MIT Study Explores How ChatGPT Is Quietly Shaping Human Thought

MIT Puts ChatGPT to the Test With Brain Scans

A new MIT Media Lab study dives deep into how ChatGPT affects our thinking literally. Using EEG scans across 32 brain regions, researchers examined the neural impact of using ChatGPT to write SAT-style essays.

ChatGPT users showed the weakest brain activity compared to using Google or writing without tools. The concern?

People may appear productive but are mentally disengaged. This isn’t just about writing, it’s about how our brains are learning to offload thought itself.

54 Participants, 3 Groups, One Big Question

The experiment involved 54 young adults, divided into three groups: one used ChatGPT, another used Google Search, and a third relied solely on their brains.

Each wrote essays over several sessions. The results were stark.ChatGPT users had the lowest engagement across linguistic, behavioral, and neural metrics.

Meanwhile, the brain-only group showed heightened creativity and satisfaction. This wasn’t about productivity but understanding how tools shape cognition over time.

ChatGPT Users Became Lazier Over Time

By the third session, many ChatGPT users simply pasted the prompt and let the AI take over. EEG readings revealed declining brain engagement with each essay. Some participants barely interacted with the writing process at all.

According to lead author Nataliya Kosmyna, “They went from refining and collaborating with ChatGPT to just outsourcing the whole task.” The decline wasn’t just behavioral, it was neurological. Minds were checking out.

Essays Scored as ‘Soulless’ and Repetitive

Two English teachers assessed the essays and described ChatGPT-generated work as “soulless.” Despite grammatical correctness and structure, the essays lacked originality and voice.

Linguistic analysis showed repetitive phrasing and uniform ideas across participants. Essentially, the same tool yielded the same bland thought.

Though it is raising concerns about AI flattening diverse thinking into conformity. Students may feel efficient, but their individuality disappears in the process.

Brain-Only Group Showed Rich Neural Connectivity

The group that wrote essays without tools had significantly more vigorous EEG activity in alpha, theta, and delta brain waves. These bands are linked to creativity, memory, and critical thinking.

Participants in this group were more curious, reflective, and deeply engaged. They also claimed more ownership of their writing. Their brains were more “alive” during the task than the AI-assisted group.

Google Search Users Fared Surprisingly Well

Interestingly, the group using Google Search showed active brain engagement and high satisfaction with their essays. They had to search, evaluate, and integrate information, key cognitive steps ChatGPT users skipped.

This contrast is essential in an era where AI chatbots replace traditional search. Google users are seeded to think critically, even when guided by external input. It’s a middle ground between full automation and complete autonomy.

Memory Loss After ChatGPT Use Is Real

In a follow-up session, ChatGPT users were asked to revise earlier essays this time without the tool. Many remembered little of their original content. EEG scans confirmed lower alpha and theta wave activity, indicating poor memory retention.

“The task was completed, but nothing stuck,” Kosmyna noted. In contrast, those who wrote unaided recalled more, showing how passive use of LLMs may erode memory pathways.

Using ChatGPT May Bypass Deep Learning Processes

The study found that ChatGPT users bypassed essential brain networks for deep learning and long-term retention. While outputs looked polished, they didn’t trigger memory systems. “It’s like eating without digesting,” Kosmyna explained.

This could have dangerous implications for how students and future professionals absorb and retain knowledge. Immediate convenience might be undermining lasting comprehension.

Switching Tools Changed Brain Patterns Instantly

In a final twist, researchers swapped the tool groups. When ChatGPT users wrote without AI, their brain engagement remained low. However, when brain-only participants tried ChatGPT, their EEGs stayed active, recalling previous memory strategies.

This suggests initial habits matter: if users start with critical thinking, they may use AI better. But it’s hard to switch gears back to effortful cognition if they begin with automation.

Ownership of Work Was Lowest in ChatGPT Group

Participants using ChatGPT reported the least sense of authorship over their work. They were less proud, less invested, and less satisfied. In contrast, the brain-only group felt a personal connection to their essays.

Kosmyna sees this as key: “When you don’t struggle, you don’t own it.” With AI, we may be sacrificing emotional investment, something crucial for meaningful learning and growth.

Neural Engagement Faded With Every ChatGPT Session

Over several months, ChatGPT users consistently underperformed in brain connectivity, behavioral quality, and linguistic depth. Their essays got shorter, less insightful, and less reflective. Brain scans revealed diminishing activity session by session.

This trend mirrors what educators fear most: as students become reliant on AI, they may lose their ability and willingness to think critically and independently.

Policy Fears Drove Early Study Release

Though the study hasn’t yet been peer reviewed, Kosmyna released it early to provoke public discussion. “I’m terrified some policymaker will say, ‘Let’s use GPT in kindergarten,’” she warned.

Her urgency reflects growing concern that young, developing brains may be especially vulnerable to AI overuse. Waiting for academic cycles might come too late for shaping responsible AI education policy.

The Loneliness Effect of ChatGPT

Previous MIT studies found that the more people talk to ChatGPT, the lonelier they feel. No matter how advanced they sound, AI chats don’t replace human interaction or emotional validation.

This loneliness could compound the cognitive passivity seen in this latest research. Dependency on AI doesn’t change thinking; it changes how we connect with ourselves and others.

AI Might Be Better Used Later, Not Sooner

The study suggests that introducing AI later in learning may be more beneficial. When students already have strong cognitive habits, AI can support rather than replace them.

The brain-only-to-AI group demonstrated this: they retained curiosity and memory while using ChatGPT. Early dependence, by contrast, weakens those very skills. The takeaway: timing matters.

Is There a Way to Use ChatGPT Responsibly?

MIT’s findings don’t suggest banning ChatGPT; instead, it is with intention. The cognitive cost is lower when learners use it to expand or refine their ideas.

But when AI is used to avoid thinking altogether, the brain disengages. This creates an urgent call for AI literacy: we must teach people how to partner with AI, not surrender to it.

Want to see how others are rethinking AI’s role in learning? Check out how Meta’s new standalone AI stacks up against ChatGPT.

ChatGPT’s Real Power Lies in How We Use It

At its core, ChatGPT isn’t inherently good or a droid. We may forget how to think if we lean on it too much. But if we use it to challenge, not replace, our mental processes, it could be transformative.

As Kosmyna’s research shows, the future of AI and education isn’t about the tech, it’s about the choices we make with it.

Curious how far ChatGPT might actually go? See what Altman envisions for its memory of your entire life.

What do you think about the MIT statement that ChatGPT affects human brains? Please share your thoughts and drop a comment.

Read More From This Brand:

• OpenAI Outage Nearly Over as ChatGPT Recovers
• OpenAI Addresses ChatGPT’s Sycophantic Tendencies
• ICYMI Must-Know Tech News From ChatGPT to GTA 6

Don’t forget to follow us for more exclusive content on MSN.

If you liked this story, you’ll LOVE our FREE emails. Join today and be the first to get stories like this one.

This slideshow was made with AI assistance and human editing.