Does ChatGPT harm critical thinking abilities? New research from MIT’s Media Lab offers concerning insights into this question. In a recent study involving 54 participants aged 18 to 39 from the Boston area, researchers investigated how the use of OpenAI's ChatGPT influences cognitive engagement and critical thinking skills compared to traditional methods.
Participants were divided into three groups: one used ChatGPT, another relied solely on Google searches, and the third group wrote essays independently, using only their brains. Each participant wrote multiple SAT-style essays while researchers measured brain activity through EEG across 32 regions.
Results revealed that the ChatGPT group consistently exhibited lower brain engagement levels and notably poorer performance on cognitive, linguistic, and behavioral measures. Over time, participants using ChatGPT grew progressively lazier, frequently resorting to copy-and-paste techniques. English teachers reviewing these essays described them as largely "soulless," noting repetitive expressions and minimal original thought.
Conversely, the group writing essays independently demonstrated the highest neural connectivity, particularly in brain waves associated with creativity, memory, and semantic processing. These participants also reported higher satisfaction and a sense of ownership over their work. The Google search group similarly showed strong cognitive engagement, suggesting traditional internet searches may still support effective learning compared to AI-generated content.
Interestingly, the researchers conducted a follow-up task where participants rewrote one of their previous essays. The ChatGPT group, now without AI assistance, struggled significantly, recalling minimal content from their earlier work. EEG readings indicated decreased brain activity linked to deep memory processing. Meanwhile, the independent-writing group, allowed to use ChatGPT in this phase, exhibited improved cognitive engagement, suggesting strategic, mindful use of AI tools might enhance rather than diminish learning.
Nataliya Kosmyna, lead author of the study, expressed concern over potential long-term developmental risks associated with early and extensive reliance on AI tools. Despite the small sample size and pending peer review, Kosmyna chose to release these findings promptly to caution against premature integration of AI into education, particularly for younger learners. She emphasized the necessity of balanced, responsible use and proactive legislation to safeguard cognitive development.
Psychiatrist Dr. Zishan Khan echoed these concerns, highlighting potential long-term cognitive and psychological consequences. Khan emphasized that overreliance on AI might weaken critical neural pathways involved in memory, resilience, and information retrieval.
An intriguing aspect emerged when social media users immediately utilized ChatGPT to summarize the study upon its release. The researchers had anticipated this and embedded intentional "AI traps," leading to LLM-generated summaries with inaccuracies—such as falsely claiming the study specified using GPT-4o, an incorrect detail fabricated by the AI.
Kosmyna’s team is currently extending this research into the software engineering domain. Early findings suggest even more pronounced adverse impacts, which could inform corporate strategies around integrating AI tools in programming roles, potentially undermining critical thinking and problem-solving capacities.
This study aligns with other emerging research indicating mixed outcomes of AI usage. A recent Harvard investigation found that while generative AI boosts productivity, it may simultaneously reduce motivation. MIT itself previously distanced from another paper overly optimistic about AI's productivity benefits.
Ultimately, as society navigates the rapid integration of generative AI, the emerging consensus underscores caution and careful consideration in educational and professional contexts. The potential risks to critical thinking and cognitive development necessitate mindful approaches, clear guidance, and rigorous research-based policymaking to ensure AI serves as a beneficial, rather than detrimental, resource.