Sendhil Mullainathan

No one would say, 'Hey, I think this medicine works, go ahead and use it.' We have testing, we go to the lab, we try it again, we have refinement. But you know what we do on the last mile? 'Oh, this is a good idea. People will like this. Let's put it out there.'

The problem with data is that it says a lot, but it also says nothing. 'Big data' is terrific, but it's usually thin. To understand why something is happening, we have to engage in both forensics and guess work.