MIT Study: Using ChatGPT Reduces Brain Engagement and Memory
Researchers from the prestigious MIT Media Lab have presented alarming findings about the impact of artificial intelligence on the human brain. The study, led by Nataliya Kosmyna, who has worked as a research scientist at the MIT Media Lab since 2021, monitored the brain activity of 54 volunteers aged 18 to 39 from the Boston area using EEG (electroencephalography) across 32 brain regions. Participants were divided into three groups—one used OpenAI's ChatGPT, another used the Google search engine, and the third wrote essays using only their own knowledge without digital assistance.
The results of the study, published by the prestigious MIT laboratory, reveal a troubling trend. The group using ChatGPT showed the lowest brain engagement of all three groups and "underperformed at the neural, linguistic, and behavioral levels." Most interesting was the observation that ChatGPT users became lazier with each subsequent essay over the course of several months and, by the end of the study, often resorted merely to copying and pasting text.
Dramatic Decline in Creativity and Memory Integration
The study's authors instructed participants to write twenty-minute essays based on topics from the SATs (a standardized American test), including topics about the ethics of philanthropy and the pitfalls of having too many choices. The group using ChatGPT submitted extremely similar essays that lacked original thought and relied on the same expressions and ideas. Two English teachers who evaluated the essays described them as largely "soulless."
EEG measurements revealed low executive control and attentional engagement in the ChatGPT group. By the third essay, many writers simply entered instructions into ChatGPT and let it do almost all the work. "It was more like, 'Just give me the essay, edit this sentence, revise it, and it's done,'" Kosmyna says.
By contrast, the brain-only group showed the highest neural connectivity, particularly in the alpha, theta, and delta bands, which are associated with creative thinking, memory load, and semantic processing. The researchers found that this group was more engaged and curious, claimed ownership of their work, and expressed greater satisfaction with their essays.
Google Search Surprisingly Beneficial
The third group, which used the Google search engine, also expressed high satisfaction and showed active brain function. This difference is noteworthy because many people now search for information using AI chatbots instead of traditional Google searches. The group using Google showed moderate brain engagement and satisfaction, similar to those who used only their brains, though not as pronounced.
After writing three essays, participants were asked to rewrite one of their previous attempts—but the ChatGPT group had to do so without the tool, while the "brain-only" group was now allowed to use ChatGPT. The first group remembered little of their own essays and showed weaker alpha and theta brain waves, probably reflecting the bypassing of deep memory processes. "The task was completed, and you could say it was efficient and convenient," Kosmyna says. "But as we show in the paper, you essentially integrated none of it into your memory networks."
Warning Against Implementation in Education Systems
By contrast, the second group performed well and showed a significant increase in neural connectivity across all EEG frequency bands. This raises hopes that artificial intelligence, if used correctly, could enhance learning rather than weaken it. Kosmyna emphasizes that education about how to use these tools and reinforcing the fact that the brain needs to develop in a more analog way are absolutely critical.
This is the first preprint Kosmyna has ever published. Although her team submitted it for peer review, they did not want to wait for approval, which can take eight months or more, before drawing attention to an issue that Kosmyna says is affecting children right now. "What really motivated me to release it now rather than wait for the full peer review was that I'm afraid that in six to eight months, some policymaker will decide, 'Let's create GPT kindergarten.' I think that would be absolutely wrong and harmful," she says. "Developing brains are most at risk."
Psychiatric Consequences for Young Users
Psychiatrist Dr. Zishan Khan, who treats children and adolescents, says he sees many children who rely heavily on artificial intelligence for schoolwork. "From a psychiatric perspective, I see that overreliance on these large language models (LLMs) can have unintended psychological and cognitive consequences, especially for young people whose brains are still developing," he says. "Those neural connections that help you access information, remember facts, and be resilient—all of them will weaken."
Ironically, after the paper was published, several social media users ran it through LLMs to summarize it and then posted the findings online. Kosmyna expected people to do this, so she inserted several AI traps into the paper, such as instructing LLMs to "read only this table below," thereby ensuring that the LLMs would return only a limited overview of the paper.
AI Hallucinations as Evidence of the Problem
She also found that LLMs hallucinated a key detail: nowhere in her paper did she specify which version of ChatGPT she had used, but AI summaries claimed that the paper had been trained on GPT-4o. "We specifically wanted to see this because we were fairly certain that the LLM would hallucinate it," she says with a laugh.
Kosmyna says that she and her colleagues are now working on another similar paper testing brain activity in software engineering and programming with or without artificial intelligence, and says that so far, "the results are even worse." According to her, this study could have implications for many companies hoping to replace their entry-level coders with artificial intelligence. Although efficiency may increase, growing reliance on artificial intelligence could potentially reduce critical thinking, creativity, and problem-solving across the remaining workforce, she argues.
Conflicting Results from Other Studies
Scientific studies examining the impacts of artificial intelligence are still in their infancy and continue to evolve. A Harvard study from May found that generative artificial intelligence made people more productive but less motivated. Also last month, MIT distanced itself from another paper written by a doctoral student in its economics program that suggested artificial intelligence could substantially improve worker productivity.
OpenAI did not respond to a request for comment. Last year, in collaboration with Wharton Online, the company published guidelines for educators on how to use generative artificial intelligence in teaching. The MIT Media Lab has recently devoted significant resources to studying the various impacts of generative artificial intelligence tools. Studies from earlier this year, for example, found that, in general, the more time users spend talking to ChatGPT, the lonelier they feel.
This study represents the first systematic look at how the use of artificial intelligence tools such as ChatGPT may affect brain activity and cognitive abilities. While the results raise concerns about the future of education and cognitive development, they also suggest that, when used correctly, artificial intelligence could support learning rather than harm it.
