I agree with you on the fact that humans can cope with lack of detailed knowledge to do things. But I think your programming analogy doesn't apply to this situation.

In your analogy, there is me the programmer and the library's programmer. If the library has a consistent behavior, I need only to know how to use it, I don't need to know how it works internally. But the library's programmer needs to know how it works. He may not need to know how the building blocks, he's using in his library, work. but he does need to know how they interact to be able to deliver new features and bugfixes. (I think) He can't add features and correct bugs by trial and error (at least not all the time :)) .

So In your analogy, I think the researchers that came up with the BERT model are the library's programmer and not the programmer.