JM JM MindMintPsychology • AI Ethics • Mental Wellness
May 9, 2026 · Jagadish Mokashi

AI and the Brain: What Research Says About Thinking and Learning

Illustration for an article about AI use, independent thinking and learning

By Jagadish Mokashi | JM MindMint

I use AI tools for research, writing, learning, and solving everyday problems. They can save time and help explain difficult ideas. But I have also noticed how easy it is to ask for an answer before trying to work through the question myself.

That raises an important question: when does AI support our thinking, and when might it replace some of the practice through which we learn?

There is no simple answer. Research is still developing, and the studies available so far examine specific tasks and groups of people. They do not prove that everyday AI use damages the brain. They do, however, offer useful lessons about learning, persistence, attention, and mental workload.

What Does AI Do to the Way We Think?

AI can change the steps we take to complete a task. Instead of recalling information, planning an answer, drafting, checking, and revising, a person may ask a chatbot to perform several of those steps.

This is an example of cognitive offloading: using an external tool, note, device, or another person to reduce the mental effort required to remember or complete something. Cognitive offloading is not automatically harmful. People have always used calendars, calculators, maps, books, and search engines to support their thinking.

The important question is what the person still practises. A calculator can help someone check arithmetic, but a learner who never practises the underlying method may struggle when the calculator is unavailable. Similarly, an AI explanation can support learning when a person questions it, checks it, and applies the idea. Simply copying an answer may provide less opportunity to practise.

For more on how technology shapes behaviour and attention, see our introduction to cyberpsychology.

What the 2025 MIT Essay-Writing Study Actually Found

In June 2025, researchers associated with the MIT Media Lab released a preliminary study titled “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” The researchers recruited 54 participants from several nearby universities and compared three conditions: writing essays with ChatGPT, using a search engine, or writing without an external tool.

The team used electroencephalography (EEG) to record brain activity during the writing task. Across the first three sessions, the researchers reported different patterns of brain connectivity among the groups. The brain-only group showed the strongest and most distributed connectivity, the search-engine group showed intermediate patterns, and the ChatGPT group showed the weakest connectivity in the reported analysis.

Only 18 participants completed a fourth crossover session. In that session, some participants who had previously used ChatGPT were asked to write without it. The researchers reported lower connectivity in some EEG networks in this group and differences in memory recall and essay ownership across conditions.

The authors used the phrase cognitive debt to describe a possible cost of outsourcing parts of a thinking task. It is their framing of preliminary findings, not a clinical diagnosis. EEG connectivity is not a direct measure of intelligence, permanent brain damage, or a person’s general ability to think.

There are important limitations. The sample was small and geographically concentrated, the task focused on essay writing, and only 18 people completed the crossover session. The authors also state that the findings should be treated cautiously and may not generalise to other tasks, age groups, or AI systems. The paper was released as a preprint, so readers should not treat it as settled evidence.

The most useful lesson is not “AI makes people less intelligent.” It is that the way we use a tool may change how much of the task we actively perform. That is a question worth studying further.

Can AI Use Create Mental Fatigue?

In March 2026, researchers from Boston Consulting Group and the University of California, Riverside described a workplace experience they called AI brain fry in Harvard Business Review.

The researchers surveyed 1,488 full-time workers in the United States. Among workers who used AI, 14% reported mental fatigue associated with excessive use, interaction with, or oversight of AI tools beyond their cognitive capacity. Participants described experiences such as mental fog, a buzzing feeling, difficulty focusing, slower decisions, and headaches.

This research is about workplace cognitive load, especially situations where people oversee multiple tools or AI agents, review large amounts of output, and remain responsible for checking decisions. It does not show that every person who asks a chatbot a question experiences brain fatigue. Nor is “AI brain fry” a formal medical diagnosis.

The findings also point to an important distinction: AI may reduce effort when it takes over routine tasks, while creating additional mental work when a person must supervise, verify, and coordinate many outputs. The design of the work matters.

What a 2026 Problem-Solving Experiment Adds

A separate set of randomized experiments, reported by UC Berkeley in October 2026, examined what happens when people use AI to help with fraction problems and reading comprehension, then have to continue without it.

Across three experiments, 1,222 online participants took part. In the first two experiments, participants worked on fraction problems; the third used an SAT-style reading-comprehension task. People with access to AI could perform better while the tool was available. When the researchers withdrew the tool, however, the AI-assisted groups showed lower accuracy or persistence on the remaining problems than comparison groups.

