The Signal and the Noise by Nate Silver
The theme of the book is that we often face temptations to see a signal in a stream of noise -- we are built to see patterns. Silver makes a strong case for resisting these tendencies and explains beautifully why it can be so hard to translate even accurate historical data into actionable information.
When it comes to economics, I welcomed the arguments that prediction markets are not a panacea, the pithy discussion of the weakening of Okun's law (something that has been bothering me for a while), and the warnings that overconfidence of forecasters has not decreased over time.
Silver tells us early in the book that more information is not always automatically beneficial (this conclusion comes up often in my own research so I am eager to agree!). Data in the hand of a hedgehog can be detrimental and an ideologue armed with a PhD degree will not translate observations into insights. Silver's encouragements to his readers to be broad (avoid a narrow specialization) are very reasonable, and his championing of the use of priors (subjective beliefs) make the voice appealing, although there are occasional passages that muddle the overall impression a little bit (for example, the strange suggestion that Isaiah Berlin's classic is not really worth reading as long as you know the gist about the hedgehog and the fox).
It is impossible not to learn new things while reading this book. I learned what the estimated number of possible chess games is, and was stunned to find out that Deep Blue was forced to begin "thinking" after just three moves (rather than relying on its inbuilt database of past games). I also stared at the estimated probabilities of a biological attack on US soil in the near future, learned what the budget of the National Weather Service is, or that more than a million small earthquakes occur every year.
The crowds are correct in this case -- the book is as fun and useful as they say it is. (Another review.)