First, I've gone back to school. I have a PhD in computer science from 1983, in which I explored spatial query processing. (General purpose ideas, but mostly applied to database systems.) My PhD has basically expired, the amount of new stuff to learn is daunting, and for a variety of other reasons, I thought it would be better to go back to school rather than study on my own. So I'm enrolled in a CS MS program to learn about AI and cognitive psychology.
Second: To a first approximation, I view the systemsy parts of computer science: CPUs, compilers, operating systems, data structures and algorithms, as having served their purpose, which was to enable AI in its current form. AI is just a different discipline. I happen to be familiar with some parts of those foundations, but I'm basically starting over with AI.
Finally: I recall what I now realize was a fork in the road. I took an AI course in 1980, I think, and we learned about A*, and simple feature detection in images, and alpha-beta pruning, and so on. Symbolic AI was up and coming, and looked to be the way to go. I remember learning about perceptrons and being intrigued. And then I learned about Minsky's proof that a perceptron can't compute XOR, so this idea of computing with a neurons was basically a dead end. Also, I liked the precision of algorithms: the answer is right or not. It has such and such worst case running time. I disliked the wishy washiness of AI, with uncertain and only probably correct answers. I distrusted the idea of tall stacks of probability calculations yielding anything other than a guess. And so I picked my path (into data structures, algorithms, databases).
Now, of course, I realize how wrong it was to dismiss the path of neural models of computation, and the usefulness of statistical methods. I have no regrets, I've had a very satisfying career. But if I were starting out now, I would study a lot more math, and hell yes, get deep into machine learning.