I agree that Rusell and Norvig AI doesn't have much penetration yet. As for Elements of Statistical Learning...
That's the canonical textbook for ML. If industry relies on splines, boosting, and support vector machines, then it is really not that far from modern academic ML research.
