Today I want to talk about a somewhat abstract topic, but one that's been feeling increasingly important lately: as AI gets smarter and smarter, will "not knowing" become a scarce commodity?

The trigger was dinner this week with a friend who works in product. He said that during requirement reviews these days, he's growing less and less willing to say "I don't know." Because someone next to him always chimes in: "Just ask AI."

At first it seemed great — AI really can answer a lot of questions that used to take forever to look up. But after a while, something feels off: when "not knowing" becomes a blank that AI can fill in anytime, people start to default to "you shouldn't not know."

Don't know a technical term? Ask AI. Don't know a specific data point? Ask AI. Don't know how a competitor's feature works? Ask AI. AI keeps raising the bar for "not knowing" higher and higher — so high you almost feel embarrassed to say "I don't know."

"Not Knowing" Used to Be Important

Not knowing isn't something to be ashamed of. On the contrary, "knowing that you don't know" is the starting point of thinking.

Have you ever had this experience? Faced with a problem, you mull it over for ages, your mind goes blank — you just don't know the answer. That state of "not knowing" is uncomfortable, but it has value. It forces you to think, to research, to discuss with people, to try and fail. In that process, you build your own understanding of the problem.

Einstein famously said: "If I had an hour to solve a problem, I'd spend 55 minutes thinking about the problem and 5 minutes thinking about solutions." What's the premise of that? You first have to "not know the answer" — only then do you have the motivation to spend 55 minutes thinking about the problem itself.

What if the answer comes out in one second? Would you still spend 55 minutes thinking about the problem itself? Probably not. You'd just take the answer and use it. You save time, but you also skip the deep thinking process that "not knowing" brings.

With AI's Answers, You Don't Know What It "Doesn't Know"

A more subtle problem: AI knows a lot, but you don't know what it "doesn't know."

Human "not knowing" is transparent. When I say "I don't know," you know I just don't know. You can follow up: "Well, what do you know?" "What direction would you guess?" We can explore back and forth at the boundary of "known" and "unknown."

AI's "not knowing" is opaque. Sometimes it talks complete nonsense with a straight face. Sometimes it misses critical information. Sometimes it treats the most probable answer as the correct answer. You don't know where its cognitive boundaries are — you don't know which things it truly understands, which it's just cobbling together, and which it's completely making up.

What does this mean? It means when you hand "not knowing" over to AI to fill in, you're essentially putting your own judgment up as collateral. You need the ability to judge whether AI's answer is right. You need to know where it might go wrong. You need your own "cognitive coordinate system" to calibrate AI's output.

And those abilities? They come precisely from all those moments of "not knowing" you've lived through before.

Protect Your "Not Knowing"

All this said, I'm not saying AI is bad. AI is great, it's useful, I use it every day. But I increasingly feel that in the age of AI, "not knowing" might become a scarce resource that needs active protection.

Specifically, there are a few things I think are worth doing:

First, set aside some time and scenarios where you "don't use AI." For example, when writing, draft it yourself first, then let AI help polish — don't let AI write it from the start. When working on a proposal, figure out the framework yourself first, then let AI fill in the details. Preserve that process of "building a framework from scratch" — that process is your thinking ability.

Second, when you encounter a problem, think about it yourself for 10 minutes before asking AI. 10 minutes isn't much, but it's enough to run the problem through your head once and build a preliminary cognitive framework. With that framework, when you look at AI's answer, you'll know if it's right, if it missed anything, and where to dig deeper. You go from "passively receiving answers" to "actively verifying answers."

Third, say "I don't know" more often. Don't be afraid to say you don't know. Not knowing is just not knowing — it's not embarrassing. What's actually embarrassing is pretending to know when you don't — whether you're doing the pretending yourself, or AI is doing it for you.

Fourth, ask "why" more, ask "what" less. "What" questions — AI answers them better than you, faster than you, more thoroughly than you. But "why" questions often require deep thinking and judgment, which AI isn't good at yet. Asking more "why" questions exercises the muscles that AI can't yet replace.

One last thought. AI makes getting answers easier and easier, and that's a good thing. But answers are cheap — thinking is rare. In an era of answer abundance, the ability to ask good questions, to tolerate "not knowing," to think independently — those abilities will become more and more valuable.

Protect your "not knowing." It's the source of your thinking.

Have a great weekend.