Most people are probably not incentivized by promotions that give them a completely different task they might not even be good at, but by the increase in salary and status.
You can certainly have the latter kind without promoting e.g. the best programmers into mediocre managers.
The paper's suggestion to promote randomly or the worst is a bit silly. It only appears better than the alternative because they assume that there is no information about someone's competence at anything except their current job, and promoting someone is equivalent to firing them in favor of a fresh hire in their model.