Yes, he is totally right.
ML is like Array map function, but more complicated. Today you also have a lot of ML libraries, so you don't need to know how it works inside. The most critical part you need in ML is data on input, then in the middle there is your model or in simple words - what do you want to extract from the data and finally at the end you will see some results.
However, to get best results in your unique case and for a bit complicated and custom scenarios, you need to know at least basics of the machine learning, math, analytics and statistics.
Moreover, for training models or making them better, you need not just data, but filtered (labeled) data so algorithm could extract value from a lot of labeled data sets, for example, if I will show you 2 pictures with some object in language you do not speak, you won't be able to learn how to say those objects, so I need also to provide you words for each picture (labels) and finally when you will have a lot of pictures with words, you will have high chance (probability) to tell me what you see on the next picture without a word I will show you. So in simple words, AI is a hype and myth, human needs to do a lot of job first.