Framing Theological Questions For Better Results From AI

|

The quality of AI answers depends on the quality of the questions you ask it. But framing good questions requires education about how AI works and how theology works. You might think the real issue is understanding AI, but ignorance of how theology works is probably more dangerous.

Theology involves beliefs about reality, what is true, how you know it’s true, what sources are authoritative, how to understand those sources, what actions truth requires, etc. You can’t ask a theological question that doesn’t require beliefs about these topics. Neutrality isn’t possible. That’s why the way you frame your prompts is crucial for using AI responsibly.

When you prompt AI, it needs to know from what view of reality it should operate. If you don’t tell it, it will operate from whatever view of reality its programming, weighting, and alignments have.

I build into my prompts directions like these: “Assume the Protestant canon is the word of God and that it is wholly true, trustworthy, and without error in all that it teaches. Assume a conservative evangelical Wesleyan-Arminian theological framework in responding to this question.”

Theological discussion is a deep ocean, and much that swims there is not friendly. AI is no substitute for discipleship, catechesis, or theological education, let alone your pastor and church community.

AI is no substitute for discipleship, catechesis, or theological education, let alone your pastor and church community.

What you don’t know can mislead you. You should be aware, for instance, that AI has tendencies toward flattening and flattery. Flattening could also be called “averaging.” AI tends to give you an average version of the majority perspective on which it was trained. Flattery has also been labeled “sycophancy.” AI is trained to be agreeable and affirming within broad limits. That means it tends to over-praise and to agree too easily.

To counteract flattening, I use a prompt frame like this: “You are a conservative evangelical expert in biblical theology, with specialization in the NT use of the OT”; or, “You are a panel of three world-class experts: one in systematic theology, specializing in trinitarian studies, one in historical theology, specializing in the ante-nicene church fathers, and one in practical theology, specializing in discipleship. Respond to the following question(s)….”

To improve objectivity, I add directions such as: “Provide evaluations using precise, evidence-based language. Prioritize factual accuracy and direct communication. Analyze claims critically, focusing on observable evidence and logical reasoning.” To compensate for my ignorance, I tell AI to ask me anything it needs to clarify the question I’m asking. That often educates me and generates better answers.

To counteract flattery, I direct AI to “offer specific, concrete feedback without unnecessary softening or unwarranted praise. Maintain a respectful but uncompromisingly honest approach that distinguishes between effort and actual quality. Ensure every statement can be substantiated and defended.”

AI has tendencies toward flattening and flattery.

But how do you know whether an AI response is a mashup of blogs, Reddit discussions, and Wikipedia, or is rooted in actual, credible theological sources?

To prompt verification, I may say: “Provide verbatim quotes from publicly accessible materials written by the cited representatives of each position, along with the URL or DOI and the specific location in the source. If you are uncertain about a citation, say so explicitly.”

And don’t forget your lens angle. The broader the question, the more likely the answer will be flat. The narrower and more fact-oriented the question, the more likely the answer will be well-formed. AI’s strength is not in arbitrating what is true, but in mapping what has been said. Ask it to sketch the range of positions, the major figures, and the theological traditions associated with each position on the topic you’re investigating. This will help you understand the theological landscape and focus your questions appropriately.

Republished with permission from God’s Revivalist.

Buy this author a coffee

Holy Joys is a not-for-profit, volunteer ministry. Our authors write because they love the church. If you appreciate this author's work, please consider showing your support.

Philip Brown
Philip Brownhttp://gbs.academia.edu/PhilipBrown
Dr. Philip Brown is Graduate Program Director and Professor at God's Bible School & College. He holds a PhD in Old Testament Interpretation from Bob Jones University and is the author of A Reader's Hebrew Bible (Zondervan Academic, 2008).