The updated paper was presented at the Conference on Language Modeling. The result is important, but it should be described accurately: it concerns short, controlled tasks and what happened immediately after AI assistance was removed. It does not prove that general intelligence falls by a fixed percentage after ten minutes, or that every kind of AI support has the same effect. The researchers’ project summary explains the experiments and their findings.

One practical implication is that assistance should not remove every opportunity to practise. Hints, explanations, worked examples, and feedback can be useful when learners still attempt the problem and check whether they can solve a similar one independently.

Why an Instant Answer Can Be So Appealing

An immediate answer can feel rewarding because it reduces uncertainty and effort. But it would be too simplistic to claim that every AI response produces a measurable dopamine release. Motivation and reward learning involve multiple brain systems, expectations, context, and individual differences.

Convenience can also shape habits. If we repeatedly reach for a tool whenever a task feels difficult, we may become less willing to sit with uncertainty or make a first attempt. This is not proof of brain damage. It is a behaviour pattern that can be noticed and changed.

Our article Why We Open Our Phones Automatically explores how cues and routines can shape everyday technology habits.

How to Use AI Without Giving Up Your Own Practice

The goal does not need to be avoiding AI. It can be using AI in a way that leaves room for your own judgement, memory, and skill development.

  • Try first. Spend a few minutes outlining your own answer or attempting the problem before asking AI for help.
  • Ask for hints before solutions. Request a clue, a question, or the next step instead of asking the tool to complete the entire task.
  • Explain the answer in your own words. After reading an AI response, close it and explain the main idea without looking.
  • Check important claims. Compare factual, medical, legal, financial, or academic information with reliable primary sources. AI can make mistakes and present them confidently.
  • Practise retrieval. Test yourself later. Try to recall the idea, solve a similar problem, or recreate the steps without assistance.
  • Use one tool with a clear purpose. Constantly switching between multiple tools and reviewing streams of output can increase workload.
  • Take a break when overloaded. If you notice fatigue, frustration, or difficulty concentrating, pause and return to the task later.
  • Keep some work independent. Writing, reading, problem-solving, and reflection without AI can help maintain confidence in your own abilities.

These are practical learning habits, not a treatment or a guarantee of improved cognitive performance. Adapt them to the task and your circumstances.

What About Children and Students?

Children and young adults are still developing knowledge, study habits, judgement, and independent problem-solving skills. That does not mean AI should automatically be banned. It means the purpose of using it should be clear.

For example, a student can ask AI to explain a difficult concept in simpler language, then close the tool and answer practice questions independently. A teacher can ask learners to show their working, discuss why an answer is correct, or compare an AI explanation with a textbook. Parents can encourage children to use AI as a study aid rather than as a substitute for every attempt.

Research on how AI use affects children and young adults is still developing. Findings from adult essay-writing or short online problem-solving experiments should not be generalised to every age group. Schools and families can focus on whether learners understand, explain, and independently apply what AI helps them produce.

For a wider perspective on how digital environments influence young people’s behaviour, explore our article on social comparison and our discussion of human bias in AI.

Can AI Also Help Us Learn?

Yes. AI can offer alternative explanations, generate practice questions, provide feedback, support language learning, and help people explore unfamiliar topics. It can also make information more accessible to people who benefit from simplified explanations or assistive support.

The difference is whether the tool supports the learner’s activity or replaces it. An explanation followed by independent practice is different from copying an answer without understanding it. A draft reviewed and improved by a person is different from submitting generated text as if it reflected skills the person has not practised.

AI is neither automatically a threat nor automatically a teacher. Its value depends on the task, the quality of the output, the user’s judgement, and whether the person remains actively involved.

One Last Thought

AI can help us work faster, explore ideas, and overcome barriers. It can also make it tempting to skip the effort that helps us learn. The research does not justify fear-based claims that AI is destroying the human brain. It does encourage us to ask better questions about attention, persistence, workload, and learning.

Use AI to extend your thinking, not to surrender your judgement. Ask questions, challenge answers, check sources, and practise doing important tasks yourself. The aim is not to compete with a machine at every task. It is to make sure that the human being using the tool continues to understand what is happening and why.

References and Further Reading

JM MindMint publishes educational psychology content. This article is for general information and is not a clinical assessment or a substitute for professional advice.

Support JM MindMint