AI Doesn't Give You Judgment, It Gives You Confidence

A research paper recently went viral on Hacker News: after receiving AI suggestions, people’s answer accuracy decreased, but their confidence in those answers increased.

This is exactly what I’ve been saying all along — humans overestimate their own creativity and self-awareness. Much of the time, it’s just a flattering self-deception.

The AI era is putting us through a shift: we’re more willing than ever to draw conclusions, but the quality of those conclusions isn’t improving in tandem. AI is simply giving us more bravado.

What Judgment Really Is

“Judgment” has never been about “reaching a conclusion.” It’s about whether you will act on that conclusion and whether you dare to bear the consequences.

When you decide whether to go with a technical solution, it’s not because you’ve “thought it through” — it’s because you know how to execute and are willing to take responsibility if it fails in production. When you decide whether to accept a job offer, the real test isn’t whether you can list pros and cons — it’s whether you’ll actually take it and can accept the life consequences if you chose wrong. The core of judgment is action and responsibility, not “outputting opinions.”

AI fills the second half perfectly but can’t touch the first half. It can hand you a conclusion, a plan, a seemingly well-reasoned analysis with astonishing ease. But it doesn’t stand in your shoes and bear any consequences. What you gain is speed in “conclusion production,” not a transfer of “consequence bearing.”

What AI Really Gives You Is Confidence

When conclusions come easily enough, you readily mistake “I have an answer” for “I made the right call.” Accuracy drops, confidence rises — and that contains real danger. It means you’re more willing than ever to bet big, but the quality of your bets hasn’t improved — they’ve gotten worse. What you gain is “courage to act,” not “correctness of action.” I call this the “illusion of competence.”

Take code review. You toss a piece of code at AI, it returns a dozen suggestions — clear, well-reasoned, with references. You skim it, think “makes sense,” and accept them all. In that moment, you’ve done two things. First, you’ve outsourced “judging whether this code should change” to the model. Second, because you don’t truly understand the cost of each suggestion, you end up hitting merge more confidently than if you’d written it yourself.

This is why I keep saying that soft skills matter far more than hard skills in the AI era. The point isn’t to reject AI — it’s to stop your brain from taking the easy way out.

What you must hold onto is that responsibility always stays with you. AI should be used to broaden your options, lower your cost of experimentation, and help you see blind spots — these it does better than humans. But it cannot make the final call or bear the consequences.

So I think the requirement for human work needs to shift: from “producing results” to “being responsible for results.” Before, you spent 80% of your energy getting the work done. Now that the work is easier to produce, you should shift that 80% of energy to: “Should this be done at all, and can I handle it if it goes wrong?”