Is it true that MIT research has found that using AI can reprogram the brain and lead to cognitive decline?
A preprint study by researchers at the Massachusetts Institute of Technology (MIT), titled 'The Brain with ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for an Essay-Writing Task,' has unintentionally generated a lot of sensationalism and has become a hot topic. Regardless of the researchers' intentions, some have pointed out that the content of the study is flawed.
[2506.08872] Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task
Commentary: 'Your Brain on ChatGPT' Preprint
https://residualinsights.com/all-hype-no-bite-your-brain-on-chatgpt-preprint/
MIT Study Finds AI Use Reprograms the Brain, Leading to Cognitive Decline | Hacker News
https://news.ycombinator.com/item?id=45114753
To clarify how using a large-scale language model (LLM) like ChatGPT affects human cognitive costs, MIT's Natalia Kosmina and her colleagues conducted an experiment in which subjects used LLMs and measured their brain waves.
It turns out that people who use AI chatbots experience a significant decrease in brain activity - GIGAZINE
The research contents were as follows:
The subjects were divided into three groups: the 'LLM group' who used LLM, the 'search engine group' who used a search engine, and the 'brain-only group' who did not use any search engine, and asked to write essays.
The essay writing session was divided into four sessions.
During the fourth session only, the LLM group was instructed not to use any tools, while the brain-only group was instructed to use the LLM.
As a result, all three groups showed significantly different neural connection patterns, with the brain-only group showing the strongest neural connections, followed by the search engine group, and the LLM group showing the weakest.
By the fourth session, the LLM group showed a decrease in specific brainwave activity, while the brain-only group showed an increase in neural connections in the occipito-parietal and prefrontal cortices.
This research spread as
Many have pointed out that this research was published on arXiv before it was peer-reviewed.
Based on this premise, it has been pointed out that the results should not be treated as sensationalized, as there were a total of 54 subjects, and only nine in each of the LLM and brain-only groups who progressed to the fourth session. This means that the sample size is small for neuroscience research, and the subjects were young, ranging in age from 18 to 39 (average age 22.9), making the results difficult to generalize.
Other points raised included whether it was appropriate to interpret a 'high level of activity' as 'good,' whether participation in the fourth session was optional, and why the results of the optional fourth session were presented as if they were the subject of a paper, even though the experiment was considered complete after participants had attended the first three sessions, and whether the fourth session was only 20 minutes long, and whether stress or pressure might be a contributing factor.
The reaction to the paper was so great that it prompted such criticism, that the authors of the paper were forced to set up a special website and post comments such as, 'The uproar is growing because many people have used LLM to summarize the paper (including errors),' 'It is incorrect to say that LLM makes people stupid,' 'The paper lists its limitations,' and 'This is the first attempt of its kind, so we hope to see many more papers and research published in the future.'
Related Posts:
in Science, Posted by log1p_kr