The AI Preacher: from the Cloud or the Spirit?

AI has infused itself into every part of our lives. Nearly everyone can feel it nipping at some aspect of their jobs. Pastors like myself are not immune. As we see uses and misuses throughout society, should AI be allowed to take over the sermon?

A few weeks back, Robert F. Kennedy’s Make American Healthy Report quickly went off the rails as observant folks discovered tell-tale clues substantial parts were generated using AI. Studies were cited that didn’t exist, some by known researchers and some by researchers who, themselves, may not exist.

Researchers refer to this as “hallucination.” Not unlike the human experience, AI can become “sure” it is remembering information while it makes it up from whole cloth.

I knew about hallucination early on in my experiments with ChatGPT and other bots, but it’s easy to lose sight of. After all, the first time I witnessed it handing me documentation for a programming interface with incredibly precise details only to discover neither the details, nor anything about the programming tool, existed, it felt like surely I was the one mistaken. How could everything be so “factually” detailed and not exist?

The researchers on the MAHA Report should have known better. So too pundits and lawyers. In often embarrassing ways, everyone from students to high-powered leaders are learning a lesson: AI can get things wrong.

But, we can see how easy it is to believe a computer when it responds authoritatively and in detail. Someday, no doubt, what it says will be true. AI is improving at an astonishing rate, and I have little doubt that hallucinations will be a quaint issue of the past soon enough.

Even so, what ought to remain the domain of human oversight? There is more, after all, than pure factual accuracy to preparing anything of substance, be it a government report or a weekly sermon.

We can all feel that in the inescapable tarnish the health report had after the debacle, even as the false citations were hastily replaced with real, published studies. Knowing the human authors hadn’t done the heavy lifting throughout makes us wonder if the research influenced the argument at all or if it merely served the predetermined outcome. After all, if the argument was written — or generated, as the case may be — citing non-existent research, how could research have influenced the argumentation?

Facts can be misused, but it is the human activity of tumbling them around to the smooth polish of a final argument that makes them helpful. For others and ourselves, as well.

We are moving too rapidly from seeing how AI can “augment” us for greater productivity to seeing how AI can just do the work for us. Most of us sense intrinsically this is a bad idea.

We want actual human beings to wrestle with big questions, both because we would be naive to entrust them to machines and because the exploration betters the mind of the one doing it.

Last week, MIT researchers published a worrisome study noting the decrease in brain function among those who depended on AI to help them write an essay. Even if AI gets every fact correct, if it does all the work, our brains will atrophy.

Even brain-flab aside, what of human creativity, compassion and inquisitiveness? It may be mimicked uncannily, but can a machine really do the same thing? We have no reason to think so.

Which brings me to sermons. If you aren’t a pastor or some other leader in a church, you may not have thought that much about sermon preparation. But, like any project fusing creativity and serious research, it demands significant time.

Good pastors start sermon preparations days, if not weeks or months, before a sermon. Sure, there are lazy pastors who ride off of their existing knowledge and natural speaking ability. But, that’s a surefire path to saying profoundly stupid or boring or damaging (or all of the above) things — or plagiarizing to escape those traps.

To delve into ancient texts to discover what they really say about something, think how they apply to the specific people one serves and develop ways to illustrate the results with relatable things from the modern experience — those are all huge time consumers.

The journey to any given Sunday’s message is often months long. I usually try to plan out a series covering a few months, wrestling with the big ideas up front. Then, over the course of weeks, I develop the messages, setting aside parts that apply to future messages. The week of a given message, I try to have an “I could preach it if I had to” message earlier in the week, while I keep praying over it and refining it — trying to see how it really ought to be and not just what could work.

Add in the other things pastors do outside of Sunday — pastoral visits, Bible studies and guiding an organization with an extraordinary number of moving parts for its size — it can be overwhelming and make AI alluring.

