That's missing the point entirely. I'm pretty sure you don't know the exact materials and the process for making the shoes you might be wearing. There are probably no shoe factories left in america. They same applies to the black box machine learning, very very few people need to know of what goes on under the hood.
edit: I don't know if you changed your comment, but black boxes exist that can solve any problem (neural nets for example).
That might be true for some very general (or common) problems (though I have doubts about that as well), but it can't possibly be true for a specialized problem such as the one described. How do you train a model if you don't understand the problem? You can't just say "unsupervised learning" and expect your algorithm to come up with a full understanding of the physics involved.
I agree with you, I just feel like that is basically the understanding of variables (names of concepts) and not maths. So, you need to understand what measurements to take/can be taken, and then set up that system and feed it to the black box machine learner.
Now, there still need to be some people who understand how things work in detail, but this is much smaller than what is being taught (most k12 do not need to know maths or physics or chemistry etc). They do need practice in abstract thinkng and this would be much better done with programming (a playful, creative, logical, and useful activity).
