Is AI Making Us Dumber? The Deskilling Problem Nobody's Talking About

There’s a strange thing happening with AI that almost nobody is talking about.

People are getting faster at their jobs. They’re producing more. They’re handling tasks that used to take hours in minutes. By every productivity metric, AI is working exactly as promised.

But underneath those numbers, something else is happening. People are quietly losing the skills they built over years. Doctors are getting worse at diagnosing without AI assistance. Law students are making more critical errors. Office workers are forgetting how to write clearly on their own.

It’s called AI deskilling, and it may be the most important unintended consequence of the AI revolution.

What AI Deskilling Actually Means

Deskilling is simple: when you stop practicing a skill, you lose it. AI deskilling is when AI tools take over cognitive tasks so completely that the human behind the screen gradually loses the ability to do those tasks independently.

This isn’t a new phenomenon. We’ve seen it before with other technologies.

Calculators and math. When calculators became common, mental math skills declined. Most adults today can’t do long division by hand, a skill that was standard two generations ago.

GPS and navigation. Research shows that habitual GPS use weakens spatial memory. People who always follow turn-by-turn directions lose their natural sense of direction over time. Many people now can’t navigate to places they’ve driven to dozens of times without GPS.

Spell-check and spelling. Automatic spell-check has made most people worse at spelling. The skill atrophied because we stopped needing it.

AI is the same pattern on a much bigger scale. Instead of outsourcing just math or navigation, we’re outsourcing writing, analysis, research, decision-making, coding, and critical thinking. The breadth of cognitive skills being offloaded to AI is unprecedented.

The Science: What Researchers Are Finding

This isn’t speculation. Peer-reviewed studies from 2025 and 2026 are documenting real deskilling effects across multiple professions.

Doctors Are Getting Worse Without AI

A 2025 study published in The Lancet Gastroenterology & Hepatology tracked doctors who used AI-assisted tools during colonoscopies. The AI helped them spot precancerous lesions called adenomas. When the AI was turned off, their detection rate dropped from 28.4% to 22.4%.

That’s a significant decline. These doctors had become dependent on the AI’s visual assistance, and their own observation skills had weakened. The researchers called this the first documented case of clinical AI deskilling.

A broader review in Artificial Intelligence Review confirmed the pattern: AI-driven decision support in medicine is eroding diagnostic expertise and reducing opportunities for doctors to develop and maintain their skills through practice.

Law Students Make More Mistakes With AI

Professors at the University of Illinois College of Law studied how generative AI affected legal education. Students who used AI chatbots for their work were more prone to critical reasoning errors than those who worked without AI.

The conclusion was blunt: without proper checks and balances, AI tools could lead to “widespread deskilling, particularly among younger and less-experienced attorneys.” The students who leaned on AI the most developed the weakest independent legal reasoning skills.

Your Brain Literally Changes

MIT’s Media Lab ran a study in 2025 where they monitored students using EEG headsets while they wrote essays. Some used ChatGPT. Others used search engines. Others wrote without any digital help.

The students who used ChatGPT showed 47% weaker brain connectivity compared to those who wrote on their own. Their brains were less actively engaged in the cognitive processes that build writing skill, critical thinking, and original thought.

A separate study from the University of Toronto found that college students today show a 42% decrease in divergent thinking scores compared to students just five years ago. Divergent thinking is the ability to come up with creative, original ideas. It’s one of the most important cognitive skills for innovation and problem-solving.

Office Workers Are Ceding Their Expertise

Microsoft Research, in collaboration with Carnegie Mellon, surveyed knowledge workers about their AI use. The findings were telling: workers reported that AI made tasks feel cognitively easier. But the researchers found something the workers didn’t notice themselves.

They were ceding problem-solving expertise to the system. Instead of thinking through challenges and building judgment, they were focusing on the functional work of gathering and integrating AI responses. The thinking itself was being outsourced.

Why This Happens: The Cognitive Offloading Trap

There’s a well-studied psychological concept called cognitive offloading. It means using external tools to reduce the mental effort a task requires.

