Study: AI Access Makes People Almost Entirely Unwilling to Admit Uncertainty
A study of more than 3,000 participants found that access to AI answers nearly eliminated people's willingness to say «I don't know», even when the AI was almost always wrong. Participants using AI felt more confident but were correct far less often than those without it.
A study involving more than 3,000 participants has found that simply having access to AI-generated answers makes people almost entirely unwilling to admit they do not know something. In one experiment, the share of participants who said «I don't know» fell from 44 percent to just 3 percent when AI was available, even though the AI's answers were almost always incorrect.
The research highlights a troubling side effect of AI assistance: it does not merely provide answers, it also appears to suppress the healthy admission of uncertainty. Participants who used AI reported feeling more confident in their responses, yet they were correct only about one-third as often as those who did not use AI. In other words, the tool inflated confidence while degrading accuracy.
The findings raise questions about how AI is reshaping the way people think, learn, and make decisions. When individuals lose the ability to say «I don't know», they may become more vulnerable to misinformation and less likely to seek reliable sources. The study suggests that the problem is not only that AI can be wrong, but that its presence can make people less willing to question their own knowledge.
According to the study, the effect was dramatic and consistent. The near-elimination of «I don't know» responses occurred even when the AI was almost always wrong. This indicates that the mere availability of an AI answer is enough to override a person's natural hesitation. The participants who used AI felt more assured, but their actual performance suffered significantly.
The research adds to a growing body of evidence about the cognitive and social impacts of artificial intelligence. As AI tools become more integrated into education, workplaces, and daily life, understanding how they influence human judgment is increasingly important. The study's results suggest that access to AI may encourage overconfidence and discourage the kind of intellectual humility that is essential for learning and critical thinking.
While the study focused on a specific experimental setting, its implications extend to broader debates about technology and society. The willingness to admit uncertainty is a cornerstone of honest inquiry, and its erosion could affect everything from public discourse to professional decision-making. The findings do not suggest that AI is inherently harmful, but they do indicate that its use can have unintended consequences for how people assess their own knowledge.
The study's authors note that further research is needed to understand the long-term effects of AI access on confidence and accuracy. For now, the results serve as a cautionary note: the convenience of instant answers may come at the cost of knowing when to say «I don't know».
5
