Programming Collective Intelligence by Toby Segaran covers the basics of machine learning and in an easy and enjoyable read. All of the examples are in Python and leave nasty, symbol-ridden equations for the appendix. The samples cover real world scenarios such as movie or product recommendations, spam filtering, dating, housing markets, web search, the stock market, etc. In the two days I spent reading the book, I definitely feel like I've learned more than enough to be dangerous.
I was also pleasantly surprised to see time spent on code to visualize the results of the algorithms discussed. There's code to draw dendrograms, networks, multidimensional scaling (2D clusters from n-dimensional data), and decision trees. There's even a discussion of how to use machine learning to improve the quality of network diagrams. Very cool stuff.
Overall, I highly recommend Programming Collective Intelligence for anyone that's ever wondered "how'd they do that?"