I took a version of this class (also taught by Norvig and Thrun) last year, and I definitely found it very enjoyable. I ended up taking it after CS 229 (which covers the mathematical underpinnings of machine learning with some rigor), so I unfortunately couldn't evaluate how good of an introductory course to AI it would be (having covered a lot of the concepts prior), but even still it was a class I enjoyed. Of particular interest was hearing the instructors draw parallels to their work (particularly with Google and the DARPA challenge), which made a lot of the theoretical concepts much more tangible and helped me recognize their practical applications.
Not very much - most of the lectures stood on their own as far as explaining the concepts, although the book was sometimes useful to consult with regard to the details of an algorithm, for example. Much of that information can be found elsewhere though, so I imagine that you could still get a lot out of this course without AIMA.
you can also see a prior version of the class here: http://www.stanford.edu/class/cs221/