 |
| The day the line disappears, you find out how much you still remember how to do yourself. |
On a modern airliner, the pilot flying you across the country actively holds the controls for about three minutes. The rest of the flight, the autopilot flies and the human watches. Mostly, this is fine. Mostly.
Then there's Air France 447. In 2009, over the Atlantic, the autopilot hit some bad sensor data and handed control back to the crew. Faced with a fairly ordinary stall, pilots who had spent years monitoring screens instead of flying couldn't recover. The plane fell for almost four minutes. 228 people died.
Pilots have a name for the habit that sets this up. They call it being a “child of the magenta line”: following the glowing route on the screen at full speed, hands in your lap, no longer quite sure how the machine gets there.
That's the real risk with AI. Not that a robot seizes control in one dramatic moment. It's that you quietly hand off one small thing at a time, until the day the line disappears and you've forgotten how to fly.
The takeover is made of little things
Nobody decides to let AI run their life. You just let it write one email. Then the tricky reply. Then the birthday caption, the meal plan, the thing you were going to say to your friend. Each handoff feels too small to matter. That's exactly why it works.
And it's happening at scale. Using ChatGPT roughly doubled among US adults in a single year. Here's the tell, though. Only 13% of Americans feel they have real control over how AI shows up in their lives, and 61% say they want more. People can feel the wheel slipping. They just can't quite find it.
And it doesn't stop at admin. People now quietly ask AI to draft the apology, the condolence, the message that was supposed to come from them. The trouble is that effort is how we read care. When a university got caught using ChatGPT to write a condolence note after a campus tragedy, the backlash was instant — because everyone understood that outsourcing the words outsources the meaning too. The small handoffs are rarely as small as they feel.
Is AI actually making us dumber?
Honest answer: the AI-specific evidence is still early, but it points one way, and we've already got decades of proof for the mechanism underneath it. When you offload a mental skill, that skill fades. This isn't a hunch. It's one of the more reliable findings in cognitive science.
Habitual GPS users have measurably worse spatial memory, and the heavier the use, the steeper the decline. There's the “Google effect” too: once we know a fact is a click away, we remember where to find it instead of the fact itself. Our memory always had limits, which is why we lean on tools (that's Miller's Law). The question is which skills we can afford to hand over.
Now the AI-specific studies are landing in the same place. Microsoft and Carnegie Mellon surveyed knowledge workers and found that the more people trusted the AI, the less critical thinking they did. An MIT team wired people up to EEG while they wrote essays with and without ChatGPT; the AI group showed the weakest brain engagement, and many couldn't quote a single line of the essay they'd just “written.” (Small study, adults only, not yet peer-reviewed, so treat it as a warning light, not a verdict.) The pattern keeps repeating: the output gets better while the person gets weaker.
We've traded skills before, and the critics were half right
The reassuring line you'll hear is that Socrates feared writing and turned out to be wrong, so relax. Except he wasn't wrong, not entirely. He predicted writing would weaken memory, and it did. The trained oral memory that once let people recite whole epics is basically extinct now. We decided the trade was worth it, and it was.
But notice the shape of it. The panic loses the big argument and wins a small, specific one. Writing didn't make us stupid; it retired a particular skill. GPS didn't make us stupid; it dulled our sense of direction. So the useful question was never “will AI ruin us.” It's narrower and more honest: which exact capability am I about to trade, and did I actually agree to the price?
The real risk isn't stupidity. It's sameness.
Here's the part almost nobody talks about. The danger isn't a world full of idiots. It's a world full of identical thoughts.
When researchers study AI-assisted writing, the outputs converge. Same ideas, sometimes the same phrasing, everyone rounding toward one average voice, an “artificial hivemind.” Any single AI answer might beat the average person, but ask a thousand people the same way and their ideas collapse into the same handful. A Georgetown neuroscientist, reading college essays after ChatGPT arrived, put it more bluntly: “Suddenly everyone's life got a lot more similar.”
Offload your thinking and you don't just get a little duller. You lose the one thing that was actually yours: your voice. And you start to sound like everyone else who typed the same prompt.
So should I just stop using AI?
No. That's neither the point nor realistic. The same tool can sharpen you or soften you, and the difference is entirely in how you hold it.
A University of Pennsylvania study made this almost too clean. Students using AI to study got 48% more practice problems right, then scored 17% worse on the real test once the AI was taken away, compared to students who never used it. Same tool. Opposite result. The only variable was whether they used it to think or to skip thinking.
AI can be a bicycle for your mind or an autopilot for it. One amplifies your own effort; the other flies while you quietly forget how. We made the augment-don't-replace case in full in How AI Can Make You an Almost Invincible Human, and the everyday version of it in How to Use AI in Your Daily Life.
The Autopilot Test
So before you hand something to AI, run it through four quick questions. Think of it as a preflight check for your own judgment.
Verification. Could I check this answer if I had to? If you can't tell whether the output is right, you're not delegating. You're gambling.
Struggle. Am I skipping the hard part, or just the boring part? Offload tedium all day. But the productive struggle is where your capability actually lives, so be careful giving that away.
Sameness. Would my answer be different from everyone else's who asked the same AI the same thing? If not, you've handed your voice to the hivemind.
Blackout. If the tool vanished tomorrow, could I still do this? Pilots who fail that question are the ones who crash.
Then keep some things deliberately manual, your own version of a pilot's manual-flying hours. Write the first draft yourself, then let AI poke holes in it. Ask it “what did I miss?” instead of “do this for me.” Keep a human firmly in the loop, which, as we argued in AI is scary, but it needs humans to coexist, is the whole point.
Convenience is the tax on capability
Every time you offload something, you pay a little capability for a little convenience. Sometimes that's a fantastic deal. Nobody needs to do long division by hand or memorize a hundred phone numbers anymore. But sometimes you're quietly selling off something you'll badly want back, and the tax is invisible right up until the bill arrives, usually at the worst possible moment, when the line disappears and you actually have to fly.
So use AI. Use it a lot. Just don't become a child of the magenta line. Automate the route, never the destination, and fly manually often enough that you still can.
If this way of thinking is your kind of thing, subscribe to the UX Breadcrumbs newsletter for more plain-English takes on living and building alongside AI without handing it the wheel.
Comments
Post a Comment