Am I the only one who feels like all that math is required in order to properly implement an ML system? Bugs in ML systems are insidious and tricky, often difficult to find and manifesting in unexpected ways based on subtle issues in the underlying data or mathematics. Having a ground up understanding of why things work the way they do is a requirement to reasoning through all the layers of the system to find and fix issues.

Maybe someday all of that will be abstracted enough that you don't need to know math to do ML (like how now you generally don't need to know machine architecture to program), but how soon will that be?