AI Cognitive Decline: How Constant AI Use Affects Young Minds.

Understanding artificial intelligence dependency, memory loss and critical thinking erosion

By Dr. Marcus Valladares Guimarães, Orthopedic Surgeon and Traumatologist · Published September 12, 2026 

Tom is 20 and studies engineering. He hands in a well-written essay on time. His professor asks him a simple question in class: “Can you quote one sentence from your essay?” Tom cannot. He wrote the essay with ChatGPT two days earlier, and almost nothing stayed in his memory.

Tom is not lazy, and he is not unusual. In a 2025 experiment at the MIT Media Lab, most students who wrote with ChatGPT had the same problem. In the first session, 83% could not quote a single sentence from essays they had just finished. Students who wrote without any tools had no such problem.

This is the core worry behind the phrase AI cognitive decline. When a machine does the thinking, does the brain slowly lose the habit? The honest answer is more interesting than the headlines. Some findings are solid. Others are early, small, or exaggerated online.

As a physician, I know one rule from rehabilitation very well. A muscle that stops working gets weaker. The brain is not a muscle, but it also adapts to what we ask of it. Here is what the research actually shows. I will also explain how to use these tools without paying a hidden price.

What cognitive offloading is, and why it is not new

Cognitive offloading means handing a mental task to something outside your head. Writing a shopping list is offloading. So is saving phone numbers in your phone or following GPS directions. Humans have always done it, and it often helps. It frees memory and attention for other things.

The problem starts when offloading replaces a skill we still need. Research showed this long before ChatGPT:

• The Google effect. In 2011, Betsy Sparrow and colleagues reported in Science that people remember less information when they expect a computer to store it. They remember where to find it instead.
• GPS and spatial memory. In 2020, Louisa Dahmani and Véronique Bohbot found in Scientific Reports that heavier lifetime GPS use went with worse spatial memory during self-guided navigation.
• Photos and memory. In 2014, Linda Henkel showed in Psychological Science that museum visitors who photographed objects remembered them less well than those who only looked.
• Brain drain. In 2017, Adrian Ward and colleagues found that the mere presence of your own smartphone on the desk reduced available working memory. The phone did not even need to be on.

Generative AI takes this one step further. A search engine gives you sources to read and judge. A chatbot gives you a finished answer. The judging, the structuring, and often the writing are done for you. That is where the new concern comes from.

It is fair to note the benefits too. AI can help a student with dyslexia organize ideas, or help a non-native speaker write clear English. It can explain a hard concept in five different ways until one makes sense. The question is not whether AI helps. It is what happens to the skills we stop practicing.

For the smartphone side of this story in adolescents, see our article on teen smartphone use and brain health.

The MIT study: what it showed and what it did not

The most quoted study on this topic came from the MIT Media Lab in June 2025 (Kosmyna et al., arXiv 2025). Nataliya Kosmyna, Pattie Maes and colleagues led it. They recruited 54 young adults from Boston-area universities.

Participants wrote essays in three groups. One group used ChatGPT. One used a search engine. One used only their own brain. Each person completed three sessions over several months. During writing, they wore an electroencephalography (EEG) cap, which records electrical activity from the scalp.

The main results:

1. Brain connectivity. The brain-only group showed the strongest and most widespread networks. Search engine users were in the middle. ChatGPT users showed the weakest connectivity.
2. Memory. In the first session, 83% of ChatGPT users could not quote from the essay they had just written. Almost everyone in the other groups could.
3. Ownership. ChatGPT users felt the least ownership of their essays.
4. Sameness. Essays in the ChatGPT group were very similar to each other within each topic. English teachers who graded them, without knowing the groups, described many as lacking a personal voice.

In a fourth session, 18 participants switched conditions. Former ChatGPT users now wrote without it. Their brain connectivity stayed lower than that of people who had never used the tool. The authors called this cognitive debt: the cost you pay later for effort you skipped earlier.

Read this study with care. It is a preprint, not yet a peer-reviewed paper. It involved 54 people, and only 18 completed the fourth session. EEG measures electrical activity at the scalp, not brain structure. Nothing in this study shows brain “atrophy” or damage. In 2026, a published comment by Stanković and colleagues also raised concerns about reporting and reproducibility. The results are a warning signal, not a verdict.

