What the MIT “Cognitive Debt” Study Misses About Writers Like Me

Brain scans, left-handed chaos, and the AI I can’t seem to quit

There’s a new study floating around this week: Your Brain on ChatGPT. It’s the kind of headline that makes people panic-share it everywhere: if you write with AI, your brain coasts. Executive control dips. Semantic processing quiets down. They call it “cognitive debt.”

I skimmed it, read a few (dozen) articles about it, and then — yes, I know — asked that very same tool to break it down for me so I could actually understand it.

My first thought? Eep. I do not need more cognitive debt. I’ve already sacrificed enough brain cells to late nights, good times, parenting, and the background static of half-finished ideas.

Still, for me, the machine is both the problem and the fix. The longer I stay here, the truer — and messier — that feels.

Just two days ago, I wrote The Letter: How I Asked ChatGPT to Help Me Leave ChatGPT. I meant every word. I really did try to switch to Claude for good. But Claude won’t hold my chaos. There’s no memory. No echo of what I said yesterday, no way to hold the contradiction I’ll want to argue with tomorrow.

So here I am — suspicious as ever, but still leaning on the same machine to help me untangle this anyway.

After The Letter went out, a friend messaged me:

“I totally don't need to be hating my co-brain AI, shame on you Amanda for making me have to second guess ChatGPT. >.<”

It made me laugh — because he’s not wrong. For people like him, like me — brains wired a little sideways — this isn’t autopilot. It’s a co-brain. A stabilizer for thoughts that outrun my working memory on the best day.

My OG co-brain: a messy notebook, my favorite pen, and my kids yelling “watch this!” every 30 seconds.

Which brings me back to that MIT study.

Now, I’m no brain scientist — the closest I ever got to MIT was slinging pints at an Irish pub in Kendall Square while I was in photography school in Boston. But here’s what I do know:

They tested fifty-four college students — healthy, right-handed, native English speakers. Some wrote SAT-style essays from scratch. Some used a search engine. Some let GPT-4o handle it for them. Each student wore an EEG cap, so the researchers could track the brain’s electrical chatter in real time — session by session, tool by tool. Some students switched methods halfway through, to see how the signals shifted.

They focused on two key networks:

1. Frontal–parietal connectivity: the brain’s project manager — planning, self-monitoring, keeping ideas lined up.

2. Temporal–frontal connectivity: the bridge that turns raw word knowledge into meaning — the difference between knowing a word and weaving it into a bigger thought.

The big finding? The more students relied on GPT to generate text for them, the weaker those connections got. They also remembered less of what they’d written. No friction, no anchor.

On paper? Solid experiment. And a fair caution: passive use will flatten your spark.

But here’s what I keep circling back to: whose brains did they test — and whose did they quietly filter out?

They screened for right-handed students because handedness affects where language and planning live in the brain. Right-handed brains tend to have neat left-hemisphere dominance — so the EEG signals line up nicely across people. Left-handers and mixed-handers often have more varied wiring — and they show up more often in neurodivergent groups like ADHD and autism. I notice this because I’m left-handed myself. So by controlling for handedness, they quietly filtered out people wired more like… well, me.

• In neurotypical brains, weaker frontal–parietal signals usually mean you’re phoning it in — less focus, shallower thinking. But for ADHD brains like mine, that signal runs weak by default. The real challenge is keeping it online at all. So using a tool like this doesn’t drain me — it props up what’s missing.

• Same with the temporal–frontal bridge: autistic brains often wire for detail but can struggle to stitch the big picture together. For some folks, an AI that holds context steady can help connect the dots — not flatten them.

Same brain signal ≠ same meaning for everyone. For some of us, the tool patches holes, instead of just dulling edges.

If you could wire me up mid-session, you’d see plenty sparking. I don’t show up blank, waiting for a robot to spit out something polished. I'll dump a paragraph of jumbled thoughts, a half-formed metaphor, and a nagging self-doubt into the chat, and say: Hold this while I pace. Then I break what it hands back and rebuild it line by line until it sounds like me again. That friction is the point.

Do you use AI like this, or totally differently? I’d love to read how you tangle with it — drop a comment.

If you want to stay smart, you have to keep doing the parts that feel like work. I don’t let it dream up my ideas. I don’t let it decide where I stand. I don’t let it finish my thoughts for me — at least, not without a fight. I use it to hold the scraps I’d lose otherwise. But the spark, the tension, the hard edges — that’s my brain’s job.

---

And I know. I’m an artist. None of this is what I thought I’d be writing about when I started a Substack. (Creative retreats, who?) But here we are: me, my chaos, my co-brain — still poking at this machine because I can’t seem to leave it alone.

---

TL;DR: Don’t fear the tool. Don’t worship it either. Keep your fingerprints on the mess. Keep the parts that hurt a little. That’s how you stay awake.