In early 2025, a study published in the journal Societies reignited a debate many had sensed intuitively but struggled to prove: the more people rely on artificial intelligence to think for them, the less they seem to think for themselves. Led by researcher Michael Gerlich, the study did not frame AI as an enemy or a dystopian force. Instead, it highlighted something far more subtle and unsettling—a slow cognitive drift caused not by AI’s power, but by our comfort with surrendering effort.
The findings were clear and uncomfortable. Individuals who frequently delegated tasks such as reasoning, evaluation, synthesis, and decision-making to AI tools scored significantly lower on validated critical thinking assessments than those who used AI sparingly. The mechanism behind this decline was not mysterious. It was something psychologists have understood for decades under a different name: cognitive offloading.
Cognitive offloading occurs when humans externalize mental work to tools. We write things down so we don’t have to remember them. We use calculators so we don’t have to compute. We rely on GPS so we don’t have to navigate. AI, however, represents a fundamental shift. It does not just store information or execute calculations—it appears to reason. And that appearance changes how deeply we engage.
What makes Gerlich’s findings particularly concerning is not just the decline itself, but who is most affected. Younger users and habitual AI dependents showed the strongest reductions in critical thinking scores. These were not people incapable of reasoning; they were people who simply skipped the process. Faced with a problem, they deferred judgment rather than exercising it. Over time, that deferral hardened into habit.
This is not a story about laziness. It is a story about efficiency reshaping cognition.
The Seduction of Effortless Intelligence
AI tools excel at presenting answers that sound complete. They structure arguments, summarize complexity, and generate fluent reasoning at a speed no human can match. For users under time pressure—students, professionals, creators—this feels like liberation. Why struggle through ambiguity when a confident response appears instantly?
The danger is not that AI is wrong all the time. The danger is that it is right often enough to earn trust without scrutiny.
In Gerlich’s survey, participants who expressed high confidence in AI outputs were significantly less likely to verify results, challenge assumptions, or refine conclusions. Trust short-circuited skepticism. Once that loop was established, effort dropped sharply. Critical engagement became optional rather than essential.
A parallel 2025 study conducted by researchers from Microsoft and Carnegie Mellon University reinforced this pattern among knowledge workers, including software developers, analysts, and social workers. Participants reported that when they trusted AI tools, they applied “little to no critical thinking” to the outputs. The tool became not an assistant, but an authority.
This distinction matters.
Tools that perform tasks still require oversight. Tools that appear to think invite abdication.
Cognitive Atrophy: When Skills Fade from Disuse
Human cognition follows a simple rule: what you don’t use, you lose.
Critical thinking is not a static trait. It is a practiced skill involving evaluation, comparison, doubt, and synthesis. When those processes are repeatedly bypassed, neural efficiency declines—not because the brain is incapable, but because it adapts to reduced demand.
Gerlich’s study likened this to muscle atrophy. A limb immobilized for months does not forget how to move—it weakens from disuse. The same applies to reasoning. If AI consistently performs the evaluative steps, the human mind reallocates energy elsewhere.
The problem is where that energy goes.
Rather than being reinvested in higher-level thinking, many users reported channeling freed cognitive capacity into passive consumption—scrolling, reacting, or multitasking—rather than deeper analysis or creativity. Efficiency did not translate into intellectual growth. It translated into mental shallowness.
This is where AI diverges sharply from earlier technologies.
Why AI Is Different from Calculators and GPS
Critics of these studies often point to historical precedent. Calculators reduced mental arithmetic. GPS weakened spatial navigation. Spellcheck eroded spelling skills. Yet society adapted. Why should AI be any different?
Because AI does not just replace a subskill. It replaces judgment itself.
A calculator gives a number. A GPS gives directions. Neither explains why. AI, on the other hand, delivers reasoned narratives. It frames premises, draws conclusions, and often anticipates objections. This creates the illusion of thinking having already been done.
The risk is not that humans will forget facts. It is that they will stop practicing evaluation.
Critical thinking is not about knowing answers. It is about knowing how to question answers.
When AI outputs arrive fully packaged—polished, persuasive, and contextually tailored—users are less inclined to ask whether assumptions are valid, sources are biased, or conclusions are incomplete. The smoother the output, the less friction there is to trigger doubt.
Friction is where thinking lives.
