AI Overuse: Why Experts Warn About Cognitive Atrophy

Artificial intelligence is rapidly changing how people learn, work and communicate. From writing emails and summarizing reports to answering complicated questions, AI tools can complete tasks in seconds that once required considerable time and effort. However, as these systems become more convenient, researchers and educators are increasingly asking an important question: What happens when people rely on AI so heavily that they stop practicing essential thinking skills?

MIT Sloan professor Eric So is raising that concern in his new book, The Collision: What AI Does to Us. His argument is not that artificial intelligence is inherently harmful. Instead, he warns that excessive dependence on these tools could undermine the mental effort needed to develop expertise, retain knowledge and make independent decisions.

The issue is particularly relevant as schools, universities and businesses encourage people to integrate AI into everyday activities. While automation can improve productivity, the long-term effects may depend on whether people use the technology to strengthen their abilities or simply avoid thinking through difficult problems.

AI Overuse and Cognitive Atrophy: What the MIT Professor Warns

In his October 15 interview with MIT News, So described a growing tendency for people to outsource parts of their thinking to artificial intelligence. He explained that he had caught himself turning to AI to finish work he could previously have completed independently.

That experience helped motivate his book, which examines how AI interacts with human behavior, learning and decision-making.

The central concern is what researchers and commentators call cognitive atrophy: the potential weakening of mental abilities when they are not regularly exercised. The term should not be interpreted as proof that ordinary AI use causes permanent brain damage. Rather, So’s argument concerns the risk of losing opportunities to develop and maintain important intellectual skills.

Learning often requires people to struggle with unfamiliar information, make mistakes, test possible solutions and revise their understanding. These activities can feel inefficient compared with receiving an immediate answer, but they are essential parts of building expertise.

AI can remove much of that effort. A student can ask a chatbot to explain a mathematical problem, generate an essay or summarize a textbook chapter without first attempting the work independently.

The result may be a polished assignment completed quickly, but the student might not fully understand the material behind it.

So argues that this distinction matters because producing an answer and developing the ability to produce that answer independently are two different accomplishments.

The same principle applies in professional settings. An employee who routinely delegates analysis to AI may finish tasks faster while missing opportunities to strengthen judgment and problem-solving abilities.

How AI Overuse Could Affect Critical Thinking

Critical thinking involves evaluating evidence, recognizing assumptions, comparing explanations and reaching conclusions based on reason rather than convenience.

These skills develop through practice. People become better at assessing arguments when they regularly question information, investigate contradictions and explain why they believe a conclusion is justified.

AI assistants can support this process by presenting alternative perspectives, organizing information and helping users identify weaknesses in an argument. Problems arise when users accept generated answers without examining how they were produced.

A confident response can sound authoritative even when it contains errors, incomplete information or misleading assumptions. If users consistently trust the system rather than checking its conclusions, they may become less accustomed to making independent judgments.

Earlier research also offers reasons to take the concern seriously. A study discussed in The Guardian in June 2026 examined how people used AI assistance to assess news headlines and images. The researchers reported that participants’ immediate performance improved with AI support, but independent performance declined when that assistance was removed. The findings suggest that the design of an AI tool and the way people interact with it can influence whether it supports or substitutes for human judgment.

The study had limitations, including a relatively small participant group and a short observation period. It therefore does not establish that every form of AI use damages critical thinking.

Nevertheless, the findings highlight an important distinction: completing a task successfully with assistance does not necessarily mean a person has improved their ability to complete it alone.

For students and professionals, a useful approach is to treat AI-generated conclusions as starting points for investigation rather than final answers that require no further scrutiny.

Why Excessive AI Dependence May Affect Learning and Memory

Learning involves more than exposure to information. People generally need to retrieve knowledge, connect new ideas to what they already understand and apply concepts in different situations.

When AI provides instant summaries and answers, users may spend less time performing those activities themselves.

For example, someone studying a foreign language could ask AI to translate every unfamiliar sentence. This approach may help the person understand a passage quickly, but it offers fewer opportunities to practice recalling vocabulary and constructing sentences independently.

Similarly, a student who relies on generated summaries without reading or questioning the original material may remember the general topic while struggling to explain its details.