My reading as a doctor is simple. The memory result is the most convincing, because it is easy to understand. When you do not produce an idea yourself, you do not remember it well. That applies to any shortcut, not only to AI.

Why young brains may be more exposed

Age seems to matter. In January 2025, Michael Gerlich of the Swiss Business School published a study in Societies (Gerlich, Societies 2025). He surveyed 666 people of different ages and interviewed 50 of them.

People who used AI tools more often scored lower on critical thinking. The link ran through cognitive offloading, meaning the habit of delegating mental work. Younger participants relied most on AI and had the lowest critical thinking scores. Higher education went with better critical thinking, whatever a person’s AI use.

This is a correlation. It cannot prove that AI caused lower scores. People with weaker critical thinking may simply rely on AI more.

A stronger design came from a field experiment. Hamsa Bastani and colleagues at the University of Pennsylvania ran it (Wharton report on Bastani et al.). They worked with nearly 1,000 high school math students in Turkey. The results appeared in PNAS in 2025:

• Students with a standard ChatGPT-like tool solved 48% more practice problems correctly.
• When the tool was taken away, the same students scored 17% lower on the exam than students who never had it.
• A second group used an AI tutor that gave hints instead of answers. They improved by 127% during practice and did as well as the control group on the exam.

This is the most useful finding in the whole field. The tool itself was not the problem. The way it was used was. Answers handed over directly harmed learning. Hints that kept students thinking did not.

Many parents will recognize this pattern at home. A 15-year-old asks a chatbot to solve the algebra homework. Every answer is correct, and the homework looks perfect. Two weeks later, the same student fails the class test. The homework measured the tool. The test measured the student.

Why would young people be more vulnerable? The prefrontal cortex, the brain region behind planning, self-control, and reasoning, keeps maturing into the mid-twenties. During these years, the brain strengthens circuits that get used and trims those that do not. Skipping hard thinking during this window may matter more than skipping it at 50, when those skills are already built. This is a reasonable hypothesis based on brain development. It has not yet been tested directly with AI.

Cognitive debt and critical thinking at work

The concern is not limited to students. In 2025, Hao-Ping Lee and colleagues at Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers. They presented their results at the CHI conference, a major meeting on human-computer interaction. The workers described 936 real examples of using generative AI at work.

The pattern was clear. The more people trusted the AI, the less critical thinking they reported. The more confident they were in their own skills, the more they checked and questioned the output. Their thinking effort did not disappear. It shifted from creating content to verifying it and fitting it into their work.

Picture two office workers who ask AI for the same report. One reads it line by line, fixes two wrong figures, and rewrites the conclusion. The other sends it straight to the manager. Both used AI. Only the first one did any real thinking, and only the first one caught the errors.

This points to a skill called metacognition, which means thinking about your own thinking. It includes knowing what you understand and what you do not. You build it by struggling with problems and noticing where you go wrong. If AI removes the struggle, people may feel they understand a topic. Often they have only read a good summary of it. The fluent answer creates an illusion of knowledge.

I see a parallel in medicine. Surgeons in training learn by doing difficult steps under supervision, again and again. A resident who only watches never develops the same judgment. Nobody would accept a surgeon who learned only from videos. We should apply the same logic to how young people learn to write, argue, and solve problems.

What about “cognitive debt”? The term is useful as a metaphor. It says that effort skipped today has a cost later. The online claim that this debt becomes permanent brain damage has no support in current research. The MIT data suggest some lasting effect after a few months. Nobody has yet measured what happens after years. We also do not know if the effect reverses once people think on their own again. Based on what we know about the brain’s plasticity, some recovery is likely.

There is also an important counterweight. In 2025, Jared Benge and Michael Scullin published a meta-analysis in Nature Human Behaviour. It pooled 57 studies with 411,430 older adults. People who used digital technology more had a lower risk of cognitive impairment, not a higher one. For older adults, computers and smartphones seem to act more like mental exercise than like a crutch. Age, and the way a tool is used, change the result.

