IMO, AI is logically quite different from Machine Learning / Statistical Learning.
The stuff mentioned on this page is largely about methods to learn from structured or unstructured data, and this is a field that has become highly relevant of late due to the data deluge. Research in these areas has progressed immensely as well, and we now have methods to mine many different types and volumes of data. If you have a good grip of statistical techniques and some basic ML ideas, you will be able to single out and pick the right technique that fits your problem, given your data type, SNR ratio, structured-ness, volume, your resource constraints, etc. Knowing a little more about ML will also allow you to change/invent new methods to suit your own problems better (e.g., a new way to compress your feature space).
I mostly agree with parent, but the divorce between AI and ML might only be temporary, and the deep learning branch of ML is targeting more general AI than just fitting linear models, c.f. for instance for a high level overview:
Great book list.
I strongly recommend getting a strong gasp of linear algebra as matrices are the a great way to think of large data in a manageable way.
also, for folks who just want to their feet wet, oreilly's programming collective intelligence is a good start.
I own them all, and I am perpetually at different stages of working through each of them. :-) I work in the way I explained in the intro - when I don't know something, I step back and go learn the background I need to move forward.
On the contrary; I find learning and production to be mutually self-reinforcing rather than conflicting.
I am a professional committed to both practicing my craft and consistently increasing my skill at my craft. For machine learning researchers, computer scientists, and engineers, a healthy ongoing dose of theory and practice is a great way to proceed.
http://www.autonlab.org/tutorials/