There is a specific kind of embarrassment that Google used to produce.
You had a question, something half-formed, and you sat in front of the search bar, not knowing what to type.
The concept was somewhere in your head, but the words refused to come.
You tried three different phrases. None of them landed.
You got results that were adjacent, off-topic, pointing somewhere you hadn’t meant to go, and you read them anyway, because you were already there.
That friction had a name, though you probably didn’t call it anything at the time.
It was the gap between knowing you don’t know something and knowing what it is you don’t know.
Bridging it required a kind of intellectual contortion.
You had to circle the problem from the outside, trying synonyms, following tangents, landing in rooms you hadn’t booked.
Sometimes it took hours, sometimes days, and then, somewhere in the wandering, a word appeared, a term, a door, and behind it was the exact world you’d been groping toward.
Nobody celebrated that moment as learning, it felt like inefficiency.
Now consider what happens when you open an AI chat.
You type the half-formed thought, exactly as it sits in your head, vague and imprecise and sprawling, and the machine answers.
Not the question you asked, but the question you probably meant.
It maps the territory before you’ve taken a single step, it gives you the vocabulary before you’ve earned it.
The experience is extraordinary, I use it constantly, there is no honest version of this argument that pretends otherwise.
But something went missing in the convenience, and I’m not sure we’ve measured it properly.
Kahneman spent decades documenting the difference between thinking that happens automatically and thinking that requires effort.
The first kind is fast, pattern-matching, associative.
The second is slow, deliberate, and costly.
We default to the first whenever we can.
The second only engages when the first fails, when the situation is novel enough or difficult enough that shortcuts don’t work.
The old Google problem, the circling and the searching and the accidental discoveries, forced the second system on.
Not because it was good pedagogy, but because you had no choice.
What AI does, structurally, is make the first system sufficient.
You no longer need to labor toward the words, now the words come to you.
The map arrives before the territory, and the map is usually good enough that you never feel the absence of the territory.
This matters more if you think the act of formulation is not just the path to knowledge but part of the knowledge itself.
Socrates did not ask questions because he lacked answers.
He asked questions because asking was the discipline.
The question was not a vehicle, it was the destination.
When Socrates cornered someone in the agora and dismantled their argument piece by piece, he wasn’t being cruel.
He was demonstrating that the premise was wrong, that the original question was badly formed.
That you had been trying to open a door that didn’t exist, and what you actually needed was to find a different wall.
The quality of the question determines everything downstream.
A bad question does not produce a bad answer, it produces an answer to a different problem, delivered with perfect confidence, and you only discover the mismatch later, when the solution doesn’t fit.
Anyone who has ever spent three weeks building the wrong thing, or three months in the wrong relationship, or three years in the wrong career, knows this particular flavor of waste.
The question at the beginning was off by a degree, by the end, the distance is enormous.
There is a version of this in leadership, too.
A conductor does not need to play every instrument.
What a conductor needs is the ability to hear when something is off, to ask the right thing of the right person at the right moment, to know what question will expose the problem.
Great leaders are often described as the people in the room who ask the best questions, not the ones who know the most.
The knowing is distributed, the questioning is the skill.
And the questioning is a skill precisely because it requires something before the question: a genuine mapping of the situation, its variables, what you know and what you don’t, what you’re assuming without realizing it.
You have to understand the shape of your ignorance before you can ask usefully about it.
This is also, not coincidentally, a form of humility.
Not the performed kind, not the disclaimer kind, but the structural kind.
The kind that says: I might be framing this wrong.
The question I think I’m asking might not be the one worth asking.
Let me hold it loosely for a moment.
That pause, that willingness to interrogate the question before sending it forward, is what separates someone who finds answers from someone who finds the right ones.
But you cannot ask the right question if you have never been through the process of not knowing how to ask.
You cannot develop the instinct for the precise question if you have always had the imprecise one answered without friction.
The skill atrophies, not dramatically, not all at once, just quietly, the way you stop using a word you used to know.
The research is beginning to catch up to the intuition.
Studies over the last year have found that people who rely heavily on AI tools score worse on measures of critical thinking, not because AI is malicious but because the friction that produced those skills has been removed.
The work was not just annoying, the work was the work.
I don’t have a clean resolution for this.
I’m not suggesting you go back to aimless Googling for the sake of the struggle.
What I do think is worth keeping visible is the specific cost that nobody lists when they describe how much easier everything has become.
The long route to the word was a form of thinking.
It was not inefficiency, it was the process by which you came to understand what you were actually asking.
When you skip it, you arrive faster.
You might also arrive somewhere slightly different from where you thought you were going, and never notice.
I Can See You
