There loads of practical real world applications for machine learning, but it’s not what the big corpos tend to push because the usefulness is generally more niche. They want to sell you universal solutions to problems that don’t exist.