I avoided AI because I feared it would make me dumb

I avoided ChatGPT for three months because I was afraid it would hollow me out

I opened ChatGPT for the first time in February 2024 at 11 p.m., staring at a blinking cursor. Three keywords, a question mark, enter. The answer came back fluent, generic, useless. I closed the tab.

For about a week I told anyone who would listen that AI was overhyped. Then I stopped talking about it and just avoided it. The tab stayed closed for three months.

The reason wasn't that the answer was bland. The reason was that opening it felt like stepping onto a moving sidewalk at the airport. Once you're on, your legs stop working. You arrive at the gate, but you didn't walk there. I didn't want to be the person who arrived at the end of a sentence without remembering how I got there.

That fear has a name now. It's called cognitive offloading.

Then I saw the MIT study that proved my fear wasn't paranoia

Half wrote essays using only their brains. The other half wrote with ChatGPT from the start.

By the end of the study, many ChatGPT users were copy-pasting entire paragraphs. When the researchers asked them to reproduce their earlier essays without tools, there was little evidence they remembered much of their work. They couldn't quote from their own essays.

Research from WIRED documented how using AI for just 10 minutes might make you lazy and dumb, with studies showing that ChatGPT users consistently underperformed at neural, linguistic, and behavioral levels compared to the brain-only group.

My cursor hovering over the ChatGPT tab in February wasn't irrational caution. It was accurate instinct. But what happens when you do cross that line?

Using AI the wrong way is like hiring a VP to do your CEO job while you watch Netflix

Your brain is the CEO. AI is a vice president you just hired.

The wrong way: you hand the VP the entire project, disappear to watch Netflix, come back when it's done, and ship whatever landed on your desk. The company runs, but you're not running it. You've abdicated. Over time, you forget what decisions feel like. You get weaker.

The right way: you draft the strategy yourself, hand the grunt work to the VP, review every line, rewrite half of it, and make the final call on what ships. The VP does the heavy lifting. You do the thinking.

Microsoft researchers found that high confidence in generative AI results in less critical thinking and less overall effort in workers. When you trust the tool completely, you stop checking its work. When you stop checking, you stop learning. The VP starts making decisions you would have caught six months ago, and you don't notice because you're not in the room anymore.

That's not efficiency. That abdication shows up first in the brain itself.

The evidence shows your brain literally dims when you offload your thinking to AI

The MIT study is not an outlier.

I read about these findings and thought about the three months I avoided ChatGPT. I was protecting the muscle I didn't want to lose. Turns out there was a group in that same study who figured out how to keep the muscle while still using the tool.

But the same study found one group kept their brain sharp while still using AI

The MIT study had a twist.

After the brain-only group finished their essays manually, the researchers gave them access to ChatGPT to revise, expand, and polish. This group had already done the thinking. AI was allowed in after the draft, not before.

Their language stayed unique. The essays didn't look soulless or identical the way the ChatGPT-first essays did.

Same tool. Different order. Opposite outcome.

An assistant professor at MIT, summarized the finding: "The takeaway is not that we should ban AI in education or workplaces. AI can clearly help people perform better in the moment, and that can be valuable. But we should be more careful about what kind of help AI provides, and when."

The "when" is the firewall. Here's what that firewall looks like in practice.

Here's what it looks like when you draft your thinking before touching AI

Before (the wrong order): You open ChatGPT with a blank slate. You type "Write a stakeholder email explaining why the timeline shifted." ChatGPT writes it. You read it once, maybe twice. It sounds fine. You paste it into Outlook and hit send.

After (the right order): Before you open ChatGPT, you write three bullet points on paper or in a blank doc: - The timeline shifted because the vendor missed two milestones, not because we were slow. - The stakeholder needs to hear that we flagged the risk two weeks ago and they didn't escalate it. - The tone has to stay neutral, but I can't let this land on my team.

Now you open ChatGPT. You paste those bullets and add: "Draft an email to a stakeholder that includes these three points. Keep the tone professional and forward-looking, not defensive."

ChatGPT drafts it. You read and catch two things: the second paragraph sounds passive-aggressive, and the closing line is vague. You rewrite both. You add one sentence ChatGPT missed entirely because it didn't know the stakeholder's history with this vendor. Then you send it.

That's the difference. You stayed in the room.

Teachers described essays written ChatGPT-first as lacking originality and depth, while the essays written brain-first and then polished with AI retained the writer's thinking and voice throughout.

You can tell which workflow someone used by whether you can hear them in the final draft. But knowing the right workflow doesn't mean you'll stick to it.

In month one you'll probably still offload your thinking (I did)

The slippery slope is real, and it starts with "just this once."

You're behind on a deadline. You tell yourself you'll draft it yourself next time. Next time arrives, and you're behind again. A month in, you realize you haven't written a first draft from scratch in three weeks. You've forgotten what it feels like to stare at a blank page and figure out what you actually think.

The tell is simple: if someone asked you to reproduce the work you did last week without the tool, could you?

If the answer is no, you've crossed the line.

I came back to ChatGPT in August 2024, six months after I left. I gave it role, background, both sides of the coin. But I still slipped. There were days I opened it first and drafted second. Those were the days I couldn't remember what I worked on by Friday.

The muscle memory of thinking is like any other muscle. If you stop using it, it weakens. If you hand the reps to someone else, they get stronger and you don't.

The diagnostic question isn't "Did I use AI?" It's "Did I think first?" And that question has a simple answer you can implement tonight.

The next time you open ChatGPT, write your answer first, even if it's messy

Before you type the prompt, write three bullet points of what you think. Not what the final version should say. What you think right now, rough and unpolished.

That's the firewall. Three bullets, 30 seconds. It keeps you in the chief editor's seat.

AI can draft the sentence. It can clean up the grammar. It can find the word you couldn't remember. But it can't think for you unless you let it. And if you let it long enough, the skill of thinking becomes something you used to do.

Draft first. Polish with AI second. Ship the version you can defend without the tool.

That's the order that keeps your brain sharp.