Can AI deliver the Word of God?
Webinar on AI-powered ministry covers the pros and cons of artificial intelligence
What is the difference between worship or a Vacation Bible School curriculum planned by AI and one written by your ministry team? What ethical and environmental concerns should play a role in when and how you use AI in your ministry? AI may be able to predict the next best word or phrase in a pastoral prayer, but can it really pray for a congregation? Programs trained on massive sets of language can write a sermon in less than a minute, but can they deliver the Word of God?
Questions about what artificial intelligence can and can’t do permeate our current culture. Last year, The New York Times published a story contest between writer Curtis Sittenfeld and ChatGPT. The literary experiment was designed to see if people could tell the difference between a story written by the human imagination and a story written by artificial intelligence.
Christianity relies heavily on story for the work of spiritual formation as well as on relationships between generations of disciples who find connection between their personal story and the narrative of its faith tradition.
Last week, the Office of Christian Formation hosted a community circle called “AI-Powered Ministry,” led by the Rev. Dr. Andy Morgan, director of family faith formation at First Presbyterian Church in Knoxville, Tennessee. During this 90-minute webinar, a group of 30 Christian formation leaders considered how AI could power their ministries and when AI could or should not. After providing a helpful handout that included a taxonomy of AI terms, Morgan explained the current landscape of AI models and how he evaluates them through his faith and moral commitments as a disciple of Christ.
Since attending the first “AI and Church Summit” in Seattle last August, Morgan has offered his presentation in multiple church contexts. Since AI is already permeating lives and ministries, he encouraged webinar attendees to be informed and thoughtful about using it. He highly recommended that people virtually attend the second ecumenical "AI and the Church Summit," set for September. When the 2025 AI and the Church Summit livestream registration is available around June 5, readers can access it here.
As an introduction, webinar participants identified where AI was operating in their lives and what they hoped it could do. Morgan suggested practical examples of using AI for ministry tasks, such as creating intergenerational games for VBS, liturgy and lesson plans. He explained which current programs could best be tailored to specific needs, like leveraging Google’s NotebookLM for precise information retrieval to better capture one’s own voice based on previous personal content.
“Processing a prompt on a large language model like ChatGPT is equivalent to wasting about a bottle of water due to the significant energy consumption and cooling requirements of its server farms,” said Morgan, highlighting the concept of digital citizenship and the need to balance AI's efficiency with the unique concerns of human relationships, especially in regard to environmental impact. As a personal practice, Morgan limits his ChatGPT use to 10 times per day. He keeps a dry-erase board near his device to mark each time.
Large language models like ChatGPT provide unique output to specific inputs. Because of that, under copyright law, those who send the prompts own the copyrights to its results. Because of this permissive gray area, and because AI is still generally misunderstood and anxiety-provoking, Morgan doesn’t cite ChatGPT when he uses it in worship materials, as he would another copyrighted source like “Feasting on the Word” published by Westminster John Knox Press. He admitted to using ChatGPT for liturgy, such as calls to confession based on the original prayers of confession that he writes. Morgan believes only those who love the people can pray for them, so he won’t let AI write his prayer. “ChatGPT does not love my people,” he said.
Morgan was candid that his theology and ethics around AI are personal, and he encouraged others to develop their own boundaries through the same discipleship and discernment their faith leads them to engage the world. There are times when the efficiency of AI and its access to information can benefit others, like when Morgan asks it to adapt his weekly lesson plan to the specific individualized education plan accommodations of a certain young person in his Christian formation program.
Because large language models were “initially trained on uncurated data scraped from the internet from sources like Reddit comments, Quora forums, Facebook comments, and other open internet blogs and social media platforms,” Morgan emphasized how AI is not value-neutral and needs a discerning mind to know when and how to use it. When these AI models began, according to Morgan, “the goal was to gather the largest possible dataset to help the AI learn statistical patterns of language and predict the most probable next word in a sentence.” This training approach means the models inherently carry biases “present in online comments, including racial, gender and Western biases,” he said.
“From a Calvinist perspective, the AI represents our inherent biases, limitations and imperfections, essentially mirroring human nature in its raw, unfiltered form,” said Morgan, who pointed out that “AI is not an objective, impartial system but a technology deeply embedded with human sin.”
Accuracy is also a concern. Morgan specifically noted that ChatGPT is approximately 67.9% accurate across different use cases, which is not highly reliable. The primary goal of these models is to predict the most probable next word in a sentence based on statistical analysis of language patterns, not to ensure absolute accuracy.
Understanding and producing human-like text is not the same as accurately representing the truth. Morgan emphasized that while it benefits tech companies to make their models more accurate, the core function is solely language generation. Perhaps the difference of purpose explains what the readers of the Sittenfeld vs. ChatGPT essay contest perceived when trying to pinpoint what made a story truly human.
The discrepancy between generating language and truth-telling is also important for those who wish to proclaim God’s Word. The readers who claimed to know the difference between a story told by a large language training model and the human imagination relied on the subtleties of relationships between words, but most importantly, the writer and her subject. A human-told story conveyed more than facts, logic or reliable observation of its world; it conveyed a longing to truly know and understand the ineffable connections between characters and the environments that shape them.
A theologian once said, “Learning is soul work,” in response to her students becoming more dependent on AI to write their theological reflections. The practice of putting human longing for divine love and perfect will into language is spiritual formation.
Like any religion, Christianity is a language people speak. There are thousands of years of material from which a large language model can draw to predict the most probable way to describe it. But the Christian tradition is not just a theological vocabulary; it’s a world its disciples inhabit, constructed through language but also built through communal action and relationship.
Through analyzing massive amounts of data, AI is trained to care about the relationships between words. But disciples of Christ are asked to search for an improbable truth through a larger model of relationship — divine love taking human form.
Like any technology or lectionary, preaching or teaching aid, there should be limits to how much power AI has within ministry.
Follow the Office of Christian Formation on Facebook for dates and topics of upcoming Community Circles.
You may freely reuse and distribute this article in its entirety for non-commercial purposes in any medium. Please include author attribution, photography credits, and a link to the original article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDeratives 4.0 International License.