How to use AI without losing your edge

Avoiding AI completely is neither realistic nor necessary. The goal is to use it in a way that keeps your brain working. These strategies follow the evidence:

• Try first, then ask. Spend 10 to 15 minutes on a problem before opening the chatbot. This builds on the “testing effect.” In 2006, Henry Roediger and Jeffrey Karpicke showed in Psychological Science that trying to recall information strengthens memory. The effort helps even when you fail.
• Ask for hints, not answers. Tell the AI to guide you step by step, as in the GPT Tutor group. This one change protected learning in the Pennsylvania study.
• Write your first draft yourself. Then use AI to critique it. You keep ownership, and you still get feedback.
• Check what it says. Compare important claims with reliable sources. This trains the judgment the Microsoft survey found at risk.
• Keep AI-free time. Block periods for reading, writing, and calculating on your own, especially for students.
• Explain it back. After using AI, close it and explain the answer in your own words. If you cannot, you did not learn it.

How do you know if you rely on AI too much? A few warning signs deserve attention:

• You cannot start a task without opening a chatbot first.
• You usually accept the first answer without checking it.
• You cannot explain what you “wrote” without looking at it.
• You feel uneasy when the tool is not available.

If several of these sound familiar, the strategies above are a good place to start.

Parents and teachers can help too. Ask children to explain how they reached an answer, not just what the answer is. Reward the process, not only the final product.

Outside the screen, the classic brain protectors still work. Physical activity supports memory and executive function, and dance adds coordination and learning on top. Our article on how dance transforms brain health explains why. Meditation trains attention, a skill every screen competes for, as described in our guide to meditation and stress. Conversation with real people forces fast, flexible thinking that no chatbot can replace.

Even games count. Strategy games demand planning and memory, as our article on board games and cognitive health shows.

What this means for Tom, and for all of us

Tom does not need to stop using ChatGPT. He needs to change the order of things. Think first, draft first, and then ask the machine for help.

The research so far points in one direction. When AI replaces thinking, memory and ownership suffer, and learning can go backwards. The MIT experiment showed it in a small group, and the Pennsylvania experiment showed it in nearly 1,000 students. When AI guides thinking, with hints and feedback, the harm largely disappears.

The strongest claims circulating online go beyond the data. No study has shown that AI shrinks the brain or causes permanent damage. The studies are young, some are small, and one of the most famous has not yet passed peer review. Older adults who use technology even appear to be protected.

Young people remain the group to watch. Their brains are still building the circuits for reasoning, and habits formed now may last. For the daily habits that protect the brain, read our article on why lifestyle matters for brain health.

The next time Tom opens a chatbot, one question can protect his brain: “Did I try this myself first?”

References

1. Kosmyna N, Hauptmann E, Yuan YT, Situ J, Liao XH, Beresnitzky AV, et al. Your brain on ChatGPT: accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv. 2025. doi:10.48550/arXiv.2506.08872
2. Gerlich M. AI tools in society: impacts on cognitive offloading and the future of critical thinking. Societies. 2025;15(1):6.
3. Bastani H, Bastani O, Sungu A, Ge H, Kabakcı Ö, Mariman R. Generative AI without guardrails can harm learning: evidence from high school mathematics. Proc Natl Acad Sci U S A. 2025;122(26):e2422633122.
4. Lee HP, Sarkar A, Tankelevitch L, et al. The impact of generative AI on critical thinking: self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In: Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. ACM; 2025.
5. Ward AF, Duke K, Gneezy A, Bos MW. Brain drain: the mere presence of one’s own smartphone reduces available cognitive capacity. J Assoc Consum Res. 2017;2(2):140-54.
6. Sparrow B, Liu J, Wegner DM. Google effects on memory: cognitive consequences of having information at our fingertips. Science. 2011;333(6043):776-8.
7. Dahmani L, Bohbot VD. Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Sci Rep. 2020;10(1):6310.
8. Henkel LA. Point-and-shoot memories: the influence of taking photos on memory for a museum tour. Psychol Sci. 2014;25(2):396-402.
9. Benge JF, Scullin MK. A meta-analysis of technology use and cognitive aging. Nat Hum Behav. 2025.
10. Roediger HL, Karpicke JD. Test-enhanced learning: taking memory tests improves long-term retention. Psychol Sci. 2006;17(3):249-55.
11. Dergaa I, et al. From tools to threats: a reflection on the impact of artificial-intelligence chatbots on cognitive health. Front Psychol. 2024. PMC11020077.
12. Stanković et al. Comment on: Your brain on ChatGPT: accumulation of cognitive debt when using an AI assistant for essay writing tasks. arXiv. 2026. doi:10.48550/arXiv.2601.00856