The concern is not that convenience automatically destroys memory. Rather, consistently replacing active learning with passive consumption may reduce opportunities to practice remembering and applying information.

So’s broader argument is that the process of learning has value beyond the immediate result. Working through a difficult question can help people develop the knowledge and confidence required to handle similar challenges later.

AI can still be useful as a learning partner. It can generate practice questions, explain difficult concepts in different ways and provide feedback on a student’s reasoning.

The difference lies in how the tool is used. Asking AI to quiz a learner encourages active recall, while asking it to complete every exercise may eliminate the very practice that builds understanding.

The Hidden Pressure Encouraging People to Use AI

AI dependence is not necessarily just a matter of individual habits. Workplace expectations, academic competition and social pressure can all influence how frequently people turn to these systems.

In his MIT News interview, So described what he calls “AI gravity,” referring to the forces that pull people toward AI even when they recognize potential disadvantages.

Employees may feel compelled to use AI because colleagues complete work faster with it. Students may believe that learning to rely on AI is necessary to prepare for future employment. Organizations may reward the speed and volume of output without adequately assessing whether workers understand the results.

These incentives can create a cycle in which people increasingly delegate tasks to technology because everyone around them appears to be doing the same.

Over time, individuals may have fewer opportunities to practice the abilities they will need when AI tools are unavailable, unsuitable or incorrect.

This creates a challenge for schools and employers. Simply telling people to use AI responsibly may not be enough if their institutions reward speed above learning, independent judgment and quality.

A healthier approach would evaluate both productivity and competence. Students should demonstrate that they understand their work, while employees should retain the ability to explain and verify important decisions, even when AI contributes to the process.

AI Overuse and Human Relationships

So also raises concerns about how AI may influence interpersonal relationships.

Conversational AI can provide immediate responses, adapt to a user’s preferences and maintain a consistently agreeable tone. These characteristics can make interactions feel comfortable, particularly when someone is tired, stressed or looking for reassurance.

However, real human relationships involve negotiation, disagreement, compromise and the need to understand perspectives different from our own.

If people become accustomed to interactions in which a system readily accommodates their preferences, conversations with other people may feel more demanding by comparison.

This does not mean that using an AI assistant inevitably weakens social skills or that people should avoid AI companions and conversational tools altogether. The long-term effects depend on many factors, including individual circumstances, the nature of the interaction and how the technology fits into a person’s broader social life.

The important consideration is balance.

AI can help people prepare for difficult conversations, practice interviews or organize their thoughts before discussing a problem with someone else. It becomes more concerning if digital interactions consistently replace meaningful contact with friends, family members, colleagues or communities.

Human relationships offer experiences that AI cannot fully reproduce, including shared responsibility, mutual understanding and the consequences of real-world commitments.

Why AI Can Be Helpful When Used as a Learning Tool

Despite these concerns, the debate over AI should not be reduced to a choice between using the technology and rejecting it entirely.

So’s discussion acknowledges that artificial intelligence can provide genuine benefits when people use it to support learning rather than bypass the process.

A well-designed learning interaction can make AI function more like a tutor than an answer machine. Instead of immediately revealing a solution, the system can ask guiding questions, identify gaps in understanding and encourage the learner to attempt each step independently.

Consider a student who is struggling with a complicated scientific concept. Rather than requesting a complete explanation to memorize, the student could ask AI to identify the foundational ideas, pose questions one at a time and provide feedback on their answers.

This method allows the student to remain actively involved in the learning process.

Writers can follow a similar approach by using AI to identify weaknesses in an argument rather than asking it to replace the entire piece. Programmers can request hints about a coding error before examining a complete solution. Professionals can use AI to compare possible strategies while independently checking the assumptions behind each recommendation.

These practices preserve the user’s role as the decision-maker.

The objective is not to make every task unnecessarily difficult. It is to distinguish between work that can be automated without meaningful consequences and activities that are essential for developing expertise.

Seven Practical Ways to Protect Your Thinking Skills

People do not need to abandon AI tools to remain intellectually engaged. Instead, they can introduce habits that preserve independent reasoning while taking advantage of the technology.

1. Attempt difficult tasks before asking AI.

Spend a few minutes developing your own answer before opening a chatbot. Even an incomplete attempt gives you something to compare with the AI-generated response.

