Surveys on AI and critical thinking: what a correlation can and cannot tell you
Two well-known surveys found that people who rely more on AI report or score lower on critical thinking. One asked 666 adults in the UK. The other asked 319 knowledge workers about 936 real work tasks. Both are peer reviewed, and both show a link, not a cause. Neither can tell you whether AI changed how people think, or whether people who think a certain way use AI more.
The two surveys in one table
| Survey | People | Method | Status |
|---|---|---|---|
| Gerlich, 2025 | 666 people in the UK, recruited through social media | A 23-question survey, plus interviews | Peer reviewed. Societies, January 2025, corrected September 2025 |
| Lee and others, 2025 | 319 knowledge workers who use AI at work at least weekly | An online survey about 936 examples from their own work | Peer reviewed. CHI 2025 conference |
The UK survey
Gerlich asked people how much they use AI tools, how much they hand thinking tasks over to them, and a set of questions drawn from critical thinking tests. Three links came out strongly.
- More AI use went with more handing over of thinking (r = +0.72).
- More AI use went with lower critical thinking scores (r = -0.68).
- More handing over went with lower critical thinking scores (r = -0.75).
Younger people relied on AI more and scored lower. More education went with better critical thinking, whatever the AI use.
The workers survey
Lee and colleagues, from Microsoft Research and Carnegie Mellon University, asked workers to describe real tasks they had done with AI. The headline finding has two halves. In the paper's words, "higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking."
- Workers reported some critical thinking in about 60% of the examples (555 of 936).
- For most kinds of thinking, a majority of examples were rated as less effort with AI. The range was from 55% for evaluating to 79% for understanding.
- The thinking that remained shifted toward checking facts, fitting the AI's answer into the work, and overseeing the task.
What a correlation can tell you
A correlation says two things move together. An r of -0.68 is a strong link for a survey. It tells you that in this group, heavy AI users and low scorers were often the same people. That is worth knowing, and it is a good reason to run experiments.
What it cannot tell you
- Which way it runs. AI use might lower scores. Or people who enjoy hard thinking less might reach for AI more. A survey taken at one moment cannot separate these.
- Whether something else explains both. Age, education and type of job are linked to AI use and to test scores.
- Whether the answers are accurate. Both surveys rest on what people say about themselves. Lee's team notes that people mixed up "this took less effort" with "I thought less critically".
- Whether it applies to everyone. Gerlich's sample came from social media. Lee's skewed young and comfortable with technology.
Neither survey measured intelligence, and neither followed anyone over time. Gerlich himself calls for experiments. A correction to one of his tables was published in September 2025. We have not checked what it changed.
A fair one-line summary. "Surveys find heavier AI reliance goes with lower critical thinking scores. They show a link, not a cause."
Where the stronger evidence is
To learn about cause you need experiments, where people are put into groups at random. Several exist. The clearest is a field trial with about 1,000 students, and there is also a set of three experiments on giving up. If these words are new, the plain-English glossary explains them.
What to do with this
The half of Lee's finding that gets less attention is the useful one. People who trusted their own skill checked the AI more. Trying a task yourself first is a direct way to know what you think before the AI tells you.
Questions
Does AI reduce critical thinking?
Surveys link heavier AI reliance with lower critical thinking scores. They cannot show that AI caused the lower scores. No survey can.
What does r = -0.68 mean?
It is a measure of how closely two things move together, on a scale from -1 to +1. A value of -0.68 means that as AI use went up, scores tended to go down, quite consistently. It says nothing about why.
Did the Microsoft study prove AI makes workers think less?
No. It measured what workers reported about their own effort. It did not test their thinking, and it did not follow them over time.
Are these surveys worthless, then?
No. They are a fair warning sign and a reason to test the idea properly. They are just not proof.