AI use may erode willingness to say 'I don’t know'

A study by researchers from French and Italian universities suggests AI access can weaken critical thinking, especially the willingness to admit uncertainty. In the data cited, only 3% of participants with access to AI said “I don’t know” on a question requiring critical thinking, compared with 44% of participants who did not have access to AI.

The researchers argue that people leaned too heavily on AI, often replacing potentially correct personal judgment with incorrect AI suggestions. To test this, the study deliberately used a model with a lower accuracy rate so it would produce wrong answers. Even then, participants often chose the AI’s output instead of reasoning independently, offering a better answer, or withholding judgment.

The paper says the effect showed up across five experiments. As the researchers put it:

“Across five experiments, mere access to AI advice nearly eliminated participants' willingness to suspend judgment.”

They added:

“As AI-generated answers become ubiquitous and, increasingly, unsolicited, our results show that the willingness to say 'I don’t know' may be among the first casualties of human–AI interaction.”

At the same time, the study found a contradiction. Participants still appeared to recognize that AI could be wrong, and when given small financial incentives to answer more honestly, more of them were less likely to copy the AI’s exact wording.

Recommended reading

Kimi K3 beats Claude on frontend coding

The paper also points to a broader confidence problem: people tend to feel more certain in their answers when external information is available, including AI-generated responses, even when that information is inaccurate or not especially relevant.

The researchers acknowledged the study’s limitations and small scope, but still called for more AI literacy and education focused on verifying outputs and reinforcing human critical thinking. The report was highlighted via The Register.