Researchers found AI advice suppressed willingness to say "I don’t know" from 44% to 3%, accuracy fell from 27% to 9%, while confidence rose from 30% to 76%.

The study, led by Valerio Capraro with Chiara Marcoccia and Walter Quattrociocchi, used questions that AI models typically mishandle, such as visual details from films. Step 3.5 Flash was deliberately chosen because it usually answered incorrectly, so any drop in judgment could not be blamed on a reliable tool.

Even with monetary incentives, participants who sought AI help became more often wrong. The willingness to admit ignorance rose only to 8% and accuracy to 16%, still far below the no‑AI baselines of 44% and 27%. Wharton researchers had earlier coined the term "cognitive surrender" for this phenomenon.

The findings raise concerns about children who grow up with AI that never says "I don’t know," especially as Google’s AI‑generated summaries replace links, a move criticized by Common Sense Media as an "unacceptable risk" for students.