I was just writing up an answer to this question when it was decided that it should be closed.
I understand why it is closed; the title alone makes it seem both a bit inflammatory and opinion based, which is generally not what we want on the site.
On the other hand, I feel like it brings up an especially important topic for statisticians; a machine learning type of approach (i.e., using cross-validation to tune hyper-parameters in overly parameterized models) seems to be doing very well at the task of prediction...so where does this leave the field of statistics?
The answer I was starting to write up was that
(1) If faced with the challenge of building a predictive model, I think statisticians should embrace what I'm calling the machine learning approach. After all, there are no hard lines between what's a statistical method and what's a machine learning method.
(2) There are a lot of very important tasks which can be addressed with statistics that have nothing to do with building black-box predictive models.
Anyways, I actually think it's a very interesting question that statisticians should think a little about, especially those early in their career. Do other users think this question could be fixed up or is it in general too open ended to belong on CV?