Biases in data can be amplified by the training process, leading to distorted — or even unjust — results. And even when a model does work, it’s not always clear why. (Deep learning algorithms are ...