A new study from researchers at MIT has ignited debate over the effects of AI tools such as ChatGPT on learning and critical thinking. 

The results of the study suggest that prolonged use of generative artificial intelligence for academic tasks, such as essay writing, could lead to diminished cognitive engagement and deteriorated intellectual capabilities, an effect the researchers called “cognitive debt.”  

“Cognitive activity scaled down in relation to external tool use,” the researchers at the Massachusetts Institute of Technology’s Media Lab write in their study.

Led by Nataliya Kosmyna, the study involved 54 participants aged 18 to 39 from the Boston area. The participants were divided into three groups and tasked with writing several SAT-style essays. One group leveraged ChatGPT, a Large Language Model built by an artificial intelligence program that is designed to understand and generate human language.

Another group used Google Search, and the final group’s members relied solely on their own knowledge and skills. Throughout the study, researchers measured participants’ brain activity via electroencephalography and analyzed the quality and originality of their writing.  

“Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity,” the researchers wrote. “Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.”

Participants who used ChatGPT exhibited significantly lower neural engagement than those in the other two groups. Over time, their essays became increasingly generic and repetitive, with researchers noting a lack of creative thought and originality.

By the third essay, many in the ChatGPT group resorted to simple copy-and-paste workflows, actively bypassing deeper cognitive processes. 

Meanwhile, the group members who wrote essays with only their brains demonstrated the highest levels of cognitive engagement. Their brain scans showed increased neural connectivity in areas associated with creativity, memory, and critical thinking. The group members also expressed more satisfaction with their work and displayed stronger recall of their own content. 

Similarly, participants in the Google Search group exhibited high levels of brain activity compared to those using ChatGPT.  

The final phase of the study involved role reversals, with ChatGPT users being asked to write their final essay without the AI tool while the brain-only group was allowed to use ChatGPT. Those transitioning to manual writing from AI struggled significantly. They had a harder time recalling their previous content and faced challenges producing essays comparable to those produced by participants who had started without AI.

Despite the concerning findings, the study’s authors urged caution in interpreting the results, citing the limited sample size and the preliminary nature of their conclusions. Only 18 participants completed the final essay-writing phase, a factor that may have affected the findings. 

Nonetheless, the study raises important questions about the educational use of generative AI. The researchers drew comparisons to the introduction of calculators in the 1970s, when educators responded to the technology by raising the complexity of exam questions, ensuring students still had to engage in high-level problem-solving. The researchers suggest a similar approach may be needed for AI.