Here is a small experiment. Think back to the last time you were stuck on something: a word you couldn’t remember, a calculation that felt complicated, a paragraph you didn’t know how to phrase. What did you do?
If you are like most people today, you reached for an AI tool. You typed your problem, got your answer in seconds, and moved on.
Now ask yourself this: Did you learn anything? Did your brain do any real work? Or did you just get the output and carry on with your day?
That tension between convenience and cognitive growth is at the heart of one of the most important questions of our time. AI is everywhere. It writes our emails, answers our questions, helps our doctors diagnose diseases, and assists our engineers in writing code. It is faster, cheaper, and often more accurate than we are. But what is it doing to us to our minds, our skills, and our ability to think?
The science is in, at least partially. And the answer is more complicated and more unsettling than either the cheerleaders or the doomsayers would have you believe.
First, let’s acknowledge what AI does very well
Before we talk about what AI might be costing us, it is only fair to talk about what it is genuinely delivering.
A Stanford University study published in April 2026 tracked ChatGPT use among more than 200,000 U.S. households. It found that using generative AI tools helps people complete “productive” digital tasks things like job hunting, travel planning, and general research 76% to 176% more efficiently than those without AI access. That is not a trivial number. For ordinary people managing busy lives, that is hours saved every week.
In the workplace, the gains are also real, though more uneven. Economists at Goldman Sachs estimate that AI could boost annual productivity growth by between 0.3 and 3.0 percentage points a year over the next decade. In 2025, economist Erik Brynjolfsson, writing in the Financial Times, pointed to new U.S. labour data showing that real GDP grew at 3.7% in Q4 of that year even as payroll growth slowed suggesting that the economy was producing more with less human effort. He called it early evidence of an “AI productivity take-off.”
For workers with less experience, the benefit appears even more pronounced. Multiple studies reviewed by the Federal Reserve Bank of Dallas found that AI boosts productivity more for less experienced workers than for highly experienced ones essentially acting as a leveller, allowing junior staff to punch above their weight.
And in science, the early results are nothing short of remarkable. In 2024, the Nobel Prize in Chemistry was awarded in part to researchers at Google DeepMind for developing an AI model that predicts protein structures, a breakthrough that had stumped biologists for 50 years. In early 2025, Google released an AI co-scientist system explicitly designed to help researchers generate new hypotheses and accelerate biomedical discoveries.
So yes in the right hands and the right contexts, AI is making us more productive, more capable, and in some corners of human knowledge, genuinely smarter.
But there is another side to this story. And it deserves to be told plainly.
The brain activity experiment that should worry everyone
In 2025, researchers at the MIT Media Lab conducted a study that has quietly become one of the most talked-about pieces of research in education and cognitive science.
The setup was straightforward: 54 participants were divided into three groups and asked to write SAT-level essays. The first group used ChatGPT. The second used Google Search. The third used nothing at all, just their own brains. Each participant was fitted with an EEG device that monitored brain activity across 32 regions. The process was repeated over multiple rounds.
The results were striking. ChatGPT users consistently showed the lowest brain engagement of all three groups. More troublingly, their brain activity actually decreased over time with each subsequent essay as though their minds were gradually switching off the more they used the tool.
Later in the study, the ChatGPT group was asked to reproduce one of their earlier essays without any tools. They could barely do it. They had retained almost nothing. The material had passed through them like water through a sieve processed, outputted, and forgotten.
The brain-only group, by contrast, was then permitted to rewrite their essay with ChatGPT access. Their work was better, more creative, more structured but critically, it still retained their original thinking and their own voice. Their baseline cognitive engagement had built a foundation that AI could then enhance.
The study was small 54 people is not a large sample and it has not yet been through full peer review. The researchers themselves acknowledged this. But they chose to release the findings publicly because they felt the question was too urgent to wait.
“Excessive reliance on AI-driven solutions,” they concluded, may contribute to what they called cognitive atrophy, a shrinking of critical thinking abilities that happens gradually, almost invisibly, as we outsource more and more of our thinking to machines.
The confidence trap
Here is where things get psychologically interesting and a little unsettling.
A peer-reviewed study published in Computers in Human Behavior in late 2025 looked at what happens to people’s self-assessment when they use AI. The results revealed a fascinating and dangerous pattern.
Using AI does improve task performance measurably and consistently. But it also causes people to overestimate how well they performed by an even larger margin. In other words, AI makes you better at things, but makes you think you are better than you actually are.
The researchers found that higher AI literacy meaning people who understood how AI works were actually worse at judging their own performance than those with less technical knowledge. The more you know about AI, the more likely you are to over-trust it and under-question yourself.
This matters enormously. A doctor who over-relies on an AI diagnostic tool may stop exercising the clinical judgment that catches the edge cases the algorithm misses. A student who uses AI to write every essay may graduate believing they are a strong writer when they have never actually developed that skill. A manager who lets AI make every data-driven decision may slowly lose the instinct to challenge the numbers.
