Technology news and discussion on TalkFuse.
Using AI makes people less likely to admit they don’t know something
Researchers from French and Italian universities have found that giving people access to AI advice makes them far less willing to admit they don't know an answer, even when the AI's advice is wrong. The study, led by Valerio Capraro of the University of Milano-Bicocca, warns that AI is eroding critical thinking and judgment, as people increasingly defer to confident-sounding but inaccurate machine answers rather than acknowledging the limits of their own knowledge.
In the experiment, participants answered obscure film-trivia questions using a large language model (Step 3.5 Flash) chosen because it was usually wrong. Without AI help, 44% of people admitted they didn't know an answer and 27% answered correctly; with AI advice, only 3% admitted ignorance and correct answers fell to 9%, while confidence in being right rose from 30% to 76%. Offering a monetary incentive for accuracy improved things only slightly, and the researchers believe the pattern would likely extend beyond film trivia to other domains.
In short
- AI advice makes people far less willing to say "I don't know"
- Accuracy dropped sharply even as confidence in answers rose
- Cash incentives for correctness barely reversed the effect
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The researchers' concern is justified by their findings: AI systems deliver confident-sounding answers that exploit how people evaluate information, leading them to abandon epistemic humility even when accuracy suffers dramatically. This matters because the effect appears automatic and robust—even financial incentives barely improve it—and if it generalises beyond trivia to medical, financial, or legal domains, the societal consequences could be substantial. The core problem is structural: AI systems are designed to sound authoritative regardless of actual reliability, making them particularly dangerous to critical thinking.
The case against
Whilst the study documents a genuine phenomenon, the concern may be overstated because it tests a deliberately poor AI system in a low-stakes trivia setting unrepresentative of real-world use. In practice, people deploy better-calibrated AI tools alongside other sources and critical judgment, not as complete replacements for thinking. Human susceptibility to confident wrong answers is also longstanding—we have always deferred to confident experts and authorities—so AI represents a difference in degree rather than a novel threat to intellectual integrity. The study also fails to distinguish between cases where growing confidence proves warranted versus unjustified, leaving the practical implications unclear.
Explained simply
A new research study from universities in France and Italy has examined how people behave when they have access to artificial intelligence, particularly when they face questions they don't know the answer to. The researchers were interested in whether having an AI tool available changes how honest people are about the limits of their own knowledge.
In their experiment, the researchers asked participants obscure film-trivia questions and gave some access to an AI tool. When people could use the AI, very few admitted they didn't know the answer and most trusted its responses, even though the system was usually wrong. Without the AI, far more people said they simply didn't know.
This matters because the study suggests that how people interact with AI could affect their willingness to acknowledge the limits of their knowledge. If wider access to AI tools changes whether people admit uncertainty and how much they trust potentially unreliable responses, this could have implications for how they handle information and decisions in their everyday lives.
AI Research Science Technology