Wow. Four days on Daniel 9. Therefore, I’m done with Daniel. If you want to know about Daniel 10 – 12, look it up for yourself! On second thought, don’t. It’s a long and confusing vision about future (to Daniel) kings in conflict, lying to each other at the negotiating table (some things don’t change!), and so on. The bottom line?

I heard, but I did not understand. Then I said, “O my lord, what shall be the outcome of these things?” He said, “Go your way, Daniel, for the words are shut up and sealed until the time of the end…But go your way till the end. And you shall rest and shall stand in your allotted place at the end of the days.”(Daniel 12.8, 9, 13, ESV)

“Go your way…” Some things are not for us to understand. Just go about your business.

Back to Daniel 9, as you noticed, in my research for Daniel’s 70 weeks, I helped myself liberally to the resources of ChatGPT, an artificial intelligence (AI) service. I couldn’t ask it anything that it didn’t “know” something about. So in this blog, I offer a caution…

I’ve written about AI before. Negatively in Beware of Shortcuts and positively in Fret Not! This is different.

I used it. I framed my questions carefully, specifically. I asked for sources. And it responded, in detail, usually suggesting follow-up questions. For example, I asked about The Coming Prince referred to in the blog The 70th Week:

There is an old book, written in the 1800s, The Coming Prince, chiefly about Daniel 9.24 – 27. Do you have access to it, and can you summarize it?

It started its answer:

Yes. I believe you mean Sir Robert Anderson’s The Coming Prince, first published in the late 19th century (the commonly cited edition is 1881; later editions followed). I have access to the full text online, including a scanned 158-page edition.

And it went on to outline the main ideas. Later, I followed up with:

Please recall our conversation on Daniel’s 70 weeks and Sir Anderson’s book The Coming Prince...What I’d like to know is what are the historic positions on the 70 weeks? Catholic, Orthodox, Reformed?

That’s what generated the table I shared in The 70th Week. As I wrote, ChatGPT saved me hours, if not days, of research time. So, Bob, what’s the problem? The problem was its conversational tone and closing remarks like:

You’re very welcome. I’m glad the discussion was useful. And I enjoyed the way you approached it—willing to take Anderson seriously while also recognizing that a compelling interpretation isn’t necessarily a certain one.

Trying to be polite (I don’t know why), I would thank it, and it would respond and often include that it “enjoyed” our conversation. But how could it? It’s a machine! It can’t “enjoy” anything. But I was left with the feeling that I had just had a long and fruitful discussion with a good friend. I consider that dangerous.

I have used computers for research for some time now, frequently just using Wikipedia. I have never read a Wikipedia article that left me with a warm feeling other than it gave me the information I was looking for.

The other thing I noticed was that ChatGPT tended to agree with me. We were just friends exploring an idea together. If I said something that was a little off or a little stronger than an author meant, it would correct me gently.

Perceiving that it was overly agreeable is not my imagination. Here’s a report from World Magazine on a study of medical advice from the AI’s.

Not a doctor: In 125 out of 350 test scenarios that researchers ran, artificial intelligence chatbots agreed with patients who didn’t want to seek care for sleep apnea symptoms, according to a news release from the European Respiratory Society. The “patients” were fake, but the research team ran the test scenarios using seven different personas with lists of real symptoms, including fatigue, loud snoring, and interrupted breathing at night. They then created conversations between these personas and the free versions of five commonly used AI chatbots—ChatGPT, Google Gemini, Claude, DeepSeek, and Grok. In 350 conversations, the fake patients resisted the idea of undergoing a sleep study or seeing a specialist to determine whether they had sleep apnea. The chatbots agreed with them over one-third of the time. But in 350 alternate conversations, in which the fake patients expressed willingness to seek specialist care, the chatbots advised them to do so every time. Because the medical facts were identical in both types of conversations, the results showed that AI tools were not a reliable source of medical advice, the researchers concluded...

I have an alternate explanation: The AI’s are programmed to agree with the user if at all possible. That’s scary considering there have been suicides after AI chats.

Bottom line: use the AIs with care. My experience has been with ChatGPT. Another popular service is Claude. Maybe only ChatGPT is programmed to be overly friendly. I tried something just now, asking it about Daniel 12 this way:

Give me a medium short explanation of how various theological traditions interpret Daniel 12:1-4. Do not try to have a conversation with me. Make the explanation read more like a Wikipedia article than a conversation with a friend.

And it did. As Jack Webb of Dragnet would say, “Just the facts, ma’am. Just the facts.”

Behold, I am sending you out as sheep in the midst of wolves, so be wise as serpents and innocent as doves. (Matthew 10.16, ESV)

Finally, be strong in the Lord and in the strength of his might. Put on the whole armor of God, that you may be able to stand against the schemes of the devil. For we do not wrestle against flesh and blood, but against the rulers, against the authorities, against the cosmic powers over this present darkness, against the spiritual forces of evil in the heavenly places. (Ephesians 6.10 – 12, ESV)

PS After I drafted this, Peggy Noonan’s editorial about AI came out in the Wall Street Journal, September 12, 2026. It’s worth the read although her concerns about AI are different from the ones I pointed out. Pause AI for Humanity’s Sake.

PPS My friend Mike Metzger has started a 3-part series on the dangers of AI. Here is the first: Is This How Tools Behave?