When artificial intelligence does the thinking, what happens to the thinker? A Patna neuropsychiatrist explains the MIT ChatGPT EEG study, newer research on cognitive offloading, persistence and deskilling, who is most at risk, and a simple Brain-First Ladder for using AI without losing your own mind.
In this article 16
- Your Brain on ChatGPT: what the MIT EEG study found
- Not just one study: AI, cognitive offloading and critical thinking
- 1. Ten minutes of AI help can reduce persistence (randomised trials, 2026)
- 2. AI tutors can harm learning, unless they are designed to teach (PNAS, 2025)
- 3. The more we trust AI, the less we think (Microsoft Research and Carnegie Mellon, 2025)
- 4. Younger heavy users show weaker critical thinking (Societies, 2025)
- 5. Even experienced doctors can be "deskilled" (Lancet Gastroenterology & Hepatology, 2025)
- Evidence at a glance
- The neuroscience of cognitive debt: why effort builds the brain
- AI dependence: who is most at risk?
- AI use among students in India: the picture at home
- How to use AI without losing your thinking skills: the Brain-First Ladder
- The bottom line
- Frequently Asked Questions
- References
- About this article
In one line
A Class 12 student finishes a two-page essay in four minutes. A young professional drafts every email, report and presentation through a chatbot. A student asks the chatbot to solve each physics numerical before even reading it. Life feels faster and easier. But a growing body of research is asking an uncomfortable question: when artificial intelligence does the thinking, what happens to the thinker?
This question went viral again recently when a widely shared post by Massimo (@Rainmaker1973) on X described an MIT Media Lab study in which students wore EEG caps while writing essays with ChatGPT, with a search engine, or with no tools at all. As a neuropsychiatrist who spends every working day thinking about how brains learn, adapt and sometimes struggle, I want to walk you through what this study actually found, what similar research around the world is showing, what it does not show, and, most importantly, how you and your children can use AI without paying for it with your own mind.
Your Brain on ChatGPT: what the MIT EEG study found
The study, titled "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task", was led by Dr Nataliya Kosmyna at the MIT Media Lab.1 Fifty-four students aged 18 to 39 from universities in the Boston area were divided into three groups. One group wrote SAT-style essays using ChatGPT, one used a search engine, and one used only their own brains. Each person wrote three essays over several months while a 32-channel EEG recorded how different brain regions were communicating with each other. In a fourth session, 18 participants switched conditions: the ChatGPT users wrote without it, and the brain-only writers were given ChatGPT.
The results were striking. The brain-only group showed the strongest and most widely distributed brain connectivity, especially in networks linked to memory, attention and planning. The search-engine group was in the middle. The ChatGPT group showed the weakest connectivity. By the third essay, many ChatGPT users were simply pasting the prompt and lightly editing the result. Most strikingly, many of them could not accurately quote a sentence from the essay they had just "written", and they described feeling little ownership of it: "not really my essay".
When the ChatGPT users were later asked to write without the tool, their brains remained under-engaged and their recall stayed weak. The researchers called this pattern "cognitive debt": the short-term ease of outsourcing thought is borrowed against your future ability to think, remember and create on your own.
The effort you skip today is the learning you lose tomorrow.
There was also a hopeful twist that received far less attention. When people who had first written on their own were later given ChatGPT, their brain engagement did not collapse; it rose, and they used the tool more thoughtfully. The order mattered: thinking first and using AI second produced a very different brain pattern from leaning on AI from the start.
A fair word of caution
This study is small, focused only on essay writing, and was released as a preprint before completing peer review. The lead author herself has said it shows no "brain rot", no stupidity and no damage: EEG measures how engaged the brain is during a task, not the size or health of brain tissue. Other scientists have published constructive criticisms of its methods.2 It is an early warning signal, not a final verdict, which is exactly why it matters that other research points the same way.
Not just one study: AI, cognitive offloading and critical thinking
Psychologists use the term cognitive offloading for handing mental work to an external aid: a calculator, a diary, a phone contact list. Offloading is not new, and it is not always bad; nobody misses memorising 50 phone numbers. In 2011, a well-known study in Science showed the "Google effect": when people expect information to remain available online, they remember where to find it rather than the information itself.3 Generative AI goes much further. It does not just store facts; it can draft, argue, summarise, calculate and decide. That means it can offload not only memory but thinking itself. Here is what the newer research shows.
1. Ten minutes of AI help can reduce persistence (randomised trials, 2026)
In a set of three randomised experiments involving 1,222 adults, researchers from Carnegie Mellon, Oxford, MIT and UCLA gave some participants an AI assistant while they solved fraction problems or SAT-style reading questions, then removed it without warning.4 With AI, people performed better. Once it was taken away, their accuracy dropped and they were more likely to skip or give up, after only about ten minutes of use. Crucially, the harm was concentrated in those who asked AI for direct answers; people who used it only for hints performed much like those who never had it. The authors suggest that AI conditions us to expect instant answers, and so quietly erodes our tolerance for the healthy struggle that learning requires.
