I don't quite understand why you're demanding this. As I see it, Watson is still pretty experimental and we're probably still learning what kind of stuff it's good or bad at. Why not build a few apps, throw it at hospitals or schools or other use cases, and see whether it actually helps anyone?

Basically, I feel like you're asking for peer-reviewed papers that show that Dropbox is a good way to share files. Nobody would write a paper about that. They would just try Dropbox.