Study: AI Advice Weakens Critical Thinking and Reduces Willingness to Admit Ignorance
A study by French and Italian researchers reveals that receiving advice from AI models, like Step 3.5 Flash, significantly reduces people's willingness to say "I don't know" from 44% to just 3%. Furthermore, relying on AI advice dropped participants' accuracy from 27% to 9% while doubling their confidence in incorrect answers. This research highlights how AI integration can erode human critical thinking and promote overconfidence in false information, posing risks for decision-making. It underscores the urgent need for educational interventions and AI literacy, especially for children growing up alongside these technologies. Even when researchers introduced monetary incentives to encourage accuracy, the percentage of participants admitting ignorance only rose to 8%, and accuracy to 16%, remaining far below the non-AI baseline. The study deliberately used a model prone to errors to ensure participants were not simply deferring to a highly reliable tool.
## BACKGROUND
This phenomenon is closely linked to automation bias, where humans disproportionately trust automated systems over their own judgment or contradictory evidence. It is further complicated by AI hallucinations, where generative models produce false or misleading information presented convincingly as fact.