Is AI Making Us Worse Thinkers? What the Research Actually Shows

Every semester, faculty in every discipline ask me some version of the same question: how do we keep students doing the thinking when AI can produce the answer?

Behind that question is a fear. Not that students will cheat, but that they will stop being able to think without help. Ninety percent of faculty in a 2026 national survey by AAC&U and Elon University said they expect generative AI to weaken students' critical thinking. Ninety-five percent worried students would become overreliant on it.

Fear is not evidence. So let's look at the evidence, which is newer, stranger and more interesting than the headlines.

Study One: Your Brain on ChatGPT

In June 2025 a team at the MIT Media Lab led by Nataliya Kosmyna published a study with a title built for the internet: "Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Task."

Fifty-four participants wrote essays over several sessions in one of three conditions: using a large language model, using a search engine or using nothing but their own minds. Researchers recorded their brain activity with EEG.

The results lined up in a tidy staircase. The brain-only group showed the strongest and most widely distributed neural connectivity. The search engine group was in the middle. The ChatGPT group showed the weakest. The more the tool did, the less the brain engaged.

Two behavioral findings were even more telling. ChatGPT users struggled to accurately quote essays they had just written, and they reported the lowest sense of ownership over their own work. Their essays also tended to sound alike.

The study has real limits. It is small, it was released as a preprint, and EEG connectivity is not the same thing as intelligence. But it gave a name to something many teachers had sensed: cognitive debt. When you let a tool do the thinking now, you may pay for it later in what you can remember, explain and own.

Study Two: Confidence in the Machine Lowers Our Guard

The second study looked at adults at work. In 2025 researchers at Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers who use generative AI at least weekly and analyzed nearly a thousand real examples of their AI-assisted tasks. The paper was presented at CHI 2025, the leading conference on human and computer interaction.

The central finding was elegant and a little unnerving. The more confident workers were in the AI, the less critical thinking they applied to its output. The more confident they were in their own expertise, the more they scrutinized what the AI produced.

The researchers also observed a shift in what critical thinking looks like. Workers spent less effort generating ideas and more effort verifying, integrating and overseeing AI output. Critical thinking did not disappear. It changed jobs, from author to editor.

That is good news only if people actually do the editing. When we trust the tool too much, we skip that step.

Study Three: Cognitive Offloading

The third study, by Michael Gerlich of SBS Swiss Business School, was published in the journal Societies in 2025. He surveyed 666 participants and interviewed fifty more about their AI use and critical thinking.

Gerlich found a strong negative correlation between frequent AI use and critical thinking scores, and the link ran through what psychologists call cognitive offloading: handing mental work to an external tool. Younger participants, ages seventeen to twenty-five, used AI the most, offloaded the most and scored lowest. Participants over forty-six relied on AI less and scored higher. Higher education was associated with stronger critical thinking regardless of AI use.

Again, honesty about limits: this is correlational. People with weaker critical thinking may simply be more likely to lean on AI. But taken together with the other studies, the direction of concern is consistent.

What the Research Does Not Say

It is important not to overclaim. None of these studies shows that AI makes people permanently less intelligent. And offloading is not new or always bad. We offload arithmetic to calculators, directions to maps and memory to calendars, and civilization survived.

The philosophers Andy Clark and David Chalmers argued in their 1998 paper "The Extended Mind" that tools can become genuine parts of our thinking. A notebook can function like memory. On that view, AI could extend the mind rather than shrink it.

The question is not whether we offload. It is what we offload and whether we keep the capacity to do it ourselves when it matters. A pilot uses autopilot. A good pilot can still fly the plane.

The Philosophical Point: Thinking Is a Practice, Not a Possession

Aristotle taught that virtues are formed by practice. We become just by doing just acts and brave by doing brave acts. The same is true of intellectual virtues. We become careful thinkers by thinking carefully, repeatedly, often with difficulty.

That is why the struggle matters. The awkward first draft, the argument that falls apart, the hour spent wrestling with a hard text: that friction is not a bug in learning. It is the learning. Psychologists Elizabeth and Robert Bjork at UCLA call these "desirable difficulties," challenges that slow performance in the moment but strengthen long-term learning.

Generative AI is extraordinarily good at removing friction. That is exactly what makes it dangerous for formation, and exactly why we have to be intentional about when to let it help.

What Teachers and Leaders Can Do

• Brain first, then bot. Ask students and employees to draft their own thinking before consulting AI. Interestingly, in the MIT study, participants who wrote on their own first and used the AI later showed stronger engagement than those who started with the tool.
• Make people defend the output. If you used AI, explain why its answer is right, where it might be wrong and what you changed. Verification is a critical thinking skill worth grading.
• Protect the hard parts. Decide which skills must stay human in your course or team, such as interpretation, argument and ethical judgment, and design practice for them without AI.
• Build calibrated trust. The Microsoft and Carnegie Mellon finding suggests the goal is not distrust but calibrated trust. Teach people to know their own expertise well enough to catch the machine's mistakes.
• Teach the humanities on purpose. Philosophy, history and literature are training grounds for exactly the skills AI cannot supply: weighing reasons, interpreting meaning and judging what matters.

The Bottom Line

Does AI hurt critical thinking? The best current evidence says it can, when we use it to skip thinking rather than to sharpen it. The tool is not the enemy. Passivity is.

The good news is that the remedy is old and proven. Practice the hard thing. Explain your reasoning. Keep your judgment in the loop. AI can make us faster. Only we can make ourselves wiser.

Sources and Further Reading

• Nataliya Kosmyna et al., "Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Task," MIT Media Lab (2025). MIT Media Lab.
• Hao-Ping (Hank) Lee et al., "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers," CHI 2025 (Microsoft Research and Carnegie Mellon University).
• Michael Gerlich, "AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking," Societies 15, no. 1 (2025).
• AAC&U and Elon University Imagining the Digital Future Center, faculty survey on generative AI (2026).
• Andy Clark and David Chalmers, "The Extended Mind," Analysis 58, no. 1 (1998).
• Elizabeth L. Bjork and Robert A. Bjork, "Making Things Hard on Yourself, But in a Good Way: Creating Desirable Difficulties to Enhance Learning" (2011).
• Aristotle, Nicomachean Ethics, Book II.

Related: AI Cheating in College Is a Symptom. Our Assessments Are the Disease. · Book a faculty or leadership workshop on thinking in the age of AI