An ethical question: How far can pastors go (as a routine) with AI/Internet/published sermons in preparing their own messages?

I can’t even repreach my own sermons, so I can’t imagine how I’d preach someone else’s. Well, I can revamp an old message of mine and preach the new version. At best, I could theoretically do that with someone else’s sermon.

The ethics: I see two factors (so far). It’s the same as the plagiarism discussions. Representing someone else’s work as your own is dishonest and also sort of theft. Flattering theft, I suppose, but still theft. So factor 1 is the honesty: be open about what you’re doing.

Factor 2: pastoral disciplines. It’s good for the Christian mind and soul to wrestle with the material directly as much as possible. It’s good for the leader’s communication skills to work through the process of organizing his thoughts for coherence but also for delivery—and in a way that fits who he is. There is no such thing as vicarious exercise, and you can’t build the muscles of clear thought, effective study, and prep for delivery using someone else’s stuff whole—acknowledged or not.

I can think of special situations where it would make sense, like the one Ed mentioned. A once in a great while thing doesn’t violate pastoral disciplines. But not being open about it violates honesty.

All this applies to AI. It’s a bunch of other people’s material reprocessed by an algorithm. Still not your own work. I’m talking about the generative part. As research tool, there are no special rules as far as I can tell. It’s as likely to err as humans in general.

I have occasionally found an outline in a commentary that worked with the message I wanted to preach. (Usually alliterative!!!) I acknowledge the source of the outline in my notes and verbally, at least I hope so! Can't remember every instance, but I've tried to make sure that is my practice.

I think you could use AI for parts - suppose you found a lengthy article that you would like to summarize. It would be impracticable to read the whole thing, you might try to summarize it yourself, but why not use an AI to do the summarizing for you? You would have to already have read the article so that you could be sure the AI did a good job, and you could even read the AI summary as a part of your message. You should again acknowledge the source and the source of the summary.

As Aaron mentioned, I find it hard to redo one of my own messages without making changes. Hopefully the changes are an improvement!

I wouldn't use AI to write a whole sermon, or use someone else's whole sermon. But I see nothing wrong with using ideas (including outline points) with acknowledgment.

I use AI (ChatGPT or Logos's Study Assistant) mostly as a research assistant and copy editor.

I occasionally meet Christians who seem to think that there is inherent virtue in doing something the hard way. It’s an attitude and a set of choices, not usually claim that the harder way is better just because it’s harder. They might say “The hard way is better,” but they wouldn’t embrace a more precise way of saying the same thing: “Inefficiency is a virtue.”

My thinking on AI and other—more centuries old—aids to Bible study and research includes this: Efficiency is a virtue. There is nothing godly about wasting time and energy. It’s poor stewardship.

So the question is, “What am I trying to achieve?” And the follow-up is, “How can I achieve it with less wasted effort?” It’s not that we should prefer ease over struggle, but we should prefer using what we’ve been given well vs. using it wastefully.

I won’t upgrade to Logos’ subscription model (the one with AI) until I have no choice. I also turned off Co-Pilot across all platforms. I really, really hate it. I fear that Logos AI tools will produce a very shallow crop of future ministers.

I fear that Logos AI tools will produce a very shallow crop of future ministers.

If they are not being trained to exegete Scripture in seminary, I agree. My late Greek professor lamented his students relying on Bible software to help them with exegesis. That’s when Logos was at version 4.

His expectation was that they would do their own translation and parsing without relying on Bible software.

I still manually translate my passages (Greek or Hebrew) as part of my exegetical process. I enjoy the challenge, and it keeps my languages fresh.

https://www.thegospelcoalition.org/article/agentic-ai-pastor-church/

Here's a key takeaway: "I tell my students to consult commentaries after they’ve done their own work, not before. The same principle applies to AI. Do your own thinking first. Then consult the tools."

There are new disciplines to learn for AI use. One of the most challenging might be the discipline of not simply believing what it says. They are not all equally error prone. My paid Claude version is wrong like 90% less often than Gemini free, for example. Gemini is like a coin toss, depending on the topic, whether it will confidently assert something 180 degrees off the truth.

But this discipline is both simple and difficult. The simple part: Humans are wrong quite frequently, and we expect that. So with a lot of what we hear/read, we do quick provisional probability-of-correctness tirage and then follow up later if that might be useful. But the AI’s frequently ‘speak’ with a level of confidence that makes them sound expert, when they are really just average at best, when it comes to simulating thought. Of course, when it comes to pulling together a ton of info, they are well above average. But ‘understanding’ a pile of data is a different thing.

So, easy: like humans they are often wrong.

More difficult: they tend to sound very trustworthy even when they are completely wrong. With humans there are often contextual flags that help us evaluate trustworthiness (though trust is always provisional at best with humans). With the AI’s the flags are not there or are more subtle.

But not always: there are things you can to do in durable memory with your instructions. With Claude, I’ve told it double check everything, push back on my reasoning, and question itself—depending on the project. So, if you are using a reasonably good model, you can tell it to always assess the probability of its ‘facts,’ and tell you when uncertainty is high.

But in my experience, Claude does this a lot more by default.

Gemini used to, but it’s gotten a lot stupider lately in the free version.

