AI doesn't know what it doesn't know. Yet your brain treats its confident, fluent answers as if they came from a trusted human expert. That mismatch is dangerous.
Working well with AI forces you to deliberately engage your System 2 thinking—the slow, analytical mode—on outputs that System 1 would let slide. Especially when the output looks fine. Because 'looks fine' is exactly the signal your brain uses to skip verification.
Here's the part almost nobody talks about: trust calibration. Your brain builds trust models from experience. When a source gives you correct information repeatedly, you trust it more and scrutinize it less. This is adaptive—it's how expertise forms. But this mechanism misfires with AI.
AI is right often enough that your trust model calibrates upward. You lower your guard. But the errors, when they come, are not distributed like a human expert's errors. A human expert's errors happen at the edges of their knowledge, where they have appropriate uncertainty and usually signal doubt. AI errors happen anywhere, delivered with the same fluency and confidence as everything else.
Your brain is building a trust model for a source that doesn't exist. The fix isn't trusting AI less overall. It's refusing to let fluency do the work of verification.