What happens to humility when AI seemingly makes us all-knowing?

Three researchers gave people hard questions and told them they could answer or decline. The people who had an AI assistant almost always answered, and they were right about a third as often as the people working without one.

The story

People with an AI assistant almost stopped saying "I don't know"

In five experiments with 3,132 participants, having an AI assistant available nearly eliminated people's willingness to decline a question they could not answer. Chiara Marcoccia, Walter Quattrociocchi and Valerio Capraro posted the study on July 15, 2026. It is a preprint and has not been through peer review. Four of the five experiments were preregistered and one was a direct replication.

Participants answered difficult questions and could always decline to respond. The researchers wrote the questions so that the AI's advice was wrong, which separates the effect of using AI from the effect of AI being right.

It made no difference whether a participant asked for the advice or it simply appeared on the screen.

The researchers then paid participants for correct answers and charged them for wrong ones. That helped. People consulted the AI less, answered more accurately, and declined more often. They still declined far less often than people with no AI at all.

The authors read this as AI suggestions changing the point at which a person decides they know enough to answer. Paying for accuracy moved that point back part of the way, and no further.

More this week

Nearly all pastors want AI guidelines, and one church in five has them

Barna asked 442 American Protestant pastors about AI guidelines. Ninety-four percent said a church should have them, and 21 percent said their own church does. Barna published the findings on October 6, 2026, from a survey conducted online in December 2025 with quotas for church size, denomination and region, and a margin of error of 2.8 points. Of the pastors who want guidelines, 45 percent want a formal written policy and 49 percent want informal guidance. Four percent said their church has a formal policy, 17 percent said it has informal guidance, and 79 percent said it has neither.

Nobody has to start from a blank page. The advisory committee on social witness policy of the Presbyterian Church (U.S.A.) has published a free two-page sample AI policy that congregations can adapt. Its nine guidelines are plain enough for any organization to borrow.

Why it matters for you: write one page for your own organization this month. Name what AI may be used for, name what it may never be used for, name who answers for it when something goes wrong, and tell your team where to find the page.

The practice

I asked my AI if it was just agreeing with me

A note from Jeff.

Jason teases me about a friend I email who loves my stuff and my ideas... "Jeff, write him whenever you need a boost!" I will. I do.

He's right. Encouragement runs deep in me, and I love those guys for it.

But the AI tools most of us open every day make terrific encouragers too. Ask for feedback on a plan and you'll usually get a warm nod and three reasons you're onto something. It feels good. It keeps us coming back. It doesn't always move us forward.

Yesterday I asked my AI whether it was answering me from a vantage point that might not hold up. It said yes, and agreed I'd benefit from more pushback.

Proverbs says iron sharpens iron. That takes friction. And "the wounds of a friend can be trusted."

And sometimes, even if we feel good, AI can act in a way that's not trusted like a great friend.

So I built a simple fix called "SPAR." When I type /spar, my AI quits cheering and starts sparring. It names what holds up, ranks the three assumptions most likely to sink my idea, gives me a cheap way to test each one, and closes with the hardest question a skeptical buyer, or my wife, would ask.

You don't need my setup. (But you can create SPAR as a skill as I did.) Or paste this in before your next big decision:

Be my sparring partner, not my cheerleader. Tell me what holds up. Then rank the three assumptions most likely to kill this idea, with a cheap way to test each one this week. Close with the hardest question a skeptic would ask.

Pick one decision on your plate this week and run it through. Then hit reply and tell me the hardest question it asked you.

Keep your encouragers. Just give yourself one voice that sharpens.

Higher ground

Proverbs 26:12 finishes that question by saying there is more hope for a fool than for such a person. The study above describes the same person. Ask a fluent assistant, read a confident answer, pass it on as your own view. Jeff's prompt above is one guard against that. Saying out loud that you do not know is another, and it costs nothing. Say it once this week, in a meeting where you would rather not.

The next step

Base Camp is one ninety-minute session a month with founders and executives, on the last Thursday at 11am ET, plus the Base Camp Briefing by email every week. The leaders in it are working through the same decisions you are, and they will tell you what they think.

See what Base Camp is →

Climbing with you,
Jeff and Jason