science.slashdot.org
MIT Experiment Finds ChatGPT-Assisted Writing Weakens Student Brain Connectivity and Memory
A study at MIT involved 54 college students who were divided into three categories: those who used only AI tools, those who used search engines, and those who used no tools at all. The study found that the group that used only AI tools showed the weakest neural connectivity across all frequency bands, while the group that used no tools showed the strongest connectivity. In the first session, 83% of the AI group could not quote any line they had just written and none of them produced a correct quote. Only nine of the 18 participants in the AI group claimed full authorship of their work, compared to 16 of the 18 participants in the no-tool group. The neural coupling in the AI group declined further over repeated use. When the participants in the AI group were later asked to write without assistance, their frontal-parietal networks remained subdued and 78% of them again failed to recall a single sentence accurately. However, when students who first wrote unaided were introduced to ChatGPT in a crossover session, they produced the highest connectivity sums in alpha, theta, beta, and delta bands, indicating intense integration of AI suggestions with prior knowledge. The study suggests that habitual reliance on large language models can lead to "cognitive debt," trading immediate fluency for weaker memory, reduced self-monitoring, and narrowed neural engagement. The authors warn that this can have negative consequences in the long run. Overall, the study highlights the importance of balancing the use of AI tools with independent thinking and writing.