A pastor of some celebrity recently tweeted — and then deleted — that he uses AI for major parts of his sermons. He said he takes his basic idea and uses ChatGPT to find scholarly material to support what he wants to say and then engaging stories to liven it up.

Current AI chatbots are certainly capable of writing a sermon — even a decent one. And, given a few ideas for where one thinks it should go, they can replace all that time planning and refining I outlined above. Why not avoid all the painstaking effort?

Now, I’m no luddite on AI, but it struck me that this pastor is shortcircuiting the intensive process precisely where the human element is most important. Taking the “raw data” of what God’s Word says and helping people understand how to apply it to their lives is at the heart of the pastoral job.

I know what those in my congregation are going through. How a point from Scripture might encourage them if drawn out more or how I need to use care on another because it strikes at someplace raw and painful. Human creativity and God’s guidance are both at play in writing a sermon. I am far from unique as a pastor in praying over the process of sermon preparation, believing God guides my work, imperfect as it is.

I can’t express how many times I had a “factually correct and even engaging” message long before Sunday, only to be convicted through prayer and preparation that it wasn’t the right message. Sometimes, even while preaching, I see someone hurting, and, knowing their story, pull from my preparations to take the message in a slightly different — but clearly needed — direction.

AI can draft a message, but without the ideas rattling around in one’s human brain, neither the typical later-in-the-week refinements nor that in-the-moment adaptation are possible.

I don’t think I’ve ever regretted the times I’ve ditched a “it could work” message. To a one, I would regret having kept them.

And, as those MIT researchers saw, the one doing the work benefits, too. I feel the conviction of places I am flawed. I learn things I previously missed. I dust off long ignored tidbits. Each week, I am better for the process.

And, should anyone have a question afterwards? I don’t have to pull out the GPT Preacher on a phone to elaborate.

We are not meant to be enfleshed robot puppets delivering a computer’s thoughts.

But, You do sound like a luddite, Tim! No, seriously, I am not. I love deploying new AI technology, including to serve my work. But, I try to use it in ways that challenge my thinking instead of replacing it.

For example, when my sermon outline — I always preach from bullet points, not a manuscript — is nearly finished, I toss it into xAi’s Grok or Anthropic’s Claude. I don’t ask it to spice it up or finish it for me. No, I ask it to critique. When one has been working on any project for a long time, it is easy to lose perspective. Having AI say, “you don’t seem to explain how you got to this point” is helpful.

When I’ve preached at a church with multiple services, my second pass often benefitted from tweaks based on questions that arose from the first time through. In a sense, AI best applied is not a co-author but that “early service feedback.”

A second key is also crucial for harnessing our uniquely human gifts, staying attuned to God’s leading and for learning from the effort: I don’t assume the AI is correct.

The aforementioned study on the use of AI found that those who first did research without AI, could leverage that familiarity when later interacting with an AI on the same subject. They observed, “AI-supported re-engagement invoked high levels of cognitive integration, memory reactivation, and top-down control.” On the other hand, those groups that depended on AI from the start “reflected reduced [brain] connectivity over time.”

I argue with the AI to see if I can make a cogent point. I feel free to disregard it, too, because in the end, its “opinion” doesn’t matter. AI is like sandpaper. Wisdom is required to know how hard to apply the refining grit to smooth out rough edges without unduly wearing things down.

For this new revolution to be a tool that enhances our work, I need to know what the end result should be and what the correct ideas are. AI can be a fine debate partner, but it shouldn’t be a ghostwriter.

If we are going to flourish as human beings in this new era, this is a point we must embrace. Sermons are just one example of the multitude of human creations that are being impacted by AI, but ought not to be consigned to AI.

My church is counting on hearing what I have prayerfully developed, not what a computer can approximate of prayerful study. If I’m preaching believing God is there and works, no amount of witticism or even factual accuracy can replace time with Him and His Word.

AI can help us make our hard work better. But, sermons should be guided by the Spirit, not the cloud.