MIT Study Raises Concerns: ChatGPT May Be Eroding Critical Thinking Skills

A new study from the Massachusetts Institute of Technology’s Media Lab is raising alarms about how reliance on ChatGPT and other large language models (LLMs) may be weakening critical thinking, memory, and creativity — especially among younger users. The research, led by MIT scientist Nataliya Kosmyna, found that people who used ChatGPT to help write essays engaged their brains significantly less than those who relied on their own knowledge or even traditional Google searches.

Although the study has not yet been peer reviewed, its findings have sparked immediate debate in the academic community, with experts warning about the risks of integrating AI into education without careful oversight.

How the Study Was Conducted

The research team recruited 54 participants between the ages of 18 and 39 from the Boston area. They divided them into three groups:

1. ChatGPT Group – Participants used OpenAI’s ChatGPT to help write their essays.

2. Google Search Group – Participants used traditional search engines to find information.

3. Brain-Only Group – Participants wrote without any digital assistance.

Each person was asked to write several 20-minute essays based on SAT-style prompts, covering topics such as the ethics of philanthropy and the psychological effects of having too many choices.

What made the experiment stand out was the use of EEG (electroencephalogram) technology to monitor brain activity across 32 regions. This allowed researchers to measure engagement, memory processing, and creativity in real time.

The Results: Lowest Engagement Among ChatGPT Users

The findings were clear and consistent. The ChatGPT group demonstrated the lowest brain engagement of the three groups. EEG readings showed weaker executive control and attentional focus, particularly in alpha, theta, and delta brain wave bands associated with creativity, memory retention, and semantic processing.

Kosmyna noted that participants using ChatGPT often began with some interaction but quickly moved to simply pasting the prompt into the chatbot, requesting a full essay, and then making minimal edits. Over time, their essays became increasingly formulaic and similar to one another, with two English teachers describing them as “soulless” and lacking originality.

By contrast, the Brain-Only Group displayed the highest neural connectivity, showing active thinking, curiosity, and greater satisfaction with their work. They also demonstrated a strong sense of ownership over their writing. The Google Search Group performed somewhere in between, showing high brain activity and satisfaction, but with a different workflow that required information gathering and synthesis.

Memory and Learning Impact

The study also tested how well participants retained knowledge from the essay tasks. After completing three essays, each participant was asked to rewrite one of them. However, the ChatGPT group was told they could not use ChatGPT this time.

The results were striking: ChatGPT users remembered little of their original essays and showed weaker alpha and theta brain waves, indicating poor deep memory integration. In simple terms, they had completed the task efficiently, but the information had not been meaningfully processed or stored in their memory.

Conversely, the Brain-Only Group, now allowed to use ChatGPT for the rewrite, retained a clear memory of their essays and demonstrated even greater brain connectivity than before. This suggested that when AI is introduced after an initial period of independent thinking, it may actually enhance rather than diminish learning.

Risks for Developing Minds

Kosmyna says she decided to release the findings before full peer review because she is concerned that policymakers might move too quickly to integrate AI into early education.

“What really motivated me to put it out now… is that I am afraid in 6-8 months, there will be some policymaker who decides, ‘let’s do GPT kindergarten.’ I think that would be absolutely bad and detrimental,” she told reporters. “Developing brains are at the highest risk.”

Psychiatrist Dr. Zishan Khan, who works with children and adolescents, agrees. He warns that heavy dependence on AI for academic work could impair the development of neural pathways essential for information recall, resilience, and problem-solving. “These neural connections that help you in accessing information, the memory of facts, and the ability to be resilient: all that is going to weaken,” he said.

Earlier MIT Findings on AI and Well-Being

This is not the first time the MIT Media Lab has examined the effects of AI tools on human behavior. Earlier in 2025, the lab published studies showing that extended conversations with ChatGPT were correlated with increased feelings of loneliness among users.

Kosmyna’s new work builds on this by focusing specifically on education and the cognitive demands of writing. With more students turning to ChatGPT for homework and assignments, she believes it is essential to understand not only how AI changes output quality, but also how it affects the mental processes behind that work.

AI Isn’t Always the Problem — It’s How We Use It

Interestingly, the study also points to scenarios where AI could be beneficial. When the Brain-Only Group used ChatGPT after generating their own ideas, they experienced increased neural activity. This suggests that AI may be most helpful as a refinement tool rather than a primary generator of content.

“The key is that your brain does need to develop in a more analog way,” Kosmyna emphasized. “Education on how we use these tools, and promoting active engagement, is absolutely critical.”

Wider Implications Beyond Education

Kosmyna’s team is already working on a follow-up study exploring how AI affects problem-solving and critical thinking in software engineering. Preliminary results suggest that programmers who rely heavily on AI tools also show reduced brain engagement and problem-solving activity — an outcome that could have significant consequences for companies considering replacing entry-level coders with AI.

While AI may boost efficiency, over-reliance could undermine the creative and analytical skills needed to maintain and improve complex systems over time.

Limitations of the Study

As with any early-stage research, this study comes with limitations. The sample size of 54 participants is relatively small, and the paper has not yet undergone peer review. This means its findings should be interpreted as preliminary rather than definitive.

However, Kosmyna defends the decision to release it now, saying that waiting eight months for peer review could mean missing an opportunity to influence policy discussions already underway.

Mixed Reactions in the AI Community

OpenAI has not commented on the study. In the past, the company has collaborated with educators to create guidance on how to integrate generative AI responsibly in classrooms. Many in the AI field argue that the technology is a tool like any other — its impact depends on how it is used.

Still, the MIT findings raise important questions for educators, policymakers, and parents about where to draw the line between helpful assistance and harmful dependency.

The Takeaway

The MIT Media Lab’s latest research underscores a growing concern: while AI tools like ChatGPT can make tasks easier and faster, they may also encourage mental shortcuts that bypass the deep thinking and memory processes essential for learning.

Used without guidance, ChatGPT risks turning students into passive editors rather than active thinkers. Used with intention — after initial brainstorming or research — it could serve as a powerful aid without replacing the cognitive work that education is meant to foster.

For now, the message from researchers is clear: if we want to preserve creativity, memory, and critical thinking, we must ensure that AI supplements rather than substitutes human effort. As Kosmyna puts it, “These tools aren’t going away. The challenge is making sure we use them in ways that make our brains stronger, not weaker.”