I’ve been circling an uncomfortable question this month: what happens to our thinking when the tools built to help us think get better than we are at everything except the one thing that matters most?
MIT researchers gave me part of the answer. In a study led by Nataliya Kosmyna, they wired up participants and tracked their brains while they wrote essays using ChatGPT, a search engine, or nothing at all. The ChatGPT group showed the weakest neural engagement of the three, and when those same people were asked to write again without help, the deficit came with them.
Michael Gerlich found something similar outside the lab. In a 2025 study surveying more than 600 people, he found that heavier AI use tracked with lower critical thinking scores, and the relationship ran through what researchers call cognitive offloading: our tendency to hand mental effort to whatever tool is closest.
Then there’s Cal Newport, whose recent writing sent me down this path in the first place. He points to reporting by Rose Horowitch in The Atlantic showing a steady collapse in reading for pleasure, and a growing share of adults who can’t paraphrase a few pages of text without losing the thread.
Three different researchers, three different methods, one conclusion. The tools we use to think faster are quietly making us worse at thinking.
I didn’t need a study to tell me this, though. I have Alex Meyer.
Alex teaches AI for a living. He’s spent more hours than almost anyone I know watching people, mostly young people, learn to work alongside these tools. And what he tells me isn’t about the technology. It’s about his students.
He’s amazed, and not in a good way, at how much his students struggle to simply think. Not to use AI. To think. Form a position. Sit with an unresolved question long enough for something original to surface. And here’s the part that stays with me: that struggle is exactly what’s crippling their ability to engage with AI intelligently. You can’t ask a sharp question of a machine if you’ve lost the muscle for asking sharp questions at all. The tool doesn’t compensate for the deficit. It exposes it.
I see a version of this every week in my own work, sitting across from leaders who are sharp, accomplished, and used to being the smartest person in the room.
I’ll ask a hard question. The kind with no clean answer, the kind that requires someone to actually wrestle with tradeoffs and own a point of view. And more often than I’d like to admit, the first move isn’t thinking. It’s reaching. “Let’s see what AI says.”
I want to be careful here, because I’m not against the tool. I use it constantly, and used well, it’s an extraordinary thinking partner. The problem isn’t the reaching. It’s the timing. When a leader consults AI after forming their own view, they’re stress testing their thinking. When they consult it instead of forming a view, they’re outsourcing the very judgment that got them the seat at the table in the first place. One sharpens the blade. The other lets it dull.
And that’s what actually worries me, more than the statistics, more than the EEG studies, more than any of it.
Because here’s the thing about being human that we don’t say out loud enough. Almost everything about us is conditioned. Our habits, our biases, our reactions, shaped by biology, upbringing, culture, algorithms, whatever came before us. But thought, real thought, the kind that forms a new idea instead of retrieving an old one, is one of the last places we get to be genuinely free. Nobody hands you that thought. You make it.
If we hand that away too, casually, voluntarily, one “let’s see what the AI says” at a time, what’s actually left?
What do we have left to engage with, to ponder, to wrestle with, that makes any of this worthwhile?