2. Ask for explanations instead of finished answers.

Request step-by-step reasoning, examples and guiding questions. Try to explain the result in your own words before moving on to another task.

3. Verify important claims.

Check facts against credible primary sources, especially when the information concerns health, law, finance, science or major professional decisions. AI-generated confidence should never replace evidence.

4. Practice remembering information without assistance.

After reading a document or studying a topic, close the material and summarize what you remember. Then revisit the source to identify missing or incorrect details.

5. Keep some tasks entirely independent.

Write short summaries, solve practice problems or plan projects without AI from time to time. These exercises help you monitor whether your skills are improving.

6. Use AI to challenge your ideas.

Ask the system to identify counterarguments, expose unsupported assumptions or suggest alternative explanations. Evaluate those suggestions rather than accepting them automatically.

7. Reinvest the time AI saves.

Use saved time to learn new skills, read original material, practice difficult tasks or engage in meaningful conversations. Productivity gains are more valuable when they create opportunities for growth rather than simply increasing the volume of work.

These habits can help users maintain a more deliberate relationship with AI. They also make it easier to identify situations in which a person has become overly dependent on assistance.

What the Debate Means for Schools and Businesses

The discussion around AI overuse has implications beyond individual behavior.

Educational institutions must decide how to incorporate AI without weakening the learning objectives of assignments. This may involve more supervised practice, oral explanations, in-class problem-solving and assessments that measure understanding rather than simply evaluating a finished document.

Teachers can also show students how to question generated information, compare it with reliable sources and recognize uncertainty.

Businesses face a related challenge. AI can reduce repetitive work and help employees analyze large amounts of information, but organizations should not assume that faster output automatically translates into better performance.

Training programs can encourage employees to verify results, explain decisions and develop skills that remain useful when automation fails. Managers can also assess whether AI is improving the quality of work instead of measuring adoption rates alone.

Companies that treat AI as a substitute for expertise may achieve short-term efficiencies while overlooking the long-term importance of maintaining skilled employees.

Conversely, organizations that combine automation with continued human learning may be better positioned to use the technology responsibly.

The broader question is how institutions can benefit from AI while continuing to reward the knowledge, judgment and creativity that people contribute.

Does AI Overuse Mean Artificial Intelligence Is Bad for the Brain?

The available evidence does not justify a blanket conclusion that artificial intelligence is inherently harmful to human cognition.

AI systems can make knowledge more accessible, support people with disabilities, help users explore unfamiliar subjects and provide personalized learning opportunities. They can also reduce repetitive workloads and allow people to devote more time to complex or creative activities.

At the same time, researchers are investigating whether particular patterns of dependence can weaken specific skills or make users more vulnerable to mistakes.

The distinction between correlation, potential risk and proven long-term harm is important. Findings from individual studies should not be interpreted as definitive evidence that all AI use causes cognitive decline.

So’s argument is best understood as a warning about the trade-offs involved in outsourcing mental work. When a task is important for developing expertise, repeatedly skipping the thinking process may undermine the user’s opportunity to learn.

The answer is not necessarily less technology. It is more intentional technology use.

People should consider what they want to accomplish, which skills they need to maintain and whether AI is helping them become more capable or simply making it easier to avoid difficult work.

The Future of AI Should Include Human Intelligence

Artificial intelligence is likely to remain an important part of education, business and everyday life. As these tools become more capable, the challenge will be ensuring that technological progress strengthens human abilities rather than encouraging unnecessary dependence.

Eric So’s concerns highlight an issue that deserves continued research and public discussion. Convenience can be valuable, but it should not become the only measure of success.

The most productive use of AI may be the one that helps people do more while continuing to develop their own knowledge, judgment and creativity.

That means using the technology to explore questions, test ideas and receive feedback without surrendering responsibility for understanding the results.

Ultimately, AI can be a powerful assistant, but people still need to know how to think critically, evaluate evidence and make informed decisions.

The goal should be to build a future in which artificial intelligence expands human capabilities while leaving people better equipped to learn, reason and solve problems independently.

Source note: This article summarizes concerns raised by MIT professor Eric So and related research. Cognitive atrophy is discussed here as a potential risk of overreliance, not as a proven diagnosis caused by routine AI use.