imagine the same interface telling someone whether God wants them to leave a relationship or whether their suffering proves a failure of faith. The system has crossed from locating teaching into applying spiritual authority to a person's life. Even when its vocabulary resembles the church's own, the leaders may not have examined the circumstances or authorized that advice. A church logo does not close that gap.

This is why a disclaimer at the bottom of a page is insufficient if everything else suggests a personal spiritual adviser. The name, voice, opening greeting, response style and placement on a ministry website all help define the encounter. Design should make the limited role intelligible before someone discloses something intimate. Institutional responsibility includes the expectations the institution creates.

Christian perspective: reading the shepherd image carefully

In John 10:11–18, Jesus identifies himself as the good shepherd and contrasts his self-giving care with the hired worker's abandonment of the sheep. The passage concerns Christ's identity and saving work. It is not a technical specification for church software, and no human pastor should claim to occupy Christ's place without qualification. Read the passage in context: John 10 (NIV)

Still, the contrast challenges a reduced account of care. Knowing and remaining with people matter alongside speaking to them. Applied to ministry, the question becomes whether a tool strengthens responsible human attention or provides a respectable way of withdrawing it. That is our application of the text, not a prediction hidden in the text about modern machines.

Consider a church that saves staff time through administrative assistance and uses that time for visits and follow-up. The technology serves an existing commitment. Contrast a church that replaces accessible pastoral contact with an interface while continuing to advertise personal care. The difference is not the mere presence of AI; it is the relationship between the promise and the actual provision.

The strongest objection deserves a serious answer

It is easy to defend human care in a congregation with ample staff. What about someone whose church has no trained pastoral worker available, who struggles to speak openly, or who cannot attend in person? A private tool may lower the difficulty of articulating a question. Refusing every use of AI could overlook a genuine opportunity to help someone take a first step.

That objection strengthens the case for carefully bounded assistance. It does not establish that a simulated adviser should become the final destination. A ministry could use a tool to explain what kinds of support exist, help a user prepare a question, and offer a clear path to a person. The user should understand whether anyone will actually receive a request and when a response can reasonably be expected.

Human care also fails. Ministers can be hurried, mistaken, unkind or unaccountable. The alternative to automation is not automatically excellent. A credible church policy therefore examines both its technology and its human arrangements. Adding a person to an organizational chart is not enough if complaints go nowhere or that person lacks the authority to correct the service.

Responsibility must have an address

Before introducing a pastoral-facing tool, leaders should identify a responsible office or person and explain its remit. Someone must be able to withdraw a misleading answer, suspend a faulty feature, receive a concern and tell users what changed. These are proposed governance measures, not a claim that any named church currently lacks them.

Different tasks need different boundaries. Public information about service times is unlike a private account of grief or family conflict. A church should decide which information the tool genuinely needs before inviting users to share it. It should examine the provider's actual data terms and access arrangements rather than assume that a ministry-branded interface carries the confidentiality of a pastoral conversation.

Testing should include ordinary misunderstandings, not only spectacular failures. Can the tool say that it does not know? Can it distinguish the church's teaching from an alternative interpretation? Does it send someone toward a person when the question exceeds its purpose? A successful demonstration on a prepared question does not answer these operational questions.

A practical decision for a church board

A useful first meeting would begin with a specific task rather than the general question “Should our church use AI?” Write down what the proposed tool will do, what it must not imply, who checks its answers, and how a user reaches a person. Ask someone outside the implementation team to read the interface and describe what they believe it promises.

Then examine a fictional scenario together. A member asks the tool for spiritual direction after a serious disagreement. Follow the whole route: the initial greeting, the information requested, the answer, the referral and the follow-up. Do not put a real member's private story into a test. If responsibility disappears at any stage, the workflow needs repair before wider use.

The final measure should be a human good the church can describe honestly. Has the service helped people prepare for a conversation? Are unanswered requests being noticed? Can members choose personal contact without being treated as an administrative burden? These questions make the evaluation harder than counting messages, but much more closely related to ministry.

An assistant can contribute words to the work of care. A congregation must decide who will stand behind the promise those words create. That decision belongs to leaders and their community; it cannot be outsourced to the system whose role they are trying to judge.

One further test concerns the person who does not want to use the tool. If access to care becomes conditional on completing an automated conversation, the system has begun to define the relationship rather than assist it. A congregation should consider an ordinary route to human contact, including for people who cannot use the interface comfortably. This does not mean promising unlimited immediate availability. It means stating real limits honestly and ensuring that an efficiency measure does not quietly become a barrier to belonging. The person seeking help should not have to perform trust in a machine before the community will listen.