2. AI tutors can harm learning, unless they are designed to teach (PNAS, 2025)
In a field experiment with nearly a thousand high-school maths students in Turkey, students using a standard ChatGPT-style tutor scored 48% higher in practice sessions. But when the AI was removed for the exam, they scored 17% lower than students who had never used it. A second group used a "GPT Tutor" built to give hints rather than answers; they improved even more in practice and showed no loss on the exam.5 Revealingly, the AI users believed they had learned more than they actually had: a dangerous illusion of competence.
| AI tutor group | Practice sessions (with AI) | Exam (without AI) |
|---|---|---|
| GPT Base (gives answers) | 48% higher | 17% lower |
| GPT Tutor (hints only) | 127% higher | No significant loss |
3. The more we trust AI, the less we think (Microsoft Research and Carnegie Mellon, 2025)
A survey of 319 knowledge workers, covering 936 real work tasks, found that people who trusted AI more applied less critical thinking to its output, while those confident in their own abilities thought more carefully. Work shifted from solving problems to merely checking and tidying the machine's answers, and those relying on AI produced less varied ideas for the same task.6 Many workers admitted they could no longer judge AI output in areas where they lacked expertise.
4. Younger heavy users show weaker critical thinking (Societies, 2025)
A UK study of 666 people found a clear negative link between frequent AI use and critical-thinking scores, explained largely by cognitive offloading. The pattern was strongest among 17- to 25-year-olds, while higher education appeared protective.7 This study shows an association rather than proof of cause, but it fits the experimental evidence above.
5. Even experienced doctors can be "deskilled" (Lancet Gastroenterology & Hepatology, 2025)
Perhaps the most sobering evidence comes from medicine. In four Polish endoscopy centres, experienced specialists began using an AI system that highlights polyps during colonoscopy. When they later performed colonoscopies without AI, their detection rate of pre-cancerous growths had fallen from 28.4% to 22.4%, comparing the three months before AI was introduced with the three months after.8 This was described as the first real-world clinical evidence of AI-related deskilling, in experts with thousands of procedures behind them.
Evidence at a glance
| Study | Who was studied | Key finding |
|---|---|---|
| MIT Media Lab EEG study (2025, preprint) | 54 students writing essays over 4 months | ChatGPT users showed the weakest brain connectivity, poorest recall of their own essays and low ownership |
| Liu et al. (2026, preprint) | 1,222 adults, 3 randomised experiments | About 10 minutes of AI help, then removal: accuracy fell and people gave up sooner, mainly those who took direct answers |
| Bastani et al., PNAS (2025) | About 1,000 high-school maths students | Unrestricted AI: 17% lower on exams without AI; guarded "hints" tutor: no loss |
| Microsoft Research and CMU, CHI (2025) | 319 knowledge workers, 936 tasks | More trust in AI, less critical thinking; more self-confidence, more critical thinking |
| Gerlich, Societies (2025) | 666 UK adults | Heavier AI use linked with weaker critical thinking, via cognitive offloading; strongest in ages 17 to 25 |
| Budzyń et al., Lancet Gastroenterol Hepatol (2025) | 19 experienced endoscopists, 1,443 procedures | Unaided polyp detection dropped from 28.4% to 22.4% after AI exposure |
The neuroscience of cognitive debt: why effort builds the brain
Why should skipping mental effort matter so much? Because, for the brain, effort is not a cost of learning; it is the learning. Here is how I explain it to students and families in my clinic.
Use it or lose it
Neural pathways strengthen with repeated use and weaken with disuse. The prefrontal cortex, our centre for planning, reasoning and self-control, is trained by struggling with problems, much like a muscle in a gym. If a machine lifts the weights, the muscle does not grow.
The generation effect
We remember what we produce far better than what we merely read. Writing a sentence yourself forces the brain to retrieve, connect and encode ideas. Copying a polished paragraph leaves a faint trace, which is why the MIT students could not recall their own essays.
Desirable difficulty
Learning that feels slightly hard (recalling, trying, making mistakes and correcting them) produces deeper and longer-lasting memory than learning that feels smooth. AI removes exactly this productive friction.
The instant-reward trap
The brain's reward system loves quick results. When answers arrive in seconds, our tolerance for frustration shrinks: the reduced persistence seen in the 2026 trials. Persistence, more than talent, predicts long-term achievement.
There is also a quieter psychological cost: the illusion of competence. Because AI-assisted work looks impressive, we believe we understand it. The PNAS study showed exactly this: students overestimated their learning.5 In real life, that illusion breaks in the examination hall, the interview room or the operation theatre, where the tool is not available. (Our article on the flow trap explores the same gap between feeling right and being right.)
