Many people and companies are working on this. The major blocker, in my opinion, is that we don't really track outcomes rigorously and with out those, we have n

Many people and companies are working on this. The major blocker, in my opinion, is that we don't really track outcomes rigorously and with out those, we have no way of categorizing the training examples. Yes we do some outcomes tracking but I would be interested in examining collection bias in those sets and how it affects the results. I find the claim that their NLP generated database is 98% accurate. That would be an amazing leap forward. I hope it is true but their is a lot of exaggeration and outright bunk in this field