“Brain rot,” Oxford’s Word of the Year in 2024, refers to both the mental degradation from consuming too much unchallenging or trivial content online and to the content itself. Even though “brain rot” has been in current use only since the early 2000s, the earliest known example is in Thoreau’s 1854 book Walden: “While England endeavors to cure the potato-rot, will not any endeavor to cure the brain-rot, which prevails so much more widely and fatally?”
The MIT Media Lab recently conducted a study of the cognitive cost of using large language models (LLMs) such as ChatGPT or search engines instead of just our unassisted brains.
Assigned to draft an essay, students were divided into three groups: unlimited use of ChatGPT; unlimited use of search engines (but no LLMs); and no use of any electronic media, only their brains.
The researchers used electroencephalography to record students’ brain activity to assess cognitive engagement (extent to which individuals invest mental effort and attention in learning or problem-solving activities) and cognitive load (amount of mental effort and resources required to process information and perform tasks), in addition to conducting interviews with students after drafting those essays. Here’s what they found.
The brain-only group exhibited the strongest, widest-ranging brain networks. The search engine group showed intermediate engagement. The ChatGPT group elicited the weakest overall brain activity.
As a practical matter, this means that the brain-only group had the highest memory recall and reengagement with the subject of the essay. The search engine group had some memory recall, and the ChatGPT group fell behind in the ability to quote from their own essays written just minutes before.
The MIT authors cautioned that this study involved a limited number of participants and, because their findings were context-dependent, no generalizations can be made. They even urge against the use of words like “stupid,” “dumb,” “brain rot” and “harm” as not a fair conclusion to be drawn from their work.
However, as one research scientist at MIT’s Media Lab, Nataliya Kosmyna, observed: “Our brains love shortcuts; it’s in our nature. But your brain needs friction to learn. It needs to have a challenge. Simply put, we’re very good at cognitive offloading.” For example, you use GPS to find your way on a car trip, right?
Teachers and professors are concerned that students’ use of LLMs weakens important academic skills such as conducting research, writing, reading comprehension and critical thinking. In a recent survey, about half of students report concerns about their school’s uses of LLMs such as feeling less connected to their teachers and being exposed to extreme views they otherwise wouldn’t encounter.
Teachers and professors have adopted three differing approaches to their students’ use of LLMs and, more generally, any artificial intelligence: prohibition, permission within guidelines and encouragement to use LLMs. Considering the benefits and risks of using AI, which approach is best, and why?
Prohibition: This is the most difficult to enforce, and requires the teacher or professor to be aware when a student is using an LLM. To avoid such use, professors motivate their students to eschew LLMs in completing writing assignments by emphasizing the importance of students expressing ideas in their own voice, not just regurgitating what they find online.
To make prohibition work, teachers and professors teach the process of writing rather than just assigning papers: requiring students to draft with pen and paper in the classroom, showing all edits as they hone their work and grading students in part based on classroom performance, while banning phones and laptops from the classroom.
Permitted within guidelines: Acknowledging students’ perceived need to use LLMs, some teachers and professors allow their use, but only as a starting point. This requires students to scan and turn in their mark-ups from their first drafts and meetings with students individually to discuss their thought and writing processon the way they proceeded from the spoon-fed LLM draft to their own writing.
Encouragement: Some teachers and professors, acknowledging that use of LLMs simply can’t be prohibited or permitted only with guidelines, take the position that they must prepare their students for the future – and the future is a world in which we will all use LLMs for a variety of purposes. Thus, the notion is to prepare students for the proper and appropriate use of LLMs. If a student merely copies and pastes something from an LLM, the teacher or professor will know it by the anodyne language and analysis and by the use of AI-detecting software.
According to a recent poll, between May and December 2025, the percentage of high school and college students using AI for homework increased from 48 to 62 percent, even though about 66 percent said technology harmed critical-thinking. This increase might reflect the fact that AI startups appeal directly to students via social media videos featuring influencers who coach students how to use generative AI to write school essays.
If you were teaching today, which approach would you adopt?
Stewart Edelstein, a Stockbridge resident, is author of several books, including “Dubious Doublets: A Delightful Compendium of Unlikely Word Pairs of Common Origin, from Aardvark/Porcelain to Zodiac/Whiskey.”