When MIT research scientist Nataliya Kosmyna began reviewing intern applications recently, she noticed a disturbing trend. The cover letters were identical in their “soulless” perfection, she said.
Long, polished, and oddly detached, they bore the unmistakable fingerprints of large language models (LLMs) like ChatGPT.
But it wasn’t just the writing that was changing.
In her classroom, Kosmyna noticed students were forgetting information faster than ever. “Cognitive offloading,” the outsourcing of mental tasks to algorithms, is now the subject of research.
Experts warn that by letting AI do the thinking, we may be eroding our creativity, our memory, and potentially our long-term brain health.
In a recent study at the MIT Media Lab, Kosmyna monitored the brainwaves of students tasked with writing essays. The participants were split into three groups: those using ChatGPT, those using basic search engines, and those not using technology.
The results were startling. While the brains of those working unaided were “on fire” with widespread activity, the ChatGPT group saw a staggering 55% drop in brain activation.
“The brain didn’t fall asleep, but there was much less activation in areas corresponding to creativity and information processing,” Kosmyna told the BBC. Furthermore, the AI users struggled to remember what they had “written” and felt no ownership over their work, she said.
The phenomenon isn’t entirely new. The Google Effect long ago proved that we are less likely to remember facts we know can be easily found online. However, LLMs represent a more profound shift: “cognitive surrender.”
A University of Pennsylvania study suggests users are increasingly prone to accepting AI-generated answers without scrutiny, even when they contradict their own intuition. This dependency is appearing in high-stakes fields; one study found that medical professionals who relied on AI to spot colon cancer tumors saw their own diagnostic skills wither when the tool was removed.
Vivienne Ming, a computational neuroscientist and author of Robot Proof, warns that this lack of mental effort has physical consequences. Her research found that students using AI for problem-solving showed minimal “gamma wave” activity — a key marker of cognitive effort.
“Deep thinking is our superpower,” Ming explained to the BBC. “Weak gamma wave activity has been linked to cognitive decline later in life. If we don’t use it, the long-term implications are pretty strong.”
The danger isn’t the technology itself, but how we use it. A small subset of users (less than 10%) treats AI as a data-gathering tool rather than an answer machine. These individuals show higher brain activation and produce more accurate results.
To combat mental atrophy, Ming suggests seeking out “productive friction.” Instead of asking ChatGPT for an answer, she recommends the “nemesis prompt”: telling the AI to act as a lifelong enemy that must find every flaw in your argument. This forces humans to defend and refine their own ideas.
As AI becomes more integrated into daily life, the message from researchers is clear: the brain loves a shortcut, but for long-term health, we must choose the hard way. To keep our minds sharp, we must ensure that while the models get smarter, we don’t get duller.
