All technicality aside, models trained on images and their text-descriptions for blind people falls under one of the few good use-cases to Machine Learning.