The problem with a lot of the data in electronic medical records is that they are not structured but in natural language. This is exactly where Watson can help.
I don't think so. It is too expensive to enroll more than a couple of thousand of participants for a GWA study, though they are often pooled for a broader analysis. Anyway, genotyping will probably cost at least $100 (depending on the technology a lot more), which puts manual entry of patient data into perspective.
An unpaid intern can be cheaper than fancy artificial intelligence!