If you have mastered the basics (e.g. Norvig's
AIMA, Hastie and Tibshirani's
Elements of Statistical Learning, Koller's
PGM), then I would suggest that the only place to really get a view of the state of the art is by reading papers.
In general, scientific books are an overview of a field, which can only occur with sufficient time for hindsight and synthesis. Even a thousand page book such as Koller's PGM will be littered with references and suggestions of papers to read for a deeper understanding.
One partial exception might be the Deep Learning book by Goodfellow and Bengio, which was made public only a month or so ago. Even this, however, is just an overview. http://www.deeplearningbook.org/