Is AI Making Us Dumber? The Evidence Is Real — but the Question Is Wrong

The headlines write themselves. “ChatGPT rots your brain.” “AI is making us stupid.” They follow a genuine MIT study that put electrodes on people’s heads while they wrote essays, and they are not entirely wrong. But they are asking the wrong question, and the wrong question produces a useless answer. AI is not making us dumber in the way a toxin makes us sick—uniformly, unavoidably. It is offering us a trade, the same trade every cognitive tool has ever offered, on unusually generous-looking terms. Whether we come out ahead depends entirely on what we agree to give up. Here is what the evidence actually shows and why the panic and the dismissal are both wrong.

What The Study Actually Found

The piece of research behind most of the alarm is a 2025 MIT Media Lab preprint with a memorable name: “Your Brain on ChatGPT: Accumulation of Cognitive Debt.” Researchers split participants into three groups to write essays—one using ChatGPT, one using a search engine, and one using only their own minds—and monitored their brain activity with EEG.

The findings were striking. The ChatGPT group showed the weakest neural connectivity of the three, and that engagement declined over repeated sessions. More tellingly, the AI users struggled to quote or remember their own essays afterwards and reported feeling less ownership over the writing. They had produced text without, in some meaningful sense, having thought it.

Also ReadEntertainmentAre Spotify and YouTube Premium Worth It? The Subscription Math→

That’s a real result, and it should give anyone pause. But three caveats matter enormously, and the headlines omitted all of them.

The Caveats The Headlines Skipped

It’s small and preliminary. Fifty-four participants, with only eighteen completing the crucial fourth session. It’s a preprint—not yet peer-reviewed. The lead researchers themselves urged caution. One study of this size proves a direction worth investigating, not a settled fact about the human brain.

There’s an innocent explanation for part of it. As researchers writing in The Conversation pointed out, the brain-only group’s rising engagement over the first three sessions may partly reflect the familiarisation effect—people get more efficient at a repeated task. When the AI group was finally asked to write unaided, they were doing it for the first time, so of course they looked less engaged. Some of the “decline” is an artifact of study design, not proof of atrophy.

Also ReadTechnologyAI and the Economy: A Complete Guide to the Boom, the Bubble, and the Fallout→

Lower brain activity isn’t automatically bad. A calculator lowers the neural effort of arithmetic too. The question is never whether a tool reduces mental effort—that’s the point of tools—but whether it reduces effort we needed to be spending.

So the honest reading of the marquee study is narrower than the headlines: writing an essay by having ChatGPT generate it engages your brain less than writing it yourself. Which, stated plainly, is not surprising. It’s barely a finding. The real story is elsewhere.

The Stronger Evidence Is About Skill, Not IQ

Set the EEG study aside, and a more convincing body of research comes into focus—and it points not at raw intelligence but at the erosion of skill through disuse.

A study of consultants found that AI boosted their short-term performance by 14 to 40 percent while simultaneously eroding the independent judgment it was meant to support. The mechanism is the interesting part: by handling the routine cases and leaving only the exceptions to humans, AI removes the everyday practice through which judgment is built and maintained. This is part of a much bigger shift in how AI is changing work itself, from individual productivity to the structure of entire industries. For a broader look at that transformation, see our guide to how AI is reshaping the US economy. So when a genuine exception arrives—when the AI is wrong or hits something it hasn’t seen—the human who’s supposed to catch it is out of practice.

The most concrete finding comes from medicine. One study found that three months after AI assistance was introduced, clinicians’ ability to detect tumors without the AI had dropped by 6 percent. Their skill hadn’t been upgraded; it had been quietly loaned to the machine. In education, a preregistered trial found an 11-percentage-point gap in retention between AI-assisted and unassisted learners forty-five days out—the AI group had processed the material less deeply and remembered less of it.

This is the real risk, and it has a name older than AI: cognitive offloading. We’ve always offloaded—to shopping lists, to calculators, to GPS. What’s new is the range. Earlier tools offloaded narrow tasks. AI can offload the synthesis and reasoning themselves, which is the part where the thinking used to happen.

The Distinction That Changes Everything

Here is where the evidence stops being scary and starts being useful, because the same research that identifies the danger also identifies the escape.

The studies converge on a distinction between two kinds of AI use:

Unstructured use — where the AI does the reasoning and synthesis for you, and you accept the output — is where the cognitive costs show up. Weaker retention, reduced critical thinking, the “metacognitive laziness” of offloading engagement itself.

Structured use — where the AI handles the mechanical parts (grammar, formatting, a first-draft scaffold, a sparring partner to argue against) while you do the reasoning — does not show the same costs, according to more recent work.

The variable isn’t AI. It’s whether you use it as a crutch or a gym. Asked to adjudicate its own effect, even ChatGPT gave the honest answer: it depends whether you engage with it “as a crutch or a tool for growth.” That distinction also matters when deciding how you use ChatGPT day to day. If you’re considering whether its paid features are actually worth using, our ChatGPT Plus review for 2026breaks down what you actually get for the money. The technology is not the verb. You are.

Why This Generation Should Pay Attention

One pattern in the research is hard to shrug off: younger users are consistently more affected than older ones. The likely reason is developmental. Someone over forty-five who uses AI to draft an email is offloading a skill they already spent decades building; the muscle exists, it’s just resting. Someone who reaches for AI before building that skill may never build it at all. You can’t atrophy a muscle you never developed.

That’s the genuinely serious version of this story, and it has nothing to do with EEG readings. A generation that offloads the struggle of early learning — the hard, boring, effortful part where competence is actually forged — risks arriving at adulthood fluent in prompting and shaky at thinking. The tool is the same; the stakes are higher the earlier you pick it up.

That concern also raises a bigger question for education: will AI replace teachers, or simply change the way they teach? As AI takes over more tasks students once learned through practice, the real challenge is making sure it supports learning without replacing the human guidance that helps students develop critical thinking.

The Verdict

So, is AI making us dumber? The framing is wrong, and the wrongness matters. AI is not lowering anyone’s intelligence. It is offering to do our thinking for us, and taking us up on that offer—reflexively, for everything, before we’ve built the skills it’s replacing—will absolutely erode the capacities we stop using. Use it to skip the effort that builds judgment, and you will lose judgment. Use it to handle drudgery while you do the reasoning, and you may well come out sharper.

The printing press, the calculator, and GPS all provoked versions of this same panic, and the panic was always half right: each did erode a specific skill (memorization, mental arithmetic, or spatial navigation) while freeing capacity for something else. AI is a bigger bet because it reaches the core function — reasoning itself — rather than a peripheral one. That’s exactly why the choice of how to use it matters more than it ever has before. The machine isn’t making us dumber. It’s asking us, constantly, whether we’d like to stop thinking. The answer is ours.

Sources

• MIT Media Lab — “Your Brain on ChatGPT: Accumulation of Cognitive Debt” (Kosmyna et al., arXiv preprint, 2025)

• The Conversation — familiarisation-effect critique of the MIT study

• Harvard Gazette, Singularity Hub — expert commentary and the cognitive-offloading framing

• International AI Safety Report 2026 — clinician tumor-detection decline (6%); 666-participant critical-thinking study

• Published consultant and education studies (retention gaps, structured vs. unstructured use)

Note: An analysis piece. Research on AI and cognition is early and evolving; findings cited are preliminary and, in the case of the MIT study, not yet peer-reviewed. This reflects New York Editor’s reading of the current evidence, not settled science.