You're right - I see now when he says 'AI Research' he really means doing applied ML. And, that can certainly be done without knowing everything. If the goal i

You're right - I see now when he says 'AI Research' he really means doing applied ML. And, that can certainly be done without knowing everything. If the goal is to make a lot of money doing applied ML, then become a consultant and aim to know 10% more than the customers. If the goal is to create models that are relatively effective, then read tutorials, play with data, experiment, and iterate. But, if the goal is to create very effective models and be able to actually explain why they work (which I think is what many companies want), then one has to understand the math. That is, showing some interesting relationships, trends, predictions, or inferences on a data analysis portal or consumer web site is one thing. But, using ML to dispense medication, regulate a medical device, drive a power plant, or identify criminal suspects - those may require different skills. (BTW I don't mean to disparage the middle group, as that's largely what I do. But, luckily I have people in the latter group who can validate what I'm doing.)