We are still a good 5 years from any real benefit of machine learning and AI systems. So much research needs to be done still. All the bots we are creating a cool toys, but any real impact is still years ago. We need the Einstein of ML to find his theory of relativity.
ML may very well be overhyped and oversold in this domain (as it is many others). But to call it BS is a sophomoric oversimplification, to say the least. And for helping people sort out the various rather nuanced technical issues involved, not very helpful.
Machine Learning has a lot to answer for. Look at all the people who use the NIST digits and then go looking for cases where people try something similar with some other characters and you will start to wonder if it would work if you tried something else.
I believe Numenta Grok[0] tries to do exactly that: applying ML to monitoring. Anyone knows if it works?