Use AI Without Getting Dumber

In an MIT study, 83% of people who wrote an essay with ChatGPT could not quote a single sentence of it minutes later.

Hi friend,

Spend some time reading what people actually write online, as opposed to what they post for an audience, and you start to notice the same worry coming up over and over this past year.

A software engineer on Substack says he has started to feel he is letting AI do his thinking for him, and no longer has the discipline to stop.

A writer describes asking ChatGPT how to solve a small problem she once would have worked out alone in five minutes, and feeling faintly ashamed afterwards.

Developers talk about colleagues who ship code all day with an assistant open and can’t fix a simple bug once it’s closed.

None of these people are against the technology, and most of them use it every day. What they’re describing is a quieter suspicion that something they used to be able to do is getting harder, and that they didn’t notice when it started.

There’s now some research on whether that suspicion is right, and the most useful part of it is not the part everyone shared.

The experiment

In 2025 a team at the MIT Media Lab led by Nataliya Kosmyna asked 54 students and young staff from universities around Boston to write short essays across several sessions over four months, while wearing EEG caps that recorded activity across the brain. One group wrote with ChatGPT, one used a search engine but no AI, and one wrote with nothing but their own heads.

The brain-only group showed the strongest, most widely connected activity. The search group sat in the middle. The ChatGPT group showed the weakest.

Then, straight after the first session, the researchers asked everyone one simple question: can you quote a sentence from the essay you just wrote?

15 out of 18

ChatGPT writers could not correctly quote their own essay.
In the other two groups, only 2 out of 18 failed.

The ChatGPT writers also felt the least ownership of what they had written, which makes sense once you realise that, in a fairly literal way, much of what was on the page had never passed through their minds.

Read this with care. The study is a preprint still going through peer review, the sample is small, and it looked at one task with one tool. The lead author has publicly asked people to stop calling it “brain damage.” It shows what happens during a specific kind of work, which is narrower than the headlines, and more useful.

The finding nobody shared

In the fourth session, the researchers swapped the groups.

People who had been writing with ChatGPT now had to write without it, and people who had been writing on their own were given ChatGPT for the first time. Those who lost the tool showed weaker engagement and struggled to work the way the brain-only group had. Those who got the tool after several sessions of thinking for themselves showed higher recall and more brain engagement than the original ChatGPT group ever had.

❝

“Same tool, same task, opposite results.
The only difference was the order.”

The people who thought first and brought AI in afterwards used it to build on something already theirs. The people who started with AI never built anything of their own for it to sit on.

A separate study from Microsoft Research and Carnegie Mellon points the same way. They surveyed 319 knowledge workers about 936 real tasks they’d used AI for, and found this:

Trust the AI more → you think less.
Trust yourself more → you check, question and correct more.

It’s a survey of what people said about themselves rather than a measurement, but it fits the MIT result uncomfortably well. The tool rewards people who show up with a view of their own, and slowly hollows out people who show up empty.

How to use it without losing your edge

I use AI every day, and I’m not going to tell anyone to stop. People who refuse it completely will lose ground to people who use it well. The point is to stay on the right side of that order.


01

Before you open the chat

Spend five minutes with the problem on your own first, on paper or in a blank note, and write your rough answer, outline or first guess, however bad. This is the brain-first condition from the MIT study, and it’s the one change most likely to keep the tool working for you. If you can’t produce anything at all, that tells you something too: this is a skill you’re currently renting, not owning.


02

While it’s open

Ask it to critique your draft, find the holes in your reasoning, or show you what you missed, instead of asking it to produce the thing from scratch. Treat every answer as a first draft from a confident junior colleague: check the facts that matter and rewrite anything that doesn’t sound like you.


03

After you close it

Run the MIT test on yourself. Without looking, explain the key points of what you just produced. If you can’t, you have work you don’t understand, and you’ll feel that gap later, in the meeting, the interview, or the moment a client asks something the document didn’t cover.

One rule on top: keep one area of your work completely AI-free, ideally the skill you most want to be known for, so there’s at least one place where the muscle is still trained at full weight.

The people writing those posts are right to be uneasy, but the research suggests they’ve misread the problem. The danger isn’t that they use AI. It’s that they’ve started reaching for it before they’ve had a single thought of their own, and that’s a habit you can reverse this week by changing nothing except the order.

Until next time
Lorenc - Founder of Success Skill

Reply

or to participate.