I agree that it might not be a large portion of all CS students, but it does appear that many of the incoming CS PhDs have chosen to go into ML (of my class, maybe 1/3), which serious distracts from interest in other subfields. It's actually a joke in my department that all of the new students are pursuing ML, while almost no one is pursuing theoretical CS or other less popular subfields.
I'm also not sure about why the exact number matters - 200 kids matters a lot when future professors are drawn from the pool of students who have successfully completed a PhD.
I think the problem applies to non-PhDs students as well. I'm seeing a lot of interest from recent non-PhD grads in subjects other than CS wanting to steer their career toward data mining. My concern is that this re-focusing will probably lead them down paths that may leave them undistinguished relative to peers who stay within the more stable but less shiny domains where they're better prepared to succeed.
I saw the same thing happen in the late 1990s as everybody and his dog got into web design while it was hot. Few of those folks are doing that now, nor did that skill translate well into other roles since the skill set isn't fundamental to other careers.
I'm not sure ML is any different, especially deep learning, since few companies have anywhere near the necessary amount of data to successfully play that game and win.
I don’t know much about industry but it seems to me there are two ways to look at ML. First is you learn some of optimization, stats, math, parallel computing, numerical methods. The second is that you learn a hell of a lot about fiddling with different network architectures and applying things to specific problems. I wonder whether the first (more fundamental) approach doesn’t have different prospects. At least in this case it can lead to career paths like a national lab.
The first path lets you become a quantitative problem solver, which is widely employable. The second path leads to being very good at specific deep learning tasks that I don't think will be in large demand in 5 years IMO, at least outside the largest tech companies with all the data. Other companies will fulfill their business needs with fewer ML-specialists, AutoML and pretrained models.
I'm also not sure about why the exact number matters - 200 kids matters a lot when future professors are drawn from the pool of students who have successfully completed a PhD.