In which case...what good is a model that predicts semi-randomly? Oh.

("But it works - when it works" is a tautology, not a useful model)

What does "semi-random" even mean? Are humans not "semi-random" in the same sense?