I agree that the things you can do with these large datasets are pretty cool. However, people ignore the exceedingly high error rates when they try to apply the

I agree that the things you can do with these large datasets are pretty cool. However, people ignore the exceedingly high error rates when they try to apply these techniques to "important" problems (e.g. identifying potential terrorists). Take his example of the quiz identifies your home region by dialect quirks. It's really an exercise in confirmation bias. For those who get an accurate result, the quiz is amazing, and they tell all their friends. For people, like me, who were told they grew up across the country from where they actually did, it's just another silly, easy to forget quiz. It's the same for the 20-questions device. We're amazed when it guesses the right answer, but so quickly forget the ones it screws up on. That's why, when I hear about how we just need a big enough dataset to identify threats to our nation or accurately predict the stock market, I worry about that 10-30%.