Isn't that the point of tools like scikit-learn? You don't need to know how to code, optimize, etc. all the algorithms, just understand how to use them.
Perhaps, but I feel like if you are trying to use a statistical tool, it would be best to know how it works. Think about if every scientist claimed a discovery when they found a result with a 90% confidence interval. Machine learning (at least in this application) is different because often the consequences are testable, verifiable, but I still think that it's better to know how it works than treat it like a black box.
There is probably a large group that lacks advanced linear algebra and statistics for learning the theory but would still be able to build useful applications using a ML library. I think the video is mainly directed at that group.
What makes you assert he is treating it like a black box? There isn't time in the presentation to go into detail, but actually linear models are inspectable, namely, you can obtain a list of features and how they are weighted. Also, as he said, the scikit-learn documentation is of high quality and explains how the models work. BTW you give an example of scientists, but like he stressed, machine learning as he applied it is a form of engineering.
