Does ChatGPT Harm Critical Thinking Abilities? A Concerning Look at AI’s Impact on Learning
- A recent MIT Media Lab study suggests that using ChatGPT for writing tasks, such as SAT essays, may significantly reduce brain engagement and critical thinking skills, especially among younger users.
- The study found that ChatGPT users displayed lower neural activity, produced unoriginal content, and became increasingly reliant on the tool over time, often resorting to copy-and-paste tactics.
- Researchers and experts warn of long-term cognitive consequences, urging for education on proper AI use and legislative action to protect developing brains from overreliance on such technologies.
The rapid integration of generative AI tools like ChatGPT into everyday life, especially in educational settings, has sparked both excitement and concern. A new study from MIT’s Media Lab, led by research scientist Nataliya Kosmyna, has raised alarming questions about the impact of these tools on critical thinking and brain development. Conducted with a small but diverse group of 54 participants aged 18 to 39 from the Boston area, the study divided subjects into three groups tasked with writing SAT essays: one using ChatGPT, another using Google Search, and a third relying solely on their own brainpower. The results, recorded via EEG scans across 32 brain regions, paint a troubling picture of diminished neural engagement among ChatGPT users, prompting urgent discussions about the role of AI in learning environments.
The findings were stark. The group using ChatGPT consistently underperformed at neural, linguistic, and behavioral levels compared to the other two groups. Their essays, assessed by two English teachers, were described as “soulless,” lacking originality and often recycling the same expressions and ideas provided by the AI. Over several months, these participants grew increasingly lazy, with many resorting to simply copying and pasting content from ChatGPT by the end of the study. Kosmyna noted a disturbing trend: by the third essay, many users were merely feeding prompts to the AI and making minimal edits, essentially outsourcing the entire creative process. In contrast, the group relying solely on their own cognition showed the highest neural connectivity, particularly in brain waves associated with creativity, memory, and semantic processing. They reported greater satisfaction with their work and a stronger sense of ownership over their ideas.
The Google Search group also demonstrated high levels of brain engagement and satisfaction, highlighting a key difference in how information-seeking behaviors impact cognitive processes. While many people now turn to AI chatbots for quick answers instead of traditional search engines, the study suggests that the active process of searching, evaluating, and synthesizing information online may better support critical thinking than the passive acceptance of AI-generated content. This distinction raises important questions about how we interact with technology and whether convenience comes at the cost of intellectual growth.
Perhaps most concerning was the follow-up task, where participants were asked to rewrite one of their essays under different conditions. The ChatGPT group, now forced to write without the tool, struggled to recall their own work and exhibited weaker brain activity in areas tied to deep memory processes. Kosmyna explained that while the task was completed efficiently with AI, the users failed to integrate the content into their memory networks, essentially bypassing the learning process. On the other hand, the brain-only group, now allowed to use ChatGPT, showed a significant increase in brain connectivity across all EEG frequency bands, suggesting that AI could potentially enhance learning if used as a supplementary tool rather than a crutch. This glimmer of hope underscores the need for balanced approaches to AI integration in education.
The urgency of these findings compelled Kosmyna to release the paper before completing the peer-review process, a first for her team. With a review potentially taking eight months or more, she feared that policymakers might implement AI-driven educational initiatives—such as “GPT kindergarten”—without fully understanding the risks. “Developing brains are at the highest risk,” she emphasized, advocating for education on responsible AI use and the importance of analog brain development. Her concerns are echoed by professionals like Dr. Zishan Khan, a psychiatrist specializing in children and adolescents, who has observed firsthand the overreliance on AI among his young patients. He warns that such dependency could weaken neural connections critical for memory, resilience, and information access, with lasting psychological and cognitive consequences.
Interestingly, the paper’s release on social media revealed another layer of irony: many users ran the study through LLMs to summarize it, missing key nuances. Anticipating this, Kosmyna embedded “AI traps” in the text, such as instructions for LLMs to limit their analysis, ensuring that automated summaries provided only superficial insights. This clever tactic highlights the very issue her research addresses—our growing tendency to let AI think for us rather than engaging with complex material ourselves.
Beyond this study, the broader implications of AI on cognition are still unfolding. Kosmyna and her team are already working on a follow-up paper examining brain activity in software engineering and programming with and without AI assistance, with preliminary results suggesting even more concerning outcomes. If reliance on AI reduces critical thinking and problem-solving skills in technical fields, industries hoping to replace entry-level workers with AI might face unintended consequences across their workforce. Meanwhile, other research offers mixed perspectives: a Harvard study from May found that generative AI boosts productivity but diminishes motivation, while another MIT paper on AI’s potential to enhance worker output was distanced by the institution itself. OpenAI, the creator of ChatGPT, did not comment on the study but has previously collaborated with Wharton Online to provide guidance for educators on leveraging AI in teaching—a move that now seems increasingly complex in light of these findings.
The MIT study, though limited by its small sample size and pre-review status, serves as a critical wake-up call. As generative AI becomes ubiquitous, society must grapple with balancing its undeniable benefits against the potential erosion of fundamental skills. Kosmyna’s call for active legislation and rigorous testing of AI tools before widespread implementation in education is a reminder that technology should serve as a partner, not a replacement, for human thought. For students, educators, and policymakers alike, the message is clear: convenience should not come at the expense of long-term brain development. As we navigate this uncharted territory, fostering curiosity, creativity, and critical thinking remains more important than ever.