For AI to truly benefit everyone, we need to address the biases in its training data. Here's how we can do it. 🧵