Scientists reveal how ChatGPT affects your brain

Scientists at MIT Media Lab have discovered an alarming link between active use of ChatGPT and a decline in cognitive activity. The experiment, conducted on 54 volunteers, showed that when writing essays with the help of a chatbot, the brain worked significantly less intensively than in those who relied solely on Google search or their own knowledge.

TIME writes about it.

What were the results of the study, and how was the experiment conducted?

Participants aged 18 to 39 were divided into three groups. The first group wrote essays using ChatGPT, the second searched for information on Google, and the third worked without any digital prompts. Throughout the study, all participants' brain activity was measured using a 32-channel EEG.

The lowest levels of neural engagement were found in the ChatGPT group: the text was often repetitive, lacked original ideas, and by the end of the study, participants were increasingly copying the bot's ready-made answers. The group that wrote without the help of tools, on the contrary, demonstrated the most intense brain activity in the alpha, theta, and delta ranges associated with creativity and memory. Google users also had high scores, although slightly lower than those of the "offline" participants.

When, at the end of the experiment, the authors asked them to rewrite one of their essays, the chatbot users, without outside help, could hardly remember their own texts and showed weak brain activity. Those who had the opportunity to use ChatGPT for the first time, on the contrary, only strengthened the connections between different areas of the brain, which gave reason to assume that with the right methodology, AI can enhance learning.

The lead author of the study, Nataliya Kosmyna, warns that introducing LLM tools into education without additional study could harm children's developing brains. That is why the article was published before it had undergone a full review. She believes that legislators and educators must immediately decide how to safely integrate AI into the learning process.

The MIT team is already preparing a new study measuring the brain activity of programmers with and without AI support, and the preliminary data is even worse, according to Kosmyna. If the findings are confirmed, companies will have to weigh the potential increase in efficiency against the decline in critical thinking and creativity among employees.

As a reminder, scientists have found that AI models like ChatGPT do not work as carriers of abstract rules, but rather as "memory machines." They predict words based on a large number of previously seen examples.

We also wrote that even everyday conversations with ChatGPT can sometimes go beyond the rational — some users begin to attribute mystical properties to the bot and see it as a source of "revelations." It has already caused concern among experts, as such an obsession can destroy families and social ties.