Heavy Reliance On AI Tools Can Negatively Impact Critical Thinking: MIT Study

A recent study from MIT's Media Lab indicates that AI tools like ChatGPT may negatively impact critical thinking in young users, leading to decreased engagement and learning effectiveness. The findings underscore the need for careful consideration of AI integration in education.

A recent study by MIT's Media Lab raises concerns about the impact of AI tools like ChatGPT on critical thinking. Researchers divided 54 participants, aged 18 to 39, into three groups.

They were tasked with writing SAT essays using either ChatGPT, Google's search engine, or no assistance. Brain activity was monitored using EEGs, revealing that ChatGPT users showed the least brain engagement and performed poorly at neural and behavioural levels.

Over several months, those using ChatGPT became increasingly reliant on the tool, often resorting to copy-pasting by the study's end. The findings suggest that large language models (LLMs) might hinder learning, especially among younger users. Although the study hasn't undergone peer review and has a small sample size, lead author Nataliya Kosmyna felt it was crucial to share the results promptly.

Potential Risks of AI in Education

Kosmyna expressed concern about potential policy decisions that could integrate AI into early education without understanding its effects. "Developing brains are at the highest risk," she stated. The MIT Media Lab has been exploring various impacts of generative AI tools, noting earlier this year that prolonged interaction with ChatGPT can increase feelings of loneliness.

The study also highlighted differences in essay quality between groups. Essays written with ChatGPT lacked originality and were described as "soulless" by English teachers. In contrast, those who wrote without assistance showed higher neural connectivity and satisfaction with their work. The Google Search group also demonstrated active brain function and satisfaction.

Implications for Future Learning

After completing their essays, participants were asked to rewrite one without their initial tool. Those who initially used ChatGPT struggled to recall their content and exhibited weaker brain waves associated with memory processes. Kosmyna noted that while tasks were completed efficiently, they didn't integrate into memory networks.

Conversely, those who initially wrote without assistance showed improved brain connectivity when later allowed to use ChatGPT. This suggests that AI could enhance learning if used correctly rather than diminish it. However, psychiatrist Dr. Zishan Khan warns of potential cognitive consequences for young people heavily relying on AI for schoolwork.

Challenges in Understanding AI Impact

Kosmyna's team inserted traps in their paper to test LLMs' ability to summarise accurately. As expected, some LLMs hallucinated details not present in the paper, such as specifying a version of ChatGPT used in the study. This highlights challenges in relying on AI for accurate information processing.

The team is now working on another study examining brain activity in software engineering with or without AI assistance. Preliminary results indicate even worse outcomes than the current study, suggesting potential implications for companies considering replacing entry-level coders with AI.

Scientific studies on AI's impact are still developing. A Harvard study found generative AI increased productivity but decreased motivation. Meanwhile, MIT distanced itself from a doctoral student's paper claiming substantial productivity improvements through AI.

OpenAI did not comment on these findings but previously collaborated with Wharton online to provide guidance for educators on leveraging generative AI in teaching last year.