Here is an interesting study from Lee and colleagues at Microsoft Research and Carnegie Mellon (2025) on AI and critical thinking. The researchers surveyed 319 knowledge workers who use tools like ChatGPT and Copilot at work every week, and they gathered 936 real examples of how those tools showed up in actual tasks. The question driving the whole thing is simple to state and hard to answer: when people use GenAI, are they still thinking critically, and what does that thinking cost them?
The answer is unsettling in a specific way.
The confidence trap in AI and critical thinking
Lee et al. found that the more workers trusted the AI, the less critical thinking they did. And the more workers trusted themselves, the more critical thinking they did. Two kinds of confidence pull in opposite directions.
The authors put it plainly: “a higher confidence in GenAI is associated with less critical thinking even though it is perceived as less effort to do so, and… a higher self-confidence is associated with more critical thinking even though it is perceived as more effort to do so” (Lee et al., 2025, n.p).
The behavior that protects your thinking feels like more work. The behavior that erodes it feels like relief. That’s a trap, and it’s built into how these tools feel to use. When the output looks polished, the instinct is to accept it. Checking feels like friction. Trusting feels like flow.
I’ve made a version of this argument before, in my post on Gerlich’s (2025) work on cognitive offloading and critical thinking: the easier a tool makes it to skip the thinking, the more likely we are to skip it. Lee et al. give that pattern a workplace face and a mechanism. It isn’t laziness. It’s confidence aimed at the machine, not the self.
Effort doesn’t vanish, it moves
Lee et al. resist the lazy headline that AI makes us think less. What they document is subtler. Critical thinking doesn’t disappear. It relocates.
For recall and comprehension tasks, effort shifts from gathering information to verifying it. For applying knowledge, it moves from solving the problem to fitting the AI’s answer into the real task. For analysis, synthesis, and evaluation, it moves from doing the work to overseeing it. They name this last move a shift from execution to “stewardship.”
As the authors describe it: “It is not that execution has disappeared altogether, nor is having high-level oversight on a task an entirely new cognitive role, but there is a shift from the former to the latter” (Lee et al., 2025, n.p).
This maps almost perfectly onto what Ranganathan and Ye (2026) argued from a different angle, that AI doesn’t reduce work so much as change its shape. The labor moves upstream, into judgment, framing, and verification. Lee et al. add the accountability piece that makes it stick. When you delegate the production, you still own the result. You steward it.
Why this should worry teachers, not just workers
The study looks at knowledge workers, but the classroom implications are hard to miss. Lee et al. report that people skip critical thinking most when a task feels unimportant, when they’re short on time, and when they lack the domain knowledge to judge whether the AI is right.
A student who can’t yet evaluate a historical argument can’t meaningfully check one an AI produces. The tool hands them a confident answer, and confidence in the tool stands in for competence they haven’t built.
This is the same worry I raised around metacognitive laziness in Fan et al.’s (2025) study, where students let the AI monitor the thinking they should be doing themselves. Lee et al. show the adult, professional version of the same slide.
The self-confidence finding points somewhere useful, though. Workers who knew their domain interrogated the output harder, not less. That tells me the protection against passive over-reliance isn’t banning the tool. It’s building the knowledge and judgment that make a person willing to question it in the first place.
References
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. _British Journal of Educational Technology, 56 (2), 489–530. https://doi.org/10.1111/bjet.13544
- Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. _Societies_, 15(1), Article 6. https://doi.org/10.3390/soc15010006
- Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25), April 26–May 1, 2025, Yokohama, Japan. ACM. https://doi.org/10.1145/3706598.3713778
- Ranganathan, A., & Ye, X. M. (2026, February 9). AI doesn’t reduce work—it intensifies it. _Harvard Business Review. https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it