Younger Minds, Greater Vulnerability
One of the most troubling aspects of Gerlich’s findings is the disproportionate impact on younger users. This is not because younger people are less intelligent, but because their cognitive habits are still forming.
Critical thinking develops through repeated exposure to uncertainty, error, and correction. When AI smooths those experiences—offering immediate coherence—it deprives developing minds of struggle. Over time, reasoning becomes something accessed rather than exercised.
This creates a generational asymmetry. Older users often learned to think deeply before delegating tasks to machines. Younger users may learn to delegate before they ever master the underlying skill.
The long-term implications are not fully known, but the trajectory is concerning. A society where fluency replaces understanding is a society vulnerable to manipulation—by algorithms, by narratives, and by those who control the tools.
Trust, Authority, and the Illusion of Neutrality
Another critical factor is perceived objectivity. Many users treat AI outputs as neutral, forgetting that algorithms are trained on human-generated data embedded with bias, omission, and cultural assumptions.
When trust in AI rises, skepticism falls—not only toward the output, but toward one’s own judgment. Users report deferring even when answers feel intuitively wrong. The machine’s confidence overrides internal doubt.
This dynamic mirrors historical patterns of authority. Humans have always deferred to institutions perceived as knowledgeable: priests, doctors, textbooks, experts. AI inherits that authority—but without accountability, transparency, or lived experience.
The tragedy is that AI’s errors are often subtle. They are not absurd enough to trigger rejection. They are plausible enough to pass unchallenged.
Critical thinking is not only about catching mistakes. It is about recognizing when certainty itself should be questioned.
Efficiency Without Wisdom
There is no denying AI’s benefits. It accelerates research, automates routine tasks, and lowers barriers to entry in complex fields. For professionals under pressure, it can be transformative.
But efficiency is not wisdom.
The Microsoft–Carnegie Mellon study revealed that high AI confidence correlated strongly with reduced effort in refining outputs. People stopped asking, “Is this the best answer?” and started asking, “Is this good enough?”
Over time, “good enough” becomes the standard.
This shift does not merely affect individual cognition. It reshapes organizational culture. Decisions become faster but thinner. Outputs become uniform. Creativity narrows. Dissent decreases.
When everyone uses the same tools trained on the same data, originality suffers—not because AI suppresses creativity directly, but because humans stop pushing beyond what is given.
A Historical Echo with Higher Stakes
The concern around AI mirrors past technological anxieties, but with amplified consequences. Writing once threatened memory. Printing threatened oral tradition. Computers threatened mental calculation.
Each time, humanity adapted—but not without loss.
What makes AI uniquely risky is its scope. It touches language, reasoning, creativity, decision-making, and social interaction simultaneously. It does not replace one cognitive function; it reshapes the environment in which all functions operate.
And unlike previous tools, AI improves rapidly, making dependence more tempting with each iteration.
The danger is not sudden collapse. It is gradual erosion.
The Balance Experts Are Urging
Neither Gerlich nor the Microsoft researchers argue for rejecting AI. The message is about how it is used.
Experts emphasize that AI should function as a cognitive mirror, not a replacement. Used correctly, it can provoke deeper thinking—by challenging assumptions, offering alternatives, and inviting critique.
This requires intentional design and disciplined use. Prompts that ask “why,” “what’s missing,” or “argue against this” can re-engage reasoning. Verification must be habitual, not optional. Reflection must follow output.
AI should slow thinking down at key moments, not speed it up indiscriminately.
The difference between augmentation and atrophy lies in whether humans remain active participants in the reasoning process.
The Quiet Choice Being Made
Every time someone accepts an AI-generated answer without scrutiny, a small choice is made. Not a dramatic one. Not a moral failure. Just a preference for ease over engagement.
But repeated thousands of times across millions of people, those small choices accumulate.
The future shaped by AI will not be decided by the technology itself, but by whether humans continue to value the discomfort of thinking. Critical thinking is not efficient. It is slow, effortful, and often frustrating.
That friction is not a flaw. It is the cost of understanding.
The studies emerging in 2025 do not signal the end of human cognition. They signal a crossroads. AI can either become a scaffold that strengthens thinking—or a crutch that weakens it.
The outcome depends less on algorithms than on habits.
And habits, once formed, are far harder to update than software.