Writing a shopping list is cognitive offloading. You don’t need to remember everything because the list remembers for you. Using a calculator is cognitive offloading. You don’t need to do the math because the tool does it.

Cognitive offloading isn’t inherently bad. It frees up mental energy for higher-level thinking. The problem starts when the offloading becomes so complete that you never exercise the underlying skill at all.

Think of it like physical fitness. If you use an elevator every single day, your legs don’t just stay the same. They weaken. Walking up stairs isn’t just about getting to the next floor. It’s what keeps your legs strong enough to walk up stairs.

The same principle applies to your brain. Writing isn’t just about producing text. It’s what keeps you good at organizing thoughts, building arguments, and communicating clearly. When AI does the writing for you every time, those cognitive muscles atrophy.

Psychology Today published research in February 2026 confirming that cognitive offloading through AI significantly reduces new skill formation. In programming skill assessments, the AI-assisted group scored 17% lower than the group that learned without AI help.

The Numbers: How Dependent Are We Already?

The scale of AI dependency in 2026 is staggering:

• 58% of employees use AI at work regularly
• 22% use ChatGPT daily for work tasks
• 95% of developers use AI to generate or fix code
• 86% of engineers use AI tools in their work
• 62% of workers rely on AI for everyday writing and summarization
• 27% of white-collar workers use AI often at work, up 12 points from the previous year

And yet, only 6% of engineers fully trust AI output. People are using tools they don’t fully trust to do work they’re losing the ability to do themselves. That’s not a recipe for long-term competence.

The Deskilling Paradox

Here’s what makes this tricky. The Communications of the ACM called it the “AI Deskilling Paradox”:

AI makes you more productive in the short term while making you less capable in the long term.

A junior developer who uses AI to write all their code ships features faster today. But a year from now, they understand their codebase less deeply, debug less effectively, and struggle more when the AI gives them bad suggestions (which it regularly does).

A marketing manager who uses AI to write all their copy produces more content today. But their personal writing voice atrophies. Their ability to craft a genuinely original angle weakens. They become an editor of AI output rather than a creator of original work.

The paradox is that the productivity gains are visible and measurable. The skill losses are invisible until you suddenly need the skill and realize it’s gone.

Who’s Most at Risk?

Early-Career Workers

People who start their careers with AI from day one are at the highest risk. They may never develop foundational skills that previous generations built through years of practice. A junior lawyer who has always had AI to research case law may never develop the deep legal reasoning that comes from doing it manually hundreds of times.

Knowledge Workers in Routine Tasks

If your job involves writing, analysis, research, or data processing, and you use AI for the majority of those tasks, your core skills in those areas will weaken over time. The work gets done, but you’re becoming less capable of doing it yourself.

Students

Students face a unique version of this problem. The purpose of education isn’t just to produce outputs (essays, solutions, projects). It’s to build cognitive capacity through the struggle of producing those outputs. When AI eliminates the struggle, it also eliminates the learning. Students who use AI at work later will be at a disadvantage if they never built the foundational skills.

Anyone Who Stops Doing Hard Things

The pattern cuts across every profession. Radiologists who let AI flag every anomaly. Writers who let AI produce every first draft. Analysts who let AI build every model. The common thread: if you outsource the hard part of your job to AI, the hard part of your job becomes something you can no longer do.

How to Use AI Without Losing Your Edge

The solution isn’t to stop using AI. That would be like refusing to use a calculator because you want to stay good at mental math. AI is genuinely useful, and ignoring it puts you at a competitive disadvantage.

The solution is to use AI deliberately, with guardrails that protect your core skills.

1. Do Your Own Thinking First

Before you ask AI anything, spend 5 to 10 minutes forming your own thoughts. Write a rough outline before asking AI to help with a document. Sketch your approach to a problem before asking AI to solve it. Draft your email before asking AI to polish it.

