The Economist ran a piece on cognitive surrender. Wharton wrote a paper. The popular take goes like this: AI is making us dumber. We’re outsourcing our brains. The kids will grow up unable to think.

I read it. Nodded. Then asked the question I should’ve asked first.

How much real thinking were we doing before? You know I love thought experiments. Let’s look at an everyday decision that has to be made by thousands every year.

A young family buying a car. The textbook version goes: identify need, list priorities, research options, compare, test drive, negotiate, buy. Six clean steps. Rational actors making a rational choice.

Nobody buys a car like that. Nobody.

Here’s what actually happens. Mom saw a Honda Pilot in the carpool line and liked it. Dad’s college buddy just bought a Tahoe and hasn’t shut up about it. The kids saw a commercial. They walk onto a lot. The salesman opens with a monthly payment, not a price. Cox Automotive data shows the average buyer visits about two dealerships before signing. Two. They sit in three cars. They buy one of those three, financed for 72 months. Then they tell their friends they “did the research.”

No AI involved. Pure analog rot.

Now add ChatGPT. Same family. They ask “best family SUV under $50K.” The model gives a confident, fluent ranking. They like that the Pilot is on the list. They ask follow-ups. The model is patient, never tired, never pushy. It feels like research. They visit one or two dealerships. They buy.

Same family. Same Pilot. Same financing.

The down payment and the monthly payment ceiling were the actual gatekeepers the whole time. Everything else is edge selection inside an already-narrow box. The “decision” was made the day they figured out they could afford a $580 payment.

Now here’s the part the popular narrative gets half right. The Wharton paper found something specific. When the AI gave correct answers, people using it outperformed people working alone. When the AI gave wrong answers, the AI users did worse than people working alone. So AI isn’t bad in every case. It’s an amplifier. Hands on the wheel, you go faster. Hands off, you could crash harder, or end up the same place without using it.

But that’s a different problem than “AI is making us dumber.” That’s a problem about what we bring to the tool.

So the question isn’t whether AI is making us dumber. It’s whether we ever had the floor we think we lost on the kinds of decisions that drive most of life. What to buy, where to live, who to marry, which job to take. Identity decisions wearing analytical clothes.

Here’s the part that bothers me. I know my own mind. I want to be Spock.🖖 I’d take green blood and pointed ears tomorrow if the deal was honest thinking. I’m not Spock. Neither are you.

In 2010 two cognitive scientists at Lund University set up a tasting booth at a supermarket. Two jams. Pick the one you like better. The shopper picks. The researchers hand them a spoonful of their pick and ask why they chose it. People give thoughtful, elaborate answers. They liked the texture. The sweetness was just right. The flavor reminded them of something.

The catch: the researchers had secretly switched the jars. The shopper was tasting the jam they rejected, explaining in detail why they chose it. Even with cinnamon-apple swapped for bitter grapefruit, fewer than half of people noticed. The rest produced fluent reasons for a choice they never made.

This wasn’t prejudice or politics or any hot-button issue. This was jam. And we still made up the story.

The science has a name for this. Choice blindness. We work backward from outcomes and construct the reasoning after the fact. The reasoning sounds airtight. It’s a story the brain tells the brain. And we believe ourselves.

Filter bubble to AI bubble. Same mind. Smoother delivery.

The kid question is harder. A child growing up with AI as the default thinking tool doesn’t have a “before” state where the biases ran without help. That’s a real concern and I don’t have a clean answer for it. The chess research already shows on-demand AI tutoring produces less than half the skill gain of limited AI tutoring. Worth watching.

For the rest of us, though, the question is what to do.

You stop trusting yourself alone in a room with your LLM’s outputs. You are a genius after all. ChatGPT said so.

The single most useful thing I’ve done with AI is not the underwriting work. It’s not the agent systems. It’s red-teaming my own thinking. I write something I believe. I tell the model to find every hole, every assumption I haven’t earned, every place I’m flattering myself. Then I read it before I can argue back.

It catches things I can’t catch. The model is just smarter than me on specific things. It also has no investment in me being right.

Yes, I can rationalize past the AI’s pushback, too. I do. But it’s harder when the criticism is laid out so eloquently, where every objection has to fight my ego on the way out.

The move isn’t “use AI for everything,” and it isn’t “fear AI.” Use it where your own mind has a known blind spot. Confirmation bias, identity defense, the urge to feel finished. Let it push.

The easy version of this story wants AI to be the villain. It’s tidy. It blames AI for a problem older than AI.

The thinking we thought we were doing wasn’t really thinking. AI didn’t take it from us. AI just turned the lights on.