I bet the part they most enjoyed was creating the initial training set.

(if it's supervised learning they use, I haven't read the paper)