I agree about scikits.learn. Olivier Grisel did a great job introducing it at Pycon and I find that the documentation makes it more inviting for those new to machine learning.
For better and for worse, the tools that have come along with the information revolution foster what seems to be a much broader, yet more shallow perspective.
For example, lets consider the fact that it takes .19 seconds to find someone's personal distillation of Darwin's Origin of Species. I can now get a summary of one of the great scientific discoveries in less than 1000 characters and be back to reading Facebook updates without blinking an eye.
I didn't read a single word from the original work. I never touched on the years of toil, thought and research that become obvious only after you hear it in Darwin's words. To draw a relevant analogy, its like we are adding layers of abstraction to information. Wikipedia is just the high-level, interpreted view that hides all the nitty gritty details we don't need to worry about anymore. So how often will we need to dive into the inner-workings in the future?
The question becomes is this satisfying? Is it "good enough" to just read the cliff notes? I hate to say it but I think yes. We will end up with an increasing number of "instant experts" who know a little about a lot. And the craftsman - the true specialists - will probably just fade away with the rest of the irrelevant details.