AI dependence: who is most at risk?
Children and teenagers. The prefrontal cortex continues maturing into the mid-twenties. Habits formed during these years shape how the brain wires itself for attention, memory and self-control. This is why the MIT lead author said developing brains are at the highest risk, and why I advise parents to treat chatbots like any powerful tool: useful, but supervised.
Students in high-pressure exam systems. In India, where board exams, NEET, JEE and university entrance tests reward deep understanding and speed under pressure, a student who has "learned" through AI may perform well on homework and poorly on exam day, exactly the pattern in the maths study. Our student mental health and exam stress guides, including smart study, not hard study, set out study methods that build real recall.
Professionals in high-stakes work. Doctors, engineers, lawyers and pilots must remain capable when technology fails. The colonoscopy study is a reminder that even experts can lose hard-won skills within months.
People with anxiety or obsessive-compulsive tendencies. In my clinical practice I increasingly see patients who ask AI chatbots the same health or "what if" question dozens of times a day. For anxiety and OCD, this endless, instant reassurance works like any other compulsion: brief relief, followed by stronger doubt. AI can quietly become a 24-hour reassurance machine. (See our guide to generalized anxiety disorder.)
AI use among students in India: the picture at home
This is not a distant Western problem. The Bharat Survey for EdTech (BaSE) 2025 by the Central Square Foundation, covering 12,500 children and 2,500 teachers across ten states and released in February 2026, found that about a third of school students who use edtech now use generative AI for learning, and that nearly three in four children who know of these tools believe they work like a search engine, unaware that they can be confidently wrong.9 At IIT Delhi, an internal survey by the institute's AI committee found that at least 80 per cent of students and 77 per cent of faculty use AI tools; the institute now asks students and staff to disclose AI assistance in their work.10 Universities in India and abroad are rethinking take-home assignments, and some are returning to oral examinations, because written work no longer reliably shows what a student has learned. (Our research digest on problematic internet use in Indian students looks at the wider screen picture.)
What the research does NOT say
It does not say AI damages brain tissue, lowers IQ or makes people "stupid". Several key studies are small, recent or still under peer review, and some show association rather than cause. And the same research shows AI can support learning when it is used to give hints, explain mistakes or extend your own thinking. The problem is not AI. The problem is replacing thinking instead of enhancing it.
How to use AI without losing your thinking skills: the Brain-First Ladder
Every study above points to the same practical lesson: how and when you use AI matters more than whether you use it. I recommend this simple ladder to students, parents and professionals.
- 1Think first.
Spend 10 to 15 minutes alone with the problem: jot ideas, an outline, a first attempt.
- 2Ask for hints, not answers.
"Where is my mistake?" instead of "Solve this."
- 3Check and challenge.
Verify facts, ask "why?", reject what you cannot explain.
- 4Explain it back.
Close the screen and teach it aloud. If you can't, you haven't learned it.
- 5Own the final word.
Your voice, your judgement, your signature.
For parents and teachers
- Ask children to show their own attempt before any AI help.
- Set AI-free study blocks and screen-free evenings.
- Value the process: ask "how did you work this out?"
- Use oral questions and handwritten work to check real understanding.
- Talk openly about AI errors and "hallucinations".
For professionals
- Draft the first version of important work yourself.
- Regularly practise core skills without AI, like a pilot in a simulator.
- Treat AI output as a junior colleague's draft: verify before you sign.
- Keep learning in your field; you cannot judge what you do not understand.
- Notice when you reach for AI out of habit, not need.
The bottom line
Artificial intelligence is one of the most powerful tools humanity has ever built, and I use it myself. But a tool is only as good as the hand, and the mind, that holds it. The emerging science suggests that when AI replaces our thinking, we may lose memory, persistence, confidence and skill; when it extends our thinking, it can make us sharper. Think first. Struggle a little. Then let the machine help. Your brain will thank you, today and twenty years from now.
If you, your child or someone you care for feels unable to study, work or make decisions without constant AI or online reassurance, or if screen dependence is affecting sleep, mood or relationships, please reach out. Help is available, and it works. Our internet and screen addiction service is a good place to start.
More from the Awareness Series
Your questions
Frequently Asked Questions
Does using ChatGPT damage the brain?
No study has shown structural brain damage or a fall in IQ. The MIT team itself warned against words like "brain rot". What the research shows is lower mental engagement while the tool does the work, and weaker memory and skill afterwards: a functional effect that depends on how you use AI.
What is cognitive debt?
A term from the MIT Media Lab study: the short-term ease of letting AI think for you is "borrowed" against your future ability. The effort you skip is the effort that would have built memory, understanding and skill.
Is it safe for students preparing for board exams, NEET or JEE to use AI?
Yes, if it is used as a coach, not a crutch. Attempt questions on your own first, use AI to find your error or explain a concept, and then solve a similar question without it. Never submit or memorise answers you cannot explain.
How do I know if I am becoming over-dependent on AI?
Warning signs include reaching for AI before trying yourself, difficulty starting tasks without it, not remembering work you "wrote", feeling anxious when it is unavailable, and accepting its answers without checking.
Can AI ever improve learning?
Yes. In the PNAS high-school study, a tutor that gave hints rather than answers produced no learning loss, and in the MIT study people who wrote on their own first and used AI later showed higher brain engagement. Order and style of use matter.
Should children use AI chatbots?
Children and teenagers are still building the brain systems for attention, memory and self-control. Supervised, limited, "brain-first" use is sensible; unsupervised homework-by-chatbot is not.
References
Inspired by a post shared by Massimo (@Rainmaker1973) on X about the MIT Media Lab study. This article summarises published research in the author's own words; findings were checked against the original research and reputable reports.
- Kosmyna N, et al. Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT Media Lab; 2025. arXiv:2506.08872 (preprint, under peer review). arxiv.org
- Jones N. Does using ChatGPT change your brain activity? Study sparks debate. Nature news, 25 June 2025; republished by Scientific American. scientificamerican.com. Used for: expert caution on interpreting the MIT preprint.
- Sparrow B, Liu J, Wegner DM. Google effects on memory: cognitive consequences of having information at our fingertips. Science. 2011;333(6043):776-778. doi:10.1126/science.1207745
- Liu G, Christian B, Dumbalska T, Bakker MA, Dubey R. AI assistance reduces persistence and hurts independent performance. arXiv:2604.04721 (preprint), 2026. arxiv.org
- Bastani H, Bastani O, Sungu A, et al. Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS. 2025;122(26):e2422633122. doi:10.1073/pnas.2422633122
- Lee H-P, 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 CHI '25, Yokohama; 2025. doi:10.1145/3706598.3713778
- Gerlich M. AI tools in society: impacts on cognitive offloading and the future of critical thinking. Societies. 2025;15(1):6. doi:10.3390/soc15010006
- Budzyń K, Romańczyk M, Kitala D, et al. Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study. Lancet Gastroenterol Hepatol. 2025;10(10):896-903. doi:10.1016/S2468-1253(25)00133-5
- Central Square Foundation. Bharat Survey for EdTech (BaSE) 2025. New Delhi; released February 2026. centralsquarefoundation.org
- Press Trust of India. AI-generated work should be disclosed to ensure transparency: IIT Delhi to students, faculty. The Week, 28 May 2025. theweek.in. Used for: IIT Delhi's internal survey and guidelines.
About this article
Written and medically reviewed by
- MD (Psychiatry) AIIMS New Delhi
- Associate Professor & HOD, Dept. of Psychiatry, RDJM Medical College, Muzaffarpur
- Ex-Assistant Professor, PMCH Patna
- Consultant, Healthy Minds India, Patna
- Written
- November 2026
- Format
- Research explained
- Evidence
- One EEG preprint, randomised and field experiments, two surveys and one observational clinical study; several are small or still under peer review
- Sources
- 10 references
- The clinical observations on anxiety, OCD and reassurance-seeking, and the Brain-First Ladder, are the author's own professional advice; they are not conclusions of the studies cited.
- Dr. Saurabh Kumar's research focuses on brain stimulation treatments in psychiatry. He is the author of The Indian Mind.
© 2026 Healthy Minds India, Patna · Research explained, October 2026
Disclaimer
This awareness article summarises published research in the author's own words for public education. It is not a substitute for individual medical advice. If screen or AI dependence is affecting study, work, sleep, mood or relationships, please speak to a qualified mental health professional. Healthy Minds India is not an emergency service.
If you need help now
Tele-MANAS (free, 24×7): 14416 or 1-800-891-4416 · Emergency: 112
Related guides
Healthy Minds India, Patna
House No. 46/B, Sahdeo Mahto Marg, Near Sri Krishna Puri Park (Children's Park), Patna 800001
In person in Patna, online across India · Mon-Wed, Fri & Sun, 9 AM-7 PM (Thu & Sat closed)
Appointments
Non-emergency: +91 8709299986 / 9060484416 · book a consultation
Services: Student mental health and exam stress · Internet and screen addiction · Anxiety and OCD care · CBT and psychotherapy · Online consultation across India
Medically reviewed by Dr. Saurabh Kumar, Consultant Neuropsychiatrist, Healthy Minds India, Patna. This article is for general education and is not a substitute for a personal consultation. To discuss your situation, book an appointment or contact the clinic.
