An MIT EEG study showed that using large language models to write papers reduces neural engagement, particularly germane cognitive load involved in building mental schemas, compared to writing without AI.
"There was a recent paper from MIT that I I actually it was it is very much what I spend a lot of my time talking about and thinking about... So there were three different categories: people who had to write the papers, you know, just with their using their brain only... case two would be I get to use search engines... and then a third would be I use LLMs to write my paper... they were using EEG to look at neural patterns across the brain to understand how much neural engagement happened... And what they found is basically the germane cognitive load is what gets impacted most by using LLMs" (said at 0:56:26)
A systematic search of PubMed and Europe PMC did not locate a peer-reviewed publication matching the described three-arm MIT EEG study comparing brain-only, search engine, and LLM writing conditions on germane cognitive load and neural engagement. While theoretical discussions and preprints explore generative AI's impact on cognitive load and learning, no indexed experimental EEG study meeting the specific design described by the speaker was identified in the biomedical literature.