ChatGPT: I don’t use it much. A bit of work in the Copilot incarnation on the job, because that’s the only one we’re allowed to use there… it is possible to get good work out of it, but it takes quite a lot of prompting and reprompting. I have found some tasks that it can repeat now with ‘agents’ and do them well enough to actually save me time. But it took a while to get it there. My experiences with ChatGPT/Copilot have not, on the whole, been positive.

I haven’t used AI for Bible study, so maybe this is a failure of imagination on my part, but I can’t see how an AI would be interfering with thoughtful study. I mean, if you had a Bible scholar sitting next to you and you said “What does this text mean?” Would you just take his word for it or would you ask, “Why do you think so?” and evaluate the argument? So if an AI told me what it means, I wouldn’t just take its word for it. I’d be, “Tell me why.” And then see if the facts are straight and reasoning is sound.

It doesn’t seem that different to me. But maybe the workflow is very different from what I’m imagining? If the AI is only pulling together what the books say and summarizing for you, it’s just doing what you would do anyway, only faster.

… and a lot of guys can’t really outline. Might as well ask the AI to fix your outline. But it’s still worth asking, “Why did you change it in this way?” You might learn how to outline.

I haven't used AI much in Logos. The main uses I've made of it are in natural language searches as opposed to a specific search using the special Logos operators. Sometimes you can find something helpful the AI way that you can't (or can't figure out the operators for) in the regular search.

It also can summarize an article or selection for you. I think I used that once to just see how it worked. I think this approach can help you decide if you want to read the whole thing or not.

Probably there are other uses, I don't use half the tools in Logos. The thing is, it's just another tool. It won't atrophy your brain to use it.

If we want to get back to the laborious methods of yesteryear, lets dust off the quill pens and invest in ink, consulting only paper resources, and handwriting all our sermons. I find the many ongoing cavalcade of articles about the dangers of AI to be overwrought and not all that helpful.

It's a tool. Use it or not, but if you decide to use it, learn how to use it and it will pay off for you.

I’ve used the Study Assistant since it was released. I think it’s a very helpful tool; however, I’ve been rather intrigued with how it selects the books it does during its search. I did some work in pneumatology recently and used the Study Assistant for research. I have several books on the Holy Spirit in my library as well as systematic theology texts, and it ends up pulling its results from obscure sources in my library that I’d never use.

I had to create a collection with just those resources then limit the search to that collection for the study assistant to use them.

So, I’m not sure how the algorithm selects which books to use, but it needs some tuning.

Yes, selection and prioritizing is definitely a key factor in ‘manual’ vs. ‘automated’ research. I have the same problem in my older no-AI Logos, though, just doing searches. There is a lot I am not interested in looking at. There are ways to hide it from searches, and I’ve done that occasionally, but then sometimes the “not a source I knew about” search hit surfaces something helpful I would probably not have known about otherwise.

So I don’t want to completely disable the broader scope.

When I’ve used various AIs for research purposes (never in Logos yet), I have sometimes just prompted it to emphasize this or that or omit this or that kind of source. With varying success. It usually does improve after a nudge or two. Seems like whatever AI is in Logos now should be promptable that way? Or will be eventually? “Focus on the resources I use most”? I wonder if it is capable of distinguishing conservative and evangelical and categories like Reformed or dispensational or classic. Surely it would recognize patristic/fathers, right?

I think the Logos AI doesn't limit itself in the prompt, you can limit it by selecting a collection, say "Church Fathers" or "Biblotheca Sacra" or similar (you have to set up your collections ahead of time).

That's with just a few quick attempts. I could be that I am not good at writing prompts, too!!

The Study Assistant has a max resource limit. In using it for some Revelation research, I discovered it is programmatically limited to 5 resources.

So, it will somehow select the top five resources (if it determines there are five or more) then summarize them. Again, its selection criteria are hit or miss. You have to guide it to the right resources.

I also subscribe to the academic journal package. But, because the study assistant is self-limited, it doesn’t seem to reference journal articles. I have to find articles through manual searches.

In other words, Don’t think the Study Assistant combs through your entire library when it provides its answers.

However, the search feature does look at all your resources.

I will likely never use AI for research of any sort.

One intriguing thing = I asked ChatGPT to review a sermon transcript of mine for fidelity to the passage, and conformity to the New Hampshire Confession. It gave me a well-reasoned, helpful 1000 word evaluation. It frightened me a bit, but the critiques were solid!

One intriguing thing = I asked ChatGPT to review a sermon transcript of mine for fidelity to the passage, and conformity to the New Hampshire Confession. It gave me a well-reasoned, helpful 1000 word evaluation. It frightened me a bit, but the critiques were solid!

I have done something similar with the Study Assistant. I've given it my sermonic points and asked it to evaluate how they align with the particular passage under consideration. It, too, was helpful in pointing out areas of misalignment.

Based on my use of AI (whether ChatGPT or Study Assistant), I would say if you understand the limitations as well as the benefits of AI, it can be a powerful tool.

I’ve also gotten some useful output pasting in outlines or transcripts and giving it a “Tell me where a skeptic would see weaknesses in my argument” and the like.

They are usually also quite ‘happy’ to argue with themselves, so you can do something like: “OK, pretend you’re on the other side of a debate and tell me what’s weak in everything you just told me.”

I’ve only done this with Gemini and Claude. I expect ChatGPT is also pretty good at critiques. It makes little/no difference to the AI whether the ‘target’ is somebody’s work or is their own.