Classical Methods vs. Learning-Based Methods

Classical methods are conventional techniques in Optimization.

Learning-based methods use Machine Learning.

My question is, should I invest time in learning classical methods, or purely rely on learning-based methods? The thing with classical methods is that you know that they work.

Some examples

I think Geoffrey Hinton kind of talks about this about the approaches to creating Intelligence:

• One is to write up the Logic to create these super intelligent machines
• The other is to learn from Biology

Everything seems to be converging to Learning-Based Methods. Control Theory was previously very much classical, but now there are things being developed such as data-driven MPC.