Except that it can put trees somewhere where there were no trees but something similar to them. Or it can put face of a more popular actor instead of an actual less popular one because it was more often present in the training dataset. No, thanks.