This was my general take as well. Although now that I think of it, there could be a "fairness" angle regarding how many training data are obtained from each po

This was my general take as well. Although now that I think of it, there could be a "fairness" angle regarding how many training data are obtained from each population group. But if that's what the author meant by fairness, it wasn't clear.