All articles

Is AI Making Us Dumber? What the Research Says About Thinking and Creativity

Is AI hurting how we think and create? A fair look at the evidence on critical thinking, memory and creativity, plus 9 rules to use AI without losing your edge.

Is AI Making Us Dumber? What the Research Says About Thinking and Creativity
On this page
  1. The research at a glance
  2. What is cognitive offloading, and is it always bad?
  3. What the evidence says about thinking and learning
  4. What the evidence says about creativity
  5. Other ways AI can quietly weaken judgment
  6. How to read "AI is rotting our brains" headlines
  7. Who is most at risk?
  8. Delegate or develop: a simple test
  9. Nine rules to use AI without dulling your mind
  10. Advice for parents, teachers and managers
  11. Advice for writers and creators
  12. Frequently asked questions
  13. The bottom line

Headlines keep claiming that AI is "destroying" our ability to think. The research does not support that dramatic claim, but it does support a narrower and more useful warning: when people hand over the thinking itself, especially while they are still learning a skill, they tend to learn less, remember less and rely more on the tool. For creativity, AI often helps individuals produce better work, yet it can make the work of many people look more alike.

So the honest answer to "Is AI making us dumber?" is: it depends on how you use it. Using AI to skip the hard part of thinking has real costs. Using it to challenge, check and extend your thinking can help. Most studies so far are small, short and often unreviewed, so nobody can say what happens over ten years.

This guide walks through the best evidence on both sides, explains why some famous studies deserve caution, and ends with nine practical rules you can follow today. If you want the technical background on why chatbots can sound sure while being wrong, read our explainer on how large language models work first.

The research at a glance

Source What it found Main caveat
MIT Media Lab EEG study (2025 preprint) Essay writers using ChatGPT showed weaker brain connectivity and less ownership of their work than writers using search or nothing Only 54 people, one task, not peer reviewed as of fall 2026
Microsoft Research and Carnegie Mellon (2025) Among 319 knowledge workers, more confidence in AI went with less critical-thinking effort Self-reported survey, shows association
Gerlich (2025) Across 666 participants, heavier AI use was linked to lower critical-thinking scores Correlation, not proof AI caused it
High school math field experiment (PNAS, 2025) Students using a plain chatbot did better in practice but worse on unassisted tests; a tutor-style version avoided the harm One subject, one age group
Doshi and Hauser (Science Advances, 2024) AI ideas made individual stories more creative, but stories became more similar across writers Short story task in an experiment
International AI Safety Report (2026) AI may harm critical thinking and memory in some contexts, but the evidence is early Evidence base is still small

What is cognitive offloading, and is it always bad?

Cognitive offloading means using a tool to reduce the mental work a task requires. Humans have always done it. We write shopping lists, use calendars and let calculators do arithmetic. The Greek philosopher Plato even worried, in a famous passage, that writing would weaken memory.

Offloading is not automatically harmful, and it often frees your mind for better work. The question is what you are handing over. Researchers have found, for example, that people who rely heavily on GPS tend to build a weaker mental map of the places they navigate. That is a reasonable trade for a delivery driver, but a poor one if you are trying to learn a city.

The same logic applies to AI. Offloading routine tasks you already understand is efficient. Offloading the very skill you are trying to build can leave you less capable when the tool is gone.

What the evidence says about thinking and learning

The MIT "Your Brain on ChatGPT" study

This is the study behind most of the headlines. Researchers at MIT asked 54 people to write essays over several sessions, using ChatGPT, a search engine, or nothing but their own brains, while measuring brain activity with EEG. The ChatGPT group showed the weakest connectivity, the lowest sense of ownership over their writing, and in the first session most could not accurately quote what they had just written. When some participants swapped groups later, those who had started with ChatGPT stayed less engaged, while those who had first written on their own showed higher engagement when they then used AI. The paper is available on arXiv.

It is a fascinating result, but treat it carefully. It is a small preprint that had not been peer reviewed at the time of writing. It covers one 20-minute task, and EEG connectivity is not the same as intelligence. As one analysis in The Chronicle of Higher Education put it, the research shows a more complicated picture than the headlines suggest.

