AI access makes people almost entirely unwilling to say “I don’t know,” study finds
What happened
A study involving over 3,000 people found that access to AI answers nearly wiped out their willingness to say “I don’t know.” In a controlled experiment, when participants could consult AI, the rate of admitting uncertainty plunged from 44 percent to just 3 percent. This drop occurred even though the AI provided incorrect answers most of the time. Interestingly, users with AI access felt more confident but were only right about a third as often as those without AI help.
Why it matters
This finding exposes a critical risk when AI tools become trusted sources of information. People tend to over-rely on AI, mistaking confidence for correctness. For builders and operators, this means AI could unintentionally encourage users or employees to commit to answers without proper verification, increasing the chance of error. Businesses relying on AI-generated responses need procedures to check outputs rather than assume AI results are reliable. Overconfidence fueled by AI can also undermine critical thinking, making it harder for teams to spot flawed information before decisions are made.
What to watch next
The next key question is whether AI systems and user interfaces can be designed to foster more honest uncertainty and skepticism rather than suppress it. Companies should monitor how AI access shifts user behavior and decision-making quality in real-world settings. Expect more research and product innovations focused on displaying AI confidence levels, encouraging “I don’t know” answers, or nudging users to double-check AI results. Regulators may also increase scrutiny around AI reliability and user risks as these behavioral impacts become clearer.
AI Quick Briefs Editorial Desk