The latest frontier in AI research is having more humans in the loop assessing just how good the models are.