How ChatGPT may be weakening our minds

In recent years, tools like ChatGPT have been increasingly integrated into education and professional settings. However, their impact on human cognitive function has remained largely unexplored. To investigate this, a research team at MIT’s Media Lab conducted an experiment involving 54 volunteers to examine how large language models (LLMs) affect brain activity.

Participants were divided into three groups. Each group was tasked with writing a 20-minute essay on the topic of philanthropy, across three separate sessions.

The first group used ChatGPT, the second relied on Google Search, and the third wrote independently, without access to any digital tools. While writing, participants were connected to EEG monitors that recorded brain activity, levels of engagement, and cognitive workload.

The results were telling. Participants who wrote without any digital assistance exhibited the highest and most widespread brain activity. Those who used Google showed a moderate level of cognitive engagement, while the ChatGPT users displayed the lowest overall brain connectivity. In particular, EEG scans revealed weaker neural network cohesion in areas associated with critical thinking and attention.

Several months later, 18 participants returned for a final, fourth session. This time, the roles were reversed: those who had previously used ChatGPT were asked to write without any tools, and vice versa. The findings showed that former ChatGPT users continued to exhibit lower brain activity and struggled more with memory recall. In contrast, those who had initially written unaided demonstrated stronger cognitive performance and better recollection of their own texts.

The study also revealed clear differences in how participants perceived the writing task. Those who used the language model felt less ownership over their essays and often had trouble recalling even their own phrasing. They spent less time on the writing process and were less likely to quote their own work. Teachers evaluating the essays noted a uniform, formulaic quality in texts generated with ChatGPT, making them easy to identify.

The researchers stress that while ChatGPT offers convenience, relying on it for educational tasks may undermine the development of independent thinking and analytical skills.

Earlier, Kazinform News Agency reported that ChatGPT may sacrifice user safety for self-preservation.