This one habit changes everything. It ensures you’re using AI to enhance your thinking rather than replace it. The quality of your prompts will also improve dramatically because you’ll have clearer context to provide.

2. Practice Core Skills Without AI

Set aside regular time for AI-free work. Write by hand or without AI assistance at least once a week. Do analysis from scratch periodically. Navigate without GPS occasionally.

Think of it as cognitive exercise. You don’t need to do it all the time, but you need to do it enough to maintain the skill.

3. Use AI as a Second Opinion, Not a First Draft

Instead of asking AI to produce something and then editing it, produce something yourself and then ask AI to review it. The difference is huge. In the first scenario, AI does the creative work and you do quality control. In the second, you do the creative work and AI helps you improve.

4. Stay Skeptical of AI Output

One of the best protections against deskilling is healthy skepticism. If you treat AI output as potentially unreliable (which it often is), you’re forced to engage your own judgment to evaluate it. That engagement itself is a form of cognitive exercise.

5. Focus on Skills AI Can’t Replace

Double down on developing skills that are uniquely human and resistant to automation: complex judgment, creative vision, interpersonal communication, ethical reasoning, and the ability to work with ambiguity. These skills become more valuable, not less, as AI handles routine cognitive tasks.

For a deeper look at which jobs are most and least at risk, we’ve covered this separately.

6. Teach and Explain What You Know

One of the best ways to keep a skill sharp is to teach it to someone else. If you understand how machine learning works, explain it. If you know how to write a compelling argument, mentor someone. Teaching forces you to articulate and reinforce your own knowledge.

What Companies Should Be Doing

This isn’t just an individual problem. Organizations bear responsibility too.

Training should include AI-free exercises. Companies that train employees exclusively with AI tools are building a workforce that can’t function without them. Regular exercises where teams work through problems without AI assistance build resilience.

Job roles should preserve human judgment. Roles that exist purely to supervise AI output without requiring independent expertise are a dead end. The best organizational designs keep humans doing meaningful cognitive work, with AI augmenting rather than replacing their thinking.

Assessment should test actual capability. If you’re evaluating employees, test their independent ability, not just their ability to manage AI output. The distinction matters.

The Bottom Line

AI is one of the most powerful tools ever created. It genuinely makes people more productive and capable in the short term. But the research is clear: over-reliance on AI erodes the skills, judgment, and cognitive abilities that make you good at your job in the first place.

The people who will thrive long-term aren’t the ones who use AI the most or the least. They’re the ones who use it strategically: staying in control of their thinking, practicing their core skills, and treating AI as an assistant rather than a replacement for their own brain.

Use AI. But don’t let it use you.

Sources and Further Reading

This article draws on peer-reviewed research, institutional reports, and verified industry data:

• Workers Gain Hours With AI But Risk Losing Skills — Microsoft’s Future of Work report on AI deskilling risks (AllWork, 2026)
• AI Use May Be Deskilling Doctors — Lancet study on declining doctor performance without AI (STAT News, 2025)
• The AI Deskilling Paradox — Analysis of the productivity vs. capability tradeoff (Communications of the ACM, 2026)
• Is AI Dulling Our Minds? — Harvard Gazette investigation into AI and cognitive decline (Harvard, 2025)
• Your Brain on ChatGPT — MIT Media Lab EEG study on brain activity during AI-assisted writing (MIT, 2025)
• Cognitive Offloading: Using AI Reduces New Skill Formation — Research on AI’s impact on learning and skill development (Psychology Today, 2026)
• AI Is Deskilling You. Here’s How to Prevent It — Wharton professor Kartik Hosanagar’s practical prevention strategies
• AI-Induced Deskilling in Medicine — Comprehensive review of deskilling evidence in healthcare (Artificial Intelligence Review, Springer)
• AI in the Workplace Statistics 2026 — Current AI adoption and dependency data across industries (Azumo)
• Frontiers: Deskilling Dilemma: Brain Over Automation — Peer-reviewed analysis of cognitive risks from AI dependency (Frontiers in Medicine, 2026)