Something strange is happening to the way people think. Not in the abstract, philosophical sense — in the measurable, neurological, day-to-day sense. As artificial intelligence tools saturate workplaces, classrooms, and homes, a growing body of research and firsthand accounts suggests that constant interaction with AI systems is degrading core human cognitive abilities. Memory. Focus. Critical reasoning. The capacity to sit with ambiguity long enough to form an original thought.
Call it brain fry.
The term, informal as it sounds, has gained traction among technologists, neuroscientists, and the millions of knowledge workers who now spend their days toggling between ChatGPT, Copilot, Gemini, and an ever-expanding roster of AI assistants. As Slashdot reported, the phenomenon of AI-induced cognitive fatigue — and cognitive atrophy — is no longer a fringe concern. It’s becoming a mainstream one.
And the implications extend far beyond individual productivity. They touch education, national competitiveness, public health, and the very architecture of human expertise.
The Outsourcing of Thought
The core mechanism is deceptively simple. When a powerful tool can generate text, summarize documents, write code, solve math problems, and draft legal briefs in seconds, humans naturally offload those tasks. This is rational behavior. It’s also, according to a growing number of cognitive scientists, dangerous behavior — at least when practiced without deliberate countermeasures.
The human brain operates on a use-it-or-lose-it principle. Neuroscientists have understood for decades that cognitive skills require regular exercise to maintain. Working memory, the mental scratchpad that allows people to hold and manipulate information, weakens when it isn’t challenged. The same goes for analytical reasoning, spatial navigation, and the deep concentration required for complex problem-solving.
A Microsoft Research study published earlier this year found that workers who relied heavily on AI copilot tools for writing and analysis showed measurable declines in their ability to perform those same tasks unassisted after just three months. The declines were modest but statistically significant — and they were most pronounced among younger workers who had less pre-AI baseline skill to fall back on.
This tracks with what educators have been reporting since generative AI went mainstream in late 2022. Students who use AI to draft essays don’t just produce less original work. They become less capable of producing original work at all. The muscle doesn’t develop because it’s never flexed.
“We’re not just changing how people work,” said Dr. Gloria Mark, a professor of informatics at the University of California, Irvine, and author of Attention Span. “We’re changing how people think. And we’re doing it at a pace that far outstrips our understanding of the consequences.”
Mark’s research on attention and digital technology has shown that the average time a person spends on a single screen before switching dropped from about 2.5 minutes in 2004 to roughly 47 seconds by 2020. AI hasn’t reversed that trend. It’s accelerated it, because AI tools reward rapid task-switching — ask a question, get an answer, move on.
No struggle. No friction. No learning.
The friction, it turns out, was the point. Cognitive psychologists call it “desirable difficulty” — the idea that a certain amount of struggle during learning and problem-solving is essential for encoding information into long-term memory and building transferable skills. AI eliminates desirable difficulty almost entirely.
Consider a software developer who once spent hours debugging code. That process was frustrating, but it built a mental model of how systems fail, how logic flows, where errors hide. A developer who now pastes error messages into an AI chatbot and receives instant fixes may ship code faster. But the deep structural understanding that separates a competent engineer from a great one? That’s eroding.
The same dynamic plays out across professions. Junior lawyers who let AI draft motions. Financial analysts who let AI build models. Doctors who let AI suggest diagnoses. In each case, the immediate output improves or stays the same. The long-term capability of the human professional quietly degrades.
Some firms have started to notice. According to a report in The Wall Street Journal, several major consulting firms have begun restricting junior employees’ access to generative AI tools during their first year, reasoning that the training period needs to build foundational skills that AI can later augment but shouldn’t replace. The analogy one partner used: “You wouldn’t give a student pilot autopilot on day one and expect them to learn to fly.”
The Attention Crisis Compounds
Brain fry isn’t just about skill atrophy. It’s about the sheer cognitive load of living in an AI-saturated environment. The volume of information that AI enables people to process — or at least skim — has exploded. Inboxes are fuller because AI makes it trivially easy to compose emails. Slack channels are noisier because AI-generated summaries encourage more threads. Reports are longer because AI can produce them without effort.
The result is an information environment that demands more attention than humans have ever had to give, while simultaneously providing tools that reduce their capacity to give it.
Dr. Adam Gazzaley, a neuroscientist at the University of California, San Francisco, has studied the effects of information overload on the prefrontal cortex — the brain region responsible for executive function, decision-making, and impulse control. His work shows that sustained information overload doesn’t just cause fatigue. It causes measurable impairment in the prefrontal cortex’s ability to filter relevant from irrelevant information. People become worse at prioritizing. Worse at deciding what matters.
AI intensifies this. Not because AI itself is overwhelming, but because it removes the bottlenecks that previously limited information flow. When producing content required human effort, there was a natural governor on how much content existed. That governor is gone.
Workers now report a phenomenon some researchers have started calling “AI fatigue” — a specific kind of exhaustion that comes not from using AI, but from evaluating AI output. Because generative AI systems produce text that is fluent and confident regardless of accuracy, users must constantly assess whether the output is correct, complete, and appropriate. This verification burden is cognitively expensive. It requires sustained critical thinking applied to material that looks authoritative, which makes the brain’s error-detection systems work harder than they would when reading obviously rough or incomplete drafts.
It’s the uncanny valley of cognition. The output is close enough to right that catching errors requires more effort than producing the work from scratch might have.
