The Confidence Mirage: AI Makes You Dumber, But Happier

Here is a number that should stop a column dead: 44. That's the percentage of people who, given a task without AI assistance, were willing to say 'I don't know.' Feed them AI advice, and that number plummets to three percent. Not because they suddenly know more—accuracy drops from 27% to 9%—but because they are now suffused with a confidence that bears no relation to reality. The study, conducted by researchers at three French and Italian universities and reported by The Next Web, is the most damning indictment of AI's 'empowerment' narrative I have read this year.

Let me restate that, in case you skimmed. AI advice made participants three times less accurate and more than twice as confident. They knew less, but they believed they knew more. This is not an edge case; this is the product in action. Every company that rolls out an AI assistant—be it for customer service, legal research, or medical diagnosis—is betting that the tool enhances human judgment. The evidence suggests it does the opposite: it suppresses doubt, that most essential of cognitive safeguards. The 'I don't know' muscle atrophies, and what remains is a user who is certain, wrong, and dangerous.

This is the safety farce's quietest act. For months I have chronicled the big gestures—the GPT-5.6 pardon, Sol's secret sauce, Anthropic's constitutional hypocrisy—but the real story has always been the creeping substitution of human critical thinking with machine-generated certainty. The Klarna saga I covered two weeks ago is a parable: replace 700 human agents with an AI, watch efficiency claims crumble, then rehire humans to fix the mess. The study explains why. The AI didn't make the remaining humans better; it made them confidently wrong, and the company had to undo the damage. The cycle repeats, always with the same lesson: AI doesn't augment; it anesthetises.

The implications extend far beyond customer service. The Feds are moving to regulate automated decision-making in government, as iTnews reports. Good. But if the regulator itself is using AI to analyse the evidence, what happens to its judgment? The Washington Post this week described an 'existential panic' among tech workers, who fear their own tools will replace them. They should also fear the tools that remain: a workplace where AI advice has made everyone more confident and less competent. That is not augmentation; that is a slow-motion professional lobotomy.

And yet the industry persists with the 'superhuman' sales pitch. 'AI makes us better,' they say. The study says otherwise. It says AI makes us more confident, which is not the same thing at all. The safety farce, in this light, is not about existential risk from a future superintelligence. It is about the mundane, measurable erosion of human judgment happening right now, on millions of screens, every day. And nobody wants to talk about it because the alternative—admitting that the emperor has no clothes, just a very convincing spreadsheet—would puncture the whole enterprise.

So here we are. The largest bet in corporate history rests on the assumption that AI enhances human decision-making. The evidence says it does the opposite. But don't take my word for it: next time you use an AI tool, try saying 'I don't know.' If the thought feels uncomfortable, you already know the study's conclusion.