The Harvard Gazette, in a piece from November 2025 drawing on faculty experts across philosophy, education, and cognitive science, summarised the concern bluntly. Professor David Dockterman of Harvard’s Graduate School of Education noted that we already know from decades of research that the tools we use during cognitive tasks change the way we do that work. We know, for example, that taking notes by hand leads to greater recall than typing. We know that predictive text changes our word choices. “Given these kinds of trends,” he said, “I’d be stunned if frequent use of large language models didn’t lead to real changes in the way users approach reasoning tasks.”
The student problem
Nowhere is this debate more acute or more consequential than in education.
According to available survey data, 86% of students now report using AI to help with their schoolwork. That number is extraordinary. It means that the overwhelming majority of young people in schools and universities today are regularly outsourcing cognitive tasks to a machine at precisely the age when their brains are supposed to be building the skills they will carry for the rest of their lives.
A study referenced by the World Bank’s education blog made the distinction sharply: students who use AI to complete tasks may produce polished, correct work but they have not done excellent thinking. As cognitive psychologist Daniel Willingham has argued in his research, memory is “the residue of thought.” If you never do the thinking, you never build the memory, and you never develop the skill.
This is not a theoretical concern. It shows up in the numbers. A study cited in The Science Survey found a negative correlation between AI tool usage and critical thinking scores meaning that the more frequently someone used AI tools, the weaker their critical thinking performance tended to be. The effect was especially pronounced among younger users.
The distinction that matters, researchers argue, is between two types of AI users. There are active users who use AI to test their own ideas, get feedback on their thinking, and understand things more deeply. And there are passive users who simply hand over the task and copy the output. The active users benefit. The passive users, over time, erode.
The productivity paradox
The picture gets even more complicated when you look at professional settings because AI does not consistently make skilled workers faster.
In one of the most surprising studies of 2025, researchers at METR conducted a randomised controlled experiment with experienced open-source software developers. They expected to confirm what seemed obvious: that giving developers access to advanced AI coding tools would make them faster.
Instead, they found the opposite. When developers used AI tools, they took 19% longer to complete tasks than those working without AI. The researchers explored 20 possible explanations for this, and found evidence that experienced developers were, in some cases, spending more time checking and correcting AI-generated code than they would have spent simply writing it themselves.
This does not mean AI is useless in software development other research shows clear benefits in many contexts. But it does suggest something important: AI is not a simple accelerator that makes everyone better at everything. Its benefits depend heavily on how it is used, what the task is, and how much expertise the user already has.
So which is smarter or lazier?
After all of this, the most honest answer is: both, depending on who you are and how you use it.
The research consistently points to a fork in the road. People who bring their own knowledge, judgment, and critical thinking to the interaction with AI tend to get better outcomes, sharper thinking, stronger work, and real productivity gains. People who bypass the thinking altogether tend to get the output without the growth.
This is not a new pattern. We have been here before, with every transformative technology.
When calculators became widespread in the 1970s, educators panicked that students would forget how to do arithmetic. To some extent, they were right and to some extent, it did not matter, because being able to do long division by hand became less important than understanding what division means and when to use it. When GPS became universal, some people’s spatial navigation skills declined but getting lost on the way to a new city stopped being a meaningful life challenge.
The question with AI is whether the skills it is replacing are actually less important than the ones it might free us up to develop or whether they are foundational in ways we will only miss when they are gone.
There is reason to worry that it is the latter. Critical thinking, sustained attention, the ability to form and defend an argument, the patience to sit with uncertainty and work through a problem are not peripheral skills. They are the core of what it means to be an educated, capable person. And they are built through struggle, not through shortcuts.
What you can actually do about it
This is not a call to stop using AI. That ship has sailed, and in many ways it should have. AI is a genuinely powerful tool, and refusing to use it entirely is as misguided as refusing to use a library because you might become dependent on it.
But it is worth being intentional about how you use it.
Try this: before you ask AI to answer a question, try to answer it yourself first. Write down your own thinking, even roughly, before you read what the AI says. When AI gives you a draft, don’t just accept it engage with it, argue with it, rewrite parts of it in your own voice. Use it as a sparring partner, not a ghostwriter.
In education especially, the institutions getting this right are those that use AI to give students better feedback on their thinking not to eliminate the thinking altogether. The ones that design AI use so that students still struggle, still make mistakes, still have to retrieve information from memory, and still have to construct arguments from scratch.
The brain, like a muscle, only grows when it works. AI is a powerful assistant. But if the assistant does all the lifting, the muscle will eventually weaken and you may not notice until you really need it.
The question we should all be sitting with
There is a version of the AI future that looks genuinely exciting: one where humans are freed from drudgery, where scientists make discoveries faster, where the gap between experts and beginners narrows, and where people have more time to focus on what they find meaningful.
And there is another version that is quieter and harder to see coming: one where we gradually lose the capacity to think hard about hard things, where our confidence outpaces our competence, and where we become, in the words of one cognitive scientist, very good at looking smart without becoming smarter.
Which version we get depends less on the technology than on us on the choices we make individually and collectively about how to use a tool that is, right now, reshaping the human mind in ways we are only beginning to understand.
So the next time you reach for AI, ask yourself: am I using this to think better, or am I using this to avoid thinking?
That question, unlike most things these days, is one only you can answer.
Subscribe Deshwale on YouTube