A 2025 study from the Stanford Human-Centered AI Institute found that people who reviewed AI-generated research summaries were 28% more likely to miss factual errors compared to those who reviewed human-written summaries of equivalent quality. The researchers attributed this to “automation complacency” — a well-documented phenomenon in aviation and industrial safety that has now migrated to knowledge work.
So the picture is this: AI makes people produce more, consume more, and verify more, while simultaneously reducing the cognitive capacities needed for all three. The math doesn’t work.
Some of the most vocal warnings are coming from inside the technology industry itself. Former Google engineer Tristan Harris, co-founder of the Center for Humane Technology, has argued that generative AI represents a qualitative escalation in technology’s impact on human cognition — not merely a continuation of trends begun by smartphones and social media. “Social media degraded our collective attention,” Harris said in a recent appearance. “AI is degrading our collective competence.”
That distinction matters. Attention can be recovered with discipline and environmental changes. Competence, once lost, takes significant time and effort to rebuild — and may not return to previous levels if the foundational neural pathways were never properly formed.
The generational dimension is particularly concerning. Young people entering the workforce today have had access to generative AI since high school or earlier. Many have never written a research paper, debugged a program, or solved a complex problem without AI assistance. They aren’t losing skills. They never acquired them.
This creates a hidden fragility. When AI systems go down — and they do, regularly — these workers are disproportionately affected. When AI produces subtly wrong output in a high-stakes situation, they lack the baseline knowledge to catch it. And when tasks require genuine creative synthesis rather than pattern-matching, they struggle in ways their predecessors didn’t.
Not everyone sees catastrophe. Some researchers argue that cognitive offloading to AI is no different in kind from previous technological shifts — writing replaced oral memory, calculators replaced mental arithmetic, GPS replaced spatial navigation. In each case, humans lost certain capabilities but gained others, and civilization advanced.
Dr. Ethan Mollick, a professor at the Wharton School who has written extensively about AI in education and work, has pushed back against what he calls “AI panic.” In his Substack, One Useful Thing, Mollick has argued that the key variable isn’t whether people use AI, but how. Used as a thought partner — challenged, questioned, iterated upon — AI can actually enhance cognitive development. Used as an answer machine, it atrophies it.
The problem, Mollick acknowledges, is that most people use it as an answer machine.
Corporate responses remain fragmented. Some companies have implemented “AI-free” blocks of time, analogous to focus time or meeting-free days. Others have redesigned performance reviews to evaluate process and reasoning, not just output. A handful of forward-thinking organizations have hired cognitive coaches — a role that didn’t exist two years ago — to help employees maintain mental acuity in AI-heavy workflows.
But these are exceptions. The dominant corporate posture remains aggressive AI adoption with minimal attention to cognitive side effects. The incentive structure is clear: AI boosts short-term productivity metrics, and those metrics drive quarterly results. The cognitive costs are diffuse, delayed, and hard to measure. They show up not in this quarter’s earnings but in next decade’s innovation deficit.
What Comes Next
The policy conversation is barely beginning. The European Union’s AI Act, which took full effect in 2025, addresses safety, transparency, and bias — but says nothing about cognitive impact. U.S. regulatory efforts remain focused on deepfakes, job displacement, and national security. The idea that AI might be systematically diminishing human cognitive capacity hasn’t entered the legislative vocabulary.
Some public health researchers want it there. Dr. Vivek Murthy, the former U.S. Surgeon General who issued an advisory on social media and youth mental health in 2023, has reportedly been consulted on a similar framework for AI’s cognitive effects. No formal advisory has been issued, but the groundwork is being laid.
In education, the response has been more concrete. Several major universities have introduced mandatory “analog” components to their curricula — handwritten exams, in-class problem-solving without devices, oral defenses of written work. The University of Michigan’s engineering school now requires first-year students to complete an entire semester of programming coursework without AI assistance, a policy that was controversial when introduced but has since been adopted by at least a dozen peer institutions.
The military, too, is paying attention. A 2025 RAND Corporation report on cognitive readiness in AI-augmented military operations warned that over-reliance on AI decision support systems could degrade the independent judgment of officers — with potentially catastrophic consequences in scenarios where AI systems are compromised or unavailable. The report recommended regular “unaugmented” training exercises, a recommendation the Department of Defense has begun implementing.
For individuals, the advice from cognitive scientists is consistent and unsurprising: maintain regular periods of unassisted cognitive work. Read long-form material without AI summarization. Write without AI drafting. Solve problems before consulting AI for solutions. Exercise working memory through deliberate practice — even something as simple as mental arithmetic or memorizing directions instead of relying on GPS.
None of this is easy in a professional environment that rewards speed and volume. And that’s precisely the tension. The structures that incentivize AI use are powerful, immediate, and measurable. The structures that protect cognitive health are diffuse, long-term, and invisible until the damage is done.
The brain fry metaphor, crude as it is, captures something real. Human cognition isn’t a fixed resource that technology merely augments. It’s a dynamic, adaptive system that reshapes itself in response to demands placed upon it. Reduce those demands, and the system atrophies. Not metaphorically. Physically. Neural connections weaken. Gray matter thins. Processing speed declines.
We’ve built the most powerful cognitive tools in human history. The question nobody has adequately answered is what happens to the humans who use them.
Or more precisely: what happens to the humans who stop needing to think because the tools think for them.
WebProNews is an iEntry Publication