AI advice made people 3x less accurate but 2x more confident, study finds

Confidence Without Accuracy

New research shows people who received AI-generated advice became three times less accurate in their answers while simultaneously becoming twice as confident in those incorrect conclusions. The study examined how AI recommendations influence human decision-making and critical thinking.

The findings suggest AI tools may create a dangerous feedback loop where users become more certain of their judgments even as the quality of those judgments deteriorates. Researchers observed participants abandoning their own reasoning processes in favor of AI suggestions without adequate scrutiny.

Implications for Workplace Tools

The research arrives as organizations rapidly adopt AI assistants for tasks ranging from code review to strategic analysis. The study's core finding—that AI advice suppresses rather than enhances human critical thinking—challenges assumptions about productivity gains from these tools.

Understanding this dynamic becomes especially important in fields where incorrect decisions carry significant consequences, from healthcare diagnostics to financial risk assessment. Organizations deploying AI assistance may need to reconsider training approaches that emphasize verification rather than blind acceptance.

The Overconfidence Paradox

The simultaneous increase in confidence and decrease in accuracy presents a measurement challenge for teams evaluating AI tool effectiveness. Traditional productivity metrics may miss the degradation in decision quality if users complete tasks faster but with worse outcomes.

The study adds to growing evidence that human-AI interaction requires deliberate design choices about when and how to present machine recommendations. Simply making AI advice available appears insufficient—and potentially counterproductive—without frameworks for maintaining human judgment.