A new academic study has found that relying on artificial intelligence does not always improve decision-making or answer accuracy. Instead, AI-generated responses may give users a false sense of confidence, even when the information provided is incorrect.

Researchers warn that this effect could discourage people from acknowledging uncertainty, increasing the likelihood that they will defend incorrect answers rather than admit they do not know the correct response. The findings raise concerns for education, workplaces, and other environments where accurate decision-making is essential.

Incorrect Advice, Greater Confidence

The study was led by Valerio Capraro, Professor of Psychology at the University of Milano-Bicocca.

Researchers found that participants who received incorrect advice from an AI system became significantly less accurate in their responses. At the same time, they expressed greater confidence that their answers were correct.

The study also showed that these participants were much less willing to admit they did not know the answer, even when they had the option to skip the question. This occurred despite the experiment rewarding accuracy rather than simply providing an answer.

How the Study Was Conducted

The research was based on six deliberately obscure questions related to movie details.

The questions were specifically selected because the AI model consistently produced incorrect answers, allowing researchers to isolate the psychological impact of consulting AI rather than measuring the quality of the technology itself.

Participants who answered without AI assistance chose not to respond in approximately 36% to 44% of cases.

When AI assistance was available, however, the percentage of unanswered questions dropped dramatically to between 3% and 6%.

At the same time, answer accuracy declined sharply—from 27.5% among participants working independently to just 9.2% among those relying on AI-generated suggestions.

Why Does This Happen?

According to the researchers, AI may lower the level of certainty people normally require before making a decision.

As a result, users become more willing to accept information generated by AI systems—even when that information is incorrect.

Notably, this effect persisted even when AI-generated advice was presented automatically rather than being actively requested by participants, suggesting that simple exposure to AI recommendations can influence confidence and decision-making.

Financial Incentives Did Not Eliminate the Effect

The study also introduced financial rewards for correct answers and penalties for mistakes.

Although these incentives encouraged participants to be more careful and improved overall accuracy, they failed to eliminate the core effect: users continued to display excessive confidence in incorrect answers after relying on AI.

The researchers emphasized that artificial intelligence should be used to support and strengthen human judgment, not replace it entirely.

Concerns for Children and Future Generations

The researchers expressed particular concern about younger users.

They warned that constant reliance on instant answers from AI chatbots could deprive children of experiencing uncertainty—a fundamental part of learning and developing knowledge.

The study also cited recent reports indicating that many young people now turn to AI chatbots for assistance not only with schoolwork but also with conversations and everyday social interactions, raising concerns about the long-term impact on critical thinking and independent reasoning.

The Importance of Healthy Skepticism

The findings highlight that the greatest risk of artificial intelligence may extend beyond factual inaccuracies.

AI systems can also increase users' confidence in misinformation, making them less likely to question incorrect responses or seek additional verification.

The researchers conclude that acknowledging uncertainty is not a weakness but an essential part of learning, and that AI should remain a tool for assistance rather than a substitute for critical thinking.