It's one of the few areas where you can verify results. That fits nicely into training models. They aren't just making judgement calls on what would be nice, it

It's one of the few areas where you can verify results. That fits nicely into training models. They aren't just making judgement calls on what would be nice, it's "what can we do?".