The Microsoft and Carnegie Mellon survey

Researchers surveyed 319 knowledge workers about 936 real examples of AI-assisted work. People who had more confidence in AI reported using less critical thinking on those tasks, while people with more confidence in their own abilities reported more. The nature of thinking also changed: effort shifted from producing information to verifying, integrating and supervising what the AI produced. That is not necessarily worse, but it is a different skill, and it only works if people actually do the checking.

The correlation studies

A mixed-methods study of 666 participants found that heavier AI-tool use was associated with lower scores on critical-thinking measures, with the link explained largely by cognitive offloading and strongest among younger users. This kind of finding shows a pattern but cannot prove cause. People who already find thinking tedious may simply be the heaviest AI users.

The high school math experiment

One of the clearest tests comes from a field experiment in a high school math course published in PNAS in 2025. Students who practiced with a standard chatbot did better on practice problems but about 17 percent worse on later tests taken without AI. A version designed as a tutor, which gave hints and guardrails instead of answers, avoided that harm. The lesson is not "AI is bad for learning". It is "AI that gives answers undermines learning, and AI that makes you work can protect it".

Deskilling in medicine

The International AI Safety Report 2026 notes a study in which clinicians’ ability to detect tumours without AI assistance dropped by about 6 percent three months after AI support was introduced. It is one study in one field, but it shows that skills can fade when a tool does part of the job. The report itself stresses that research on AI and critical thinking is still at an early stage.

What the more hopeful studies show

The picture is not all negative. A 2026 systematic review of 63 studies on AI and students’ critical thinking found that almost half reported a significant improvement, mostly when AI use was structured, with guided prompts, reflection and assessment of the student’s own performance. The MIT study’s own twist points the same way: people who thought first and used AI afterward showed higher engagement. The tool matters less than the sequence.

What the evidence says about creativity

AI helps the individual, and narrows the crowd

The best-known study, by Anil Doshi and Oliver Hauser in Science Advances in 2024, asked people to write short stories with or without AI-generated ideas. Stories written with AI ideas were rated as more creative, better written and more enjoyable, and the boost was largest for writers who scored lower on creativity tests. But the stories also became more similar to each other: similarity rose by about 10.7 percent in the group that used a single AI idea. The authors summed it up as more individual creativity at the risk of less collective novelty. You can read the original paper through its DOI.

Where the human enters the process matters

A follow-up line of research, summarised in MIT Sloan Management Review, found that handing creative control to AI produced the most similar outputs, working as co-pilots reduced the effect somewhat, and keeping humans in charge of the early idea stage preserved much more diversity, close to the variety of fully human work. Another study suggests that homogenization is not inevitable when AI is deliberately set up to offer diverse perspectives.

Why this matters beyond one story

If millions of people draw their first ideas from the same few models, the range of ideas in headlines, product names, school essays and marketing copy can shrink. Each piece may look polished, while the whole starts to sound the same. That is a cultural risk even if no single person’s mind is "damaged".

Other ways AI can quietly weaken judgment

  • Overconfidence in fluent answers. A response that sounds sure invites less checking. Models can state errors with perfect confidence.
  • Lost nuance. A 2025 study found that language models often drop qualifiers when summarising research, presenting findings as more general than the evidence supports.
  • Less persistence. Some experiments suggest that when answers are always one prompt away, people try less before giving up.
  • Anchoring. Once you see an AI suggestion, it is hard to think of ideas that are far away from it.

How to read "AI is rotting our brains" headlines

Before you share an alarming study, ask six quick questions:

  1. How many people took part? Dozens is a hint, not a proof.
  2. Has it been peer reviewed? Preprints are early drafts.
  3. Was it measured or self-reported? Surveys capture opinions about thinking, not thinking.
  4. Is it cause or correlation? Heavy AI users may differ from light users in other ways.
  5. What was the task? A 20-minute essay says little about a career.
  6. Does it say how people used the AI? "Used AI" covers everything from copying answers to debating them.

Who is most at risk?

  • Learners building a skill, such as students, new programmers and new writers, because struggling is part of how learning sticks.
  • Heavy daily users with high trust in AI, who may stop checking.
  • Younger users, where several studies find the strongest associations, though causes are unclear.
  • People who use AI to avoid every uncomfortable task, which can feed procrastination. Our guide on how to stop procrastinating explains why avoiding the hard part tends to make it harder.

Delegate or develop: a simple test

Before you reach for AI, ask: "Is this a task I want done, or a skill I want to build?"

Often fine to delegate Better to develop yourself first
Reformatting, summarising your own notes Learning a new subject or language
Boilerplate code or emails you already know how to write Core skills in your profession
Brainstorming extra options after you have your own Forming your own opinion on a topic
Finding errors in work you wrote Writing your first draft
Explaining a concept in a different way Solving practice problems

Nine rules to use AI without dulling your mind

  1. Think first, then ask. Write your own outline, answer or list of ideas before you open the chatbot. Even five minutes of unaided thinking changes how you use the answer.
  2. Use AI as a critic, not a replacement. Ask it to find holes in your argument, spot assumptions or play devil’s advocate on your draft.
  3. Ask for hints, not solutions, when learning. Try prompts like "give me a hint, not the answer" or "quiz me on this". The tutor-style results in the math study suggest this protects learning.
  4. Verify before you trust. Spot-check facts, numbers, quotes and sources. Treat confident wording as a reason to check more, not less.
  5. Keep unplugged practice. Regularly write, calculate, navigate or code without help, the way athletes still train without equipment.
  6. Put it in your own words. Do not paste. Rewrite, then explain the idea out loud without looking. If you cannot, you have not learned it.
  7. Generate ideas before you see AI ideas. List ten of your own first, then use AI to expand or challenge them. This helps protect originality and avoids anchoring.
  8. Notice what you stop doing. If you have not drafted, calculated or figured something out yourself in weeks, that is the skill most likely to fade.
  9. Teach it back. Explaining a topic to someone else is one of the best ways to find out whether you truly understand it.

Advice for parents, teachers and managers

  • Parents: Ask children to explain how they got an answer, not only what the answer is. Encourage AI as a tutor that asks questions.
  • Teachers: Consider assignments that include a process, such as drafts, oral explanations or in-class work, so that thinking is visible even when AI exists.
  • Managers: Build checking into workflows, for example a required "what did you verify?" step, and keep people practicing core skills so the team does not lose them.

Advice for writers and creators

  • Write your own first draft, or at least your own outline and angle.
  • Use AI to challenge your ideas, not to supply them.
  • Add what only you have: personal experience, original reporting, a distinctive voice.
  • If your work reads like everyone else’s, your audience, and search engines, have little reason to prefer it.

Frequently asked questions

Is AI making us dumber?

There is no good evidence that everyday AI use is lowering human intelligence overall. There is growing evidence that heavy, uncritical use can weaken learning, memory and independent judgment, especially for people still building a skill.

Did MIT prove ChatGPT damages the brain?

No. The MIT study found lower measured brain engagement during one essay task in a small group, and it had not been peer reviewed. It is a useful signal, not proof of lasting damage.

Does AI kill creativity?

Not on its own. Research shows AI can improve individual creative output, but it can also make many people’s output more similar. Starting with your own ideas and using AI afterward appears to protect diversity.

Is AI bad for children’s learning?

It can be if it simply provides answers. Tutor-style tools that guide children with hints and questions performed much better in the studies we have, and adult guidance matters.

Should I stop using AI?

Not necessarily. The research points toward using it deliberately: think first, check the output and keep practicing important skills yourself.

How can AI help me learn better?

Ask it to quiz you, explain concepts in different ways, critique your work or generate practice problems, then try them without help.

Will we get more or less capable over time?

Nobody knows yet. The studies so far are short, and the technology is changing quickly. That is why habits matter more than predictions.

The bottom line

AI is not an intelligence-destroying machine, and it is not a harmless shortcut either. It is a powerful tool that rewards people who stay in charge of the thinking. Do your own thinking first, use AI to challenge it, verify what it tells you and keep practicing the skills you care about. If you want to protect yourself from the other side of the AI era, where fakes and scams are getting harder to detect, read our guide on how to spot AI-generated images, videos and voice clones.

Sources and further reading: the MIT Media Lab preprint "Your Brain on ChatGPT", Microsoft Research and Carnegie Mellon’s 2025 survey of knowledge workers, Gerlich (2025), the 2025 PNAS high school mathematics experiment, Doshi and Hauser (2024) in Science Advances, and the International AI Safety Report 2026. This is a fast-moving field, so we update this page as new peer-reviewed studies appear.

Get new articles by email

One short email when we publish. No spam, unsubscribe any time.