I disagree with calling ML "Statistics 3". As a field it has diverged from statistics decades ago. Even in their common ground (e.g logistic regression), each field approaches the subject in a different manner.
AI (really machine learning) enhanced infrastructure surveillance isn't the worse use of machine learning. As long as there's steady human oversight with it (which there seems to be), it's a pretty good use of it imo
Indeed, I think these kind of use cases make the most sense. The machine learning system can alert people proactively before the problems happen, but it's not doing anything operational so there's no risk of the system making a bad decision without oversight.
From what I can tell reading the article, the AI isn't actually doing anything operational like driving the trains. It's used to determine the stress on the infrastructure and to alert people to potential problems before they happen so that maintenance is done proactively.
Interesting that they're trying to replace higher-up management positions with this, instead of letting executives solely use it to gut lower-level workers from the payroll.