As Large Language Models (LLMs) and artificial intelligence services increasingly take over our cognitive burdens, researchers are sounding the alarm: the convenience of what is being called “mental outsourcing” comes with a heavy price tag.
Nataliya Kosmyna, a researcher specializing in human-computer interaction at the Massachusetts Institute of Technology (MIT), first became suspicious while reviewing internship applications. She noticed a disturbing trend—many of the letters were eerily identical. Each was long and polished, following a repetitive structure that transitioned abruptly into illogical justifications for why the applicant was a perfect fit for the role. Kosmyna concluded that these candidates were relying on LLMs like ChatGPT, Google Gemini, and Claude to draft their application materials.
Her concerns extended beyond recruitment. During her time at MIT, Kosmyna observed that students appeared to struggle with retaining recently taught material more than in previous years. She and other experts fear that over-reliance on AI is eroding fundamental cognitive skills, including critical thinking, problem-solving, and attention span. Growing evidence suggests that this “cognitive offloading” is more than just a behavioral shift; it is actively altering how our brains process information and potentially contributing to a decline in mental function.
History offers a precedent for this concern. The rise of the internet, for instance, turned deep research into a simple act of typing a keyword into a search bar. As dependence on search engines grew, studies confirmed a decline in our ability to recall specific details—a phenomenon dubbed the “Google Effect.” While some argue that the internet serves as an external memory system that frees up the human brain for other tasks, the shift toward delegating complex thought to AI is far more invasive. Memory retention and the ability to navigate complex challenges may be plummeting as a result.
While AI can effortlessly compose poetry, offer financial advice, and provide companionship, students are increasingly surrendering their academic work to these tools. This is particularly dangerous for younger generations, whose developing cognitive skills are most vulnerable to the negative impacts of AI dependency. To better understand this phenomenon, Kosmyna and her colleagues at the MIT Media Lab conducted a study with 54 students tasked with writing short essays.
The participants were divided into three groups: one using ChatGPT, one using Google search (with AI summaries disabled), and a control group that used no technology. Using brain-scanning equipment, the researchers measured neural activity as the students worked. Even though the essay prompts—covering topics like happiness, loyalty, and daily decisions—required minimal research, the findings were stark. The group working without technology showed high levels of widespread brain activity, while the Google group showed intense visual cortex engagement. In contrast, those using ChatGPT exhibited up to a 55% decrease in brain activity.
“Their brains weren’t asleep, but there was a distinct drop in activity related to creativity and information processing,” Kosmyna explained. Furthermore, the AI-reliant group struggled to recall their own work, demonstrating a lack of ownership and retention compared to those who wrote independently.
While this study undergoes peer review, its findings align with existing research. Experts at the University of Pennsylvania have identified a state of “cognitive surrender” where users accept AI output with minimal scrutiny, often overriding their own intuition. This issue extends to high-stakes fields like medicine; a multinational team found that medical professionals who used AI for colon cancer screening for three months were less capable of detecting tumors when forced to work without the technology.
The loss of originality is also becoming apparent. Professors described the AI-generated essays as “soulless,” lacking the depth and unique perspective of human thought. In follow-up tests conducted four months later, the former ChatGPT users showed lower neural connectivity even when working without AI, suggesting that their previous reliance on the tool hindered their engagement with the subject matter.
Computational neuroscientist Vivienne Ming notes that while LLMs could be powerful tools for stimulating thought, most users are using them as a substitute for mental effort. In her own research at the University of Berkeley, students tasked with predicting economic trends—like oil prices—simply copied answers from AI, showing extremely low gamma wave activity in their brains, a key indicator of cognitive effort. Ming warns that a lack of deep thinking now could have long-term implications for cognitive health, as weak gamma activity has been linked to future cognitive decline.
“Deep thinking is our superpower,” Ming emphasized. “If we don’t exercise it, the health implications are severe.” She highlights that less than 10% of participants used AI effectively—by gathering data to analyze it themselves—and these individuals showed both higher brain activity and more accurate predictions.
Mirroring concerns about GPS-induced declines in spatial memory, which have been linked to increased risks of Alzheimer’s disease, the over-reliance on LLMs could potentially accelerate cognitive decline. The proposed solution is a move toward “hybrid intelligence,” where humans perform the initial thinking and use AI only for verification and testing. By employing techniques like the “nemesis instruction”—asking AI to critique and improve our own ideas—or prioritizing “productive friction” where AI provides only questions rather than answers, we can safeguard our cognitive future.
Ultimately, while our brains are wired to crave the path of least resistance, we must remain vigilant. To ensure long-term brain health, we must continue to challenge ourselves to think, analyze, and create, rather than allowing our most valuable cognitive functions to atrophy.
Summary
Researchers are increasingly concerned that the over-reliance on Large Language Models for tasks like writing and problem-solving is leading to “cognitive offloading,” which may erode essential mental functions. Studies, including recent brain-scanning research at MIT, indicate that individuals using AI for academic tasks exhibit a significant decrease in neural activity related to creativity and information processing. This trend suggests that outsourcing complex thought to AI can result in reduced retention, lower engagement, and a decline in critical thinking skills.
Experts warn that this habit of “cognitive surrender” mirrors past dependencies on technology and could have long-term consequences for brain health, potentially accelerating cognitive decline. To mitigate these risks, researchers advocate for “hybrid intelligence,” where humans retain primary control over thinking processes and use AI only as a supplementary tool for critique or verification. Ultimately, maintaining cognitive health requires intentionally exercising the brain through deep, independent analysis rather than opting for the path of least resistance.