The report from French and Italian universities adds weight to growing concerns about artificial intelligence and human cognition. Researchers found that people with access to AI tools were far less likely to say they did not know an answer, even when the AI was intentionally inaccurate. The findings suggest that reliance on AI may quietly erode critical thinking skills.

What You Need to Know

This research highlights a phenomenon known as cognitive offloading, where humans outsource mental effort to technology. The study used a deliberately flawed AI model to test whether participants would still defer to its answers. The results show that even when people suspect AI may be wrong, they often accept its output rather than trust their own judgment or admit uncertainty.

Critical Thinking Under AI Influence

The researchers conducted five experiments with hundreds of participants. In each, a group with access to AI advice was compared with a control group that had no AI assistance. The AI model had a lower accuracy rate on purpose, producing incorrect answers for many questions. Despite this, only 3% of participants in the AI group said they did not know an answer. In the control group, 44% were willing to admit uncertainty.

Participants also showed a paradox: they felt more confident in their answers when AI was available, even though the AI was often wrong. Small financial incentives to be honest reduced but did not eliminate the tendency to follow AI suggestions. The report's authors wrote that access to AI "nearly eliminated participants’ willingness to suspend judgment."

  • Deference over doubt: Participants routinely chose AI answers over their own reasoning, even when the AI lacked credibility.
  • False confidence: Access to external information, even flawed AI output, inflated participants’ certainty in their responses.
  • Incentive effects: Honesty prompts helped slightly, but the overall pattern of AI deference remained strong.

The findings align with broader trends in human-computer interaction. As AI becomes embedded in daily tools, from search engines to writing assistants, the habit of verifying information may weaken. The study suggests that the ability to say “I don’t know” is among the first casualties of human-AI interaction.

Why This Matters

The implications extend far beyond academic experiments. In workplaces that increasingly rely on AI suggestions, employees may stop questioning outputs that seem plausible but are wrong. In education, students using AI for homework risk losing the skill of forming independent judgments. Regulators and educators face a pressing need to promote AI literacy, teaching people not just how to use the tools but when to challenge them.

The paradox of confidence combined with reduced accuracy creates a dangerous dynamic. People feel more informed while actually becoming less reliable. If left unaddressed, this could degrade decision-making in fields such as medicine, law and journalism where critical thinking is essential. The research underscores that designing AI systems to encourage healthy skepticism rather than blind acceptance is a design and policy priority.

A Broader Pattern of Cognitive Offloading

This is not the first study to flag cognitive side effects of technology. Research on GPS navigation found that drivers who relied on turn-by-turn directions developed weaker spatial memory. Similarly, search engines have been shown to reduce recall of facts people know they can find later. AI represents a more advanced form of offloading because it generates plausible answers without prompting the user to reflect.

What makes this study notable is its focus on uncertainty. Humans have long used the phrase “I don’t know” as a signal of intellectual honesty and a starting point for deeper investigation. If AI removes that instinct, it may also remove the motivation to learn. The report calls for embedding critical thinking exercises into AI interfaces, reminding users that not